@murumets-ee/imports 0.83.0 → 0.84.0

package/dist/analyze/index.mjs

@@ -1,2 +1,2 @@
-import{n as e}from"../read-feed-DGjqejw3.mjs";import{c as t}from"../parts-schema-QqLci_Pl.mjs";import{n,r}from"../schema-registry-DB5aIZMN.mjs";import{o as i,s as a}from"../matching-Sr0z4HvN.mjs";import{parse as o}from"csv-parse/sync";const s=[`	`,`,`,`;`,`|`],c=512*1024,l=200;function u(e){return e===`	`?`TAB`:e===` `?`SPACE`:`"${e}"`}function d(e){let t=null;for(let n of s){let r;try{r=o(e,{delimiter:n,columns:!1,bom:!0,skip_empty_lines:!0,relax_column_count:!0,quote:!1,to_line:200})}catch{continue}if(r.length===0)continue;let i=new Map;for(let e of r)i.set(e.length,(i.get(e.length)??0)+1);let a=0,s=0;for(let[e,t]of i)(t>s||t===s&&e>a)&&(a=e,s=t);if(a<2)continue;let c=s/r.length;(t===null||c>t.consistency||c===t.consistency&&a>t.fieldCount)&&(t={delimiter:n,fieldCount:a,consistency:c,evidence:`${u(n)} yields ${a} fields on ${Math.round(c*100)}% of ${r.length} sampled row(s)`})}return t}function f(e,t){let n=(e[0]??[]).map(e=>e.trim());if(e.length===0)return{hasHeader:!1,firstRowCells:n,evidence:`the file has no records`};let r=t(n);if(r>0)return{hasHeader:!0,firstRowCells:n,evidence:`the first row names ${r} recognisable target field(s)`};let i=n.filter(e=>e!==``),a=new Set(i).size===i.length,o=/^-?\d+(?:[., ]\d+)*$/,s=i.length>0&&!n.some(e=>o.test(e)),c=e.slice(1),l=!1;if(c.length>0)for(let e=0;e<n.length;e+=1){let t=0,n=0;for(let r of c){let i=r[e]?.trim();i===void 0||i===``||(n+=1,o.test(i)&&(t+=1))}if(n>=3&&t/n>=.8){l=!0;break}}return s&&a&&l?{hasHeader:!0,firstRowCells:n,evidence:`the first row is entirely non-numeric with distinct values while a column below it is predominantly numeric`}:{hasHeader:!1,firstRowCells:n,evidence:`the first row matches no target field and does not read as a header over the rows below — treating the feed as positional`}}function p(e,t){try{return o(e,{delimiter:t,columns:!1,bom:!0,skip_empty_lines:!0,relax_column_count:!0,quote:!1,to_line:200})}catch{return[]}}async function m(t){let{filePath:n,format:r,delimiter:i,hasHeader:a,quote:o,maxRows:s}=t,c=t.maxSamplesPerColumn??2e3,l=t.maxSamplesTotal??2e4,u=t.maxSampleRows??5,d=new Map,f=[],p=[],m=0,h=0,g=!1;try{for await(let{row:t}of e({filePath:n,format:r,delimiter:i,hasHeader:a,quote:o})){m+=1,p.length===0&&(p=Object.keys(t),!g&&p.some(e=>e.includes(`"`))&&(g=!0)),f.length<u&&f.push({...t});for(let[e,n]of Object.entries(t)){if(n===``||(!g&&n.includes(`"`)&&(g=!0),h>=l))continue;let t=d.get(e);t===void 0?(d.set(e,[n]),h+=1):t.length<c&&(t.push(n),h+=1)}if(s!==void 0&&s>0&&m>=s)break}}catch(e){return{quote:o,ok:!1,error:e instanceof Error?e.message:String(e),rows:m,sawQuoteChar:g,samples:d,sampleRows:f,columns:p}}return{quote:o,ok:!0,error:null,rows:m,sawQuoteChar:g,samples:d,sampleRows:f,columns:p}}const h=[`.`,`,`,` `],g=/^-?\d+(?:[., ]\d+)*$/;function _(){return{decimal:{".":0,",":0," ":0},grouping:{".":0,",":0," ":0},ambiguous:{".":0,",":0," ":0},ambiguousSamples:[]}}function v(e){return e===`.`||e===`,`||e===` `}function y(e){let t=[];for(let n=0;n<e.length;n+=1){let r=e[n];r!==void 0&&v(r)&&t.push({char:r,index:n})}return t}function b(e,t){let n=y(e);if(n.length===0)return;if(new Set(n.map(e=>e.char)).size>=2){let e=n[n.length-1];if(e===void 0)return;t.decimal[e.char]+=1;for(let r of n)r.char!==e.char&&(t.grouping[r.char]+=1);return}let r=n[0]?.char;if(r!==void 0){if(n.length>=2){t.grouping[r]+=1;return}if(e.length-(n[0]?.index??0)-1==3){t.ambiguous[r]+=1,t.ambiguousSamples.length<5&&t.ambiguousSamples.push(e);return}t.decimal[r]+=1}}function x(e){let n=_(),r=0,i=[];for(let t of e){let e=t.trim();e!==``&&g.test(e)&&(r+=1,i.push(e),b(e,n))}n.grouping[` `]+=n.decimal[` `]+n.ambiguous[` `],n.decimal[` `]=0,n.ambiguous[` `]=0;let a=n.decimal[`.`]+n.decimal[`,`],o=n.grouping[`.`]+n.grouping[`,`]+n.grouping[` `],s=n.ambiguous[`.`]+n.ambiguous[`,`],c={considered:e.length,numeric:r,separated:a+o},l=n.decimal[`.`],u=n.decimal[`,`];if(l>0&&u>0)return{format:null,confidence:`conflict`,evidence:`the file argues for BOTH decimal separators — ${l} value(s) put a non-3-digit run after "." and ${u} after ",". One feed is using two conventions, or the delimiter is wrong and two columns have been merged into one.`,unparsed:[],counts:c};let d=null;if(l>0?d=`.`:u>0&&(d=`,`),d===null){if(s>0)return{format:null,confidence:`ambiguous`,evidence:`every separated value has EXACTLY 3 digits after its separator (${h.filter(e=>n.ambiguous[e]>0).map(e=>`"${e}"`).join(`/`)}), e.g. ${n.ambiguousSamples.join(`, `)}. "1,234" is one-thousand-two-hundred-thirty-four or one-point-two-three-four and this file contains nothing that distinguishes them. Guessing here is a silent 1000x error, so state the format explicitly.`,unparsed:[],counts:c};let e=h.find(e=>n.grouping[e]>0);if(e!==void 0){let t=e===`.`?`,`:`.`;return{format:{decimal:t,thousands:e},confidence:`no-evidence`,evidence:`thousands "${e===` `?` `:e}" from ${o} grouped value(s), but NO value shows a decimal separator, so the decimal is a default ("${t}", chosen to differ from the grouping) rather than a reading. Check it against the supplier's own documentation before saving.`,unparsed:[],counts:c}}return{format:{decimal:`.`,thousands:``},confidence:`no-evidence`,evidence:r===0?`no numeric-shaped values were found in the sampled columns.`:`all ${r} numeric sample(s) were bare integers with no separator, so the file says nothing about its decimal separator. Defaulted to "." / none.`,unparsed:[],counts:c}}let f=h.filter(e=>e!==d&&n.grouping[e]>0),p=f[0]??``;if(f.length>1)return{format:null,confidence:`conflict`,evidence:`decimal is "${d}", but ${f.length} different characters (${f.map(e=>`"${e}"`).join(`, `)}) are each used as a thousands separator. A feed uses one.`,unparsed:[],counts:c};let m={decimal:d,thousands:p},v=[];for(let e of i)Number.isNaN(t(e,m))&&v.length<5&&v.push(e);return{format:m,confidence:`high`,evidence:`decimal "${d}" from ${d===`.`?l:u} value(s) with a non-3-digit run after it; thousands ${p===``?`none`:`"${p}"`} from ${o} grouped value(s). ${r} numeric sample(s) considered.`,unparsed:v,counts:c}}const S=1,C=new Set([`supersededno`,`supersededarticle`,`supersedednumber`,`ersatznummer`,`ersatznr`,`replacementcode`,`replacementno`]),w=[`superseding_code`,`replaces_code`];var T=class extends Error{hint;constructor(e,t){super(e),this.name=`FeedAnalysisError`,this.hint=t}};function E(e){return e===`number`?[{op:`trim`},{op:`number`}]:[{op:`trim`}]}function D(e){return e.toLowerCase().replace(/[^a-z0-9]/g,``)}async function O(e,t){let{open:n}=await import(`node:fs/promises`),r=await n(e,`r`);try{let e=Buffer.alloc(t),{bytesRead:n}=await r.read(e,0,t,0);return e.subarray(0,n).toString(`utf8`)}finally{await r.close()}}async function k(e){let{stat:t}=await import(`node:fs/promises`);return(await t(e)).size}function A(e,t){if(t===null)return{chosen:e,evidence:`no quote character appears anywhere in ${e.rows} scanned row(s), so quoting is immaterial; parsed ${e.rows} row(s) cleanly with quotes as data`,warning:null};if(e.ok&&!t.ok)return{chosen:e,evidence:`quotes are ordinary data: parsed ${e.rows} row(s) cleanly, while RFC 4180 failed after ${t.rows} row(s) — ${t.error}`,warning:null};if(!e.ok&&t.ok)return{chosen:t,evidence:`RFC 4180 quoting: parsed ${t.rows} row(s) cleanly, while treating quotes as data failed after ${e.rows} row(s) — ${e.error}`,warning:null};if(e.ok&&t.ok)return{chosen:t,evidence:`BOTH policies parse this file cleanly (${e.rows} row(s) each); chose RFC 4180 because every quote character sits at a field boundary`,warning:`Quoting is ambiguous — both policies parse this file. RFC 4180 was chosen, which STRIPS the enclosing quotes from field values; treating them as data would keep them. Check a sample row before saving, and set the policy explicitly if the other reading is right.`};throw new T(`neither quoting policy can parse this file.\n  quotes-as-data: ${e.error} (after ${e.rows} row(s))\n  RFC 4180:       ${t.error} (after ${t.rows} row(s))`,`Both failures usually mean the DELIMITER or the header verdict is wrong, not the quoting. Check the reported delimiter against the file, and re-run with an explicit --delimiter.`)}async function j(e){let{filePath:t}=e,o=e.targetSchema??`parts`,s=n(o);if(s===void 0)throw new T(`Unknown target schema "${o}".`,`Registered schemas: ${r().join(`, `)}`);let l=e.readSample??O,u=await(e.fileSize??k)(t),h=await l(t,c);if(h.trim()===``)throw new T(`${t} is empty.`);let g=d(h);if(g===null)throw new T(`Could not determine a field delimiter for ${t}.`,`No candidate (TAB, comma, semicolon, pipe) split the sample into more than one column consistently. If the feed uses another delimiter, or is a spreadsheet, map it in the admin wizard instead.`);let{delimiter:_}=g,v=_===`,`?`csv`:`tsv`,y=f(p(h,_),e=>a(e.filter(e=>e!==``),s.fields).size),{hasHeader:b}=y,S={filePath:t,format:v,delimiter:_,hasHeader:b,...e.maxRows!==void 0&&{maxRows:e.maxRows}},C=await m({...S,quote:!1}),{chosen:w,evidence:D,warning:j}=A(C,C.sawQuoteChar||!C.ok?await m({...S,quote:`"`}):null),N=[];j!==null&&N.push(j);let P=w.columns;if(P.length===0)throw new T(`${t} parsed to zero columns.`,`The file may be empty below its header.`);let F=a(P,s.fields),I=new Map(s.fields.map(e=>[e.key,e])),L=[],R=[];for(let e of P){let t=F.get(e);if(t!==void 0){let n=I.get(t);L.push({source:e,target:t,transforms:E(n?.type??`string`)});continue}R.push(M(e))}let z=L.filter(e=>I.get(e.target)?.type===`number`).map(e=>e.source),B=[];for(let e of z)for(let t of w.samples.get(e)??[])B.push(t);let V=x(B);V.format===null&&N.push(`The number format could NOT be decided (${V.confidence}): ${V.evidence} Supply { decimal, thousands } explicitly before saving.`),V.unparsed.length>0&&N.push(`${V.unparsed.length}+ sampled value(s) in the number columns do not parse under the inferred format: ${V.unparsed.join(`, `)}. Those rows will be row errors at import.`),z.length===0&&N.push(`No column mapped to a numeric target, so the number format was inferred from nothing. Check it before saving.`);let H=b?i({columns:[...P],rows:w.sampleRows}):null;return{version:1,file:{path:t,bytes:u},targetSchema:o,reader:{format:v,delimiter:_,hasHeader:b,quote:w.quote},numberFormat:V.format,numberFormatConfidence:V.confidence,columnMap:L,unmapped:R,headerSnapshot:H,dataRows:w.rows,truncated:e.maxRows!==void 0&&e.maxRows>0&&w.rows>=e.maxRows,evidence:{delimiter:g.evidence,hasHeader:y.evidence,quote:D,numberFormat:V.evidence},warnings:N}}function M(e){return C.has(D(e))?{source:e,reason:`"${e}" names a replacement code but not its DIRECTION, and the two readings are opposites: the column may hold the NEWER code that supersedes this row (superseding_code), or the OLDER code this row supersedes (replaces_code). This is deliberately not guessed — an inverted feed builds well-formed chains with valid heads and no warnings, and the merge is append-only, so re-importing the right way round does NOT repair it. Check a superseded code against the supplier's current catalogue and state the direction.`,candidates:[...w]}:{source:e,reason:`no target field matches "${e}".`,candidates:[]}}function N(e){return e.unmapped.map(e=>e.source)}export{s as CANDIDATE_DELIMITERS,c as DIALECT_SAMPLE_BYTES,l as DIALECT_SAMPLE_ROWS,S as FEED_ANALYSIS_VERSION,T as FeedAnalysisError,j as analyzeFeed,u as delimiterName,d as inferDelimiter,f as inferHasHeader,x as inferNumberFormat,p as sampleRecords,m as trialParse,N as unmappedSources};
+import{n as e}from"../read-feed-DsLgRDNR.mjs";import{c as t}from"../parts-schema-QqLci_Pl.mjs";import{n,r}from"../schema-registry-DB5aIZMN.mjs";import{o as i,s as a}from"../matching-Sr0z4HvN.mjs";import{parse as o}from"csv-parse/sync";const s=[`	`,`,`,`;`,`|`],c=512*1024,l=200;function u(e){return e===`	`?`TAB`:e===` `?`SPACE`:`"${e}"`}function d(e){let t=null;for(let n of s){let r;try{r=o(e,{delimiter:n,columns:!1,bom:!0,skip_empty_lines:!0,relax_column_count:!0,quote:!1,to_line:200})}catch{continue}if(r.length===0)continue;let i=new Map;for(let e of r)i.set(e.length,(i.get(e.length)??0)+1);let a=0,s=0;for(let[e,t]of i)(t>s||t===s&&e>a)&&(a=e,s=t);if(a<2)continue;let c=s/r.length;(t===null||c>t.consistency||c===t.consistency&&a>t.fieldCount)&&(t={delimiter:n,fieldCount:a,consistency:c,evidence:`${u(n)} yields ${a} fields on ${Math.round(c*100)}% of ${r.length} sampled row(s)`})}return t}function f(e,t){let n=(e[0]??[]).map(e=>e.trim());if(e.length===0)return{hasHeader:!1,firstRowCells:n,evidence:`the file has no records`};let r=t(n);if(r>0)return{hasHeader:!0,firstRowCells:n,evidence:`the first row names ${r} recognisable target field(s)`};let i=n.filter(e=>e!==``),a=new Set(i).size===i.length,o=/^-?\d+(?:[., ]\d+)*$/,s=i.length>0&&!n.some(e=>o.test(e)),c=e.slice(1),l=!1;if(c.length>0)for(let e=0;e<n.length;e+=1){let t=0,n=0;for(let r of c){let i=r[e]?.trim();i===void 0||i===``||(n+=1,o.test(i)&&(t+=1))}if(n>=3&&t/n>=.8){l=!0;break}}return s&&a&&l?{hasHeader:!0,firstRowCells:n,evidence:`the first row is entirely non-numeric with distinct values while a column below it is predominantly numeric`}:{hasHeader:!1,firstRowCells:n,evidence:`the first row matches no target field and does not read as a header over the rows below — treating the feed as positional`}}function p(e,t){try{return o(e,{delimiter:t,columns:!1,bom:!0,skip_empty_lines:!0,relax_column_count:!0,quote:!1,to_line:200})}catch{return[]}}async function m(t){let{filePath:n,format:r,delimiter:i,hasHeader:a,quote:o,maxRows:s}=t,c=t.maxSamplesPerColumn??2e3,l=t.maxSamplesTotal??2e4,u=t.maxSampleRows??5,d=new Map,f=[],p=[],m=0,h=0,g=!1;try{for await(let{row:t}of e({filePath:n,format:r,delimiter:i,hasHeader:a,quote:o})){m+=1,p.length===0&&(p=Object.keys(t),!g&&p.some(e=>e.includes(`"`))&&(g=!0)),f.length<u&&f.push({...t});for(let[e,n]of Object.entries(t)){if(n===``||(!g&&n.includes(`"`)&&(g=!0),h>=l))continue;let t=d.get(e);t===void 0?(d.set(e,[n]),h+=1):t.length<c&&(t.push(n),h+=1)}if(s!==void 0&&s>0&&m>=s)break}}catch(e){return{quote:o,ok:!1,error:e instanceof Error?e.message:String(e),rows:m,sawQuoteChar:g,samples:d,sampleRows:f,columns:p}}return{quote:o,ok:!0,error:null,rows:m,sawQuoteChar:g,samples:d,sampleRows:f,columns:p}}const h=[`.`,`,`,` `],g=/^-?\d+(?:[., ]\d+)*$/;function _(){return{decimal:{".":0,",":0," ":0},grouping:{".":0,",":0," ":0},ambiguous:{".":0,",":0," ":0},ambiguousSamples:[]}}function v(e){return e===`.`||e===`,`||e===` `}function y(e){let t=[];for(let n=0;n<e.length;n+=1){let r=e[n];r!==void 0&&v(r)&&t.push({char:r,index:n})}return t}function b(e,t){let n=y(e);if(n.length===0)return;if(new Set(n.map(e=>e.char)).size>=2){let e=n[n.length-1];if(e===void 0)return;t.decimal[e.char]+=1;for(let r of n)r.char!==e.char&&(t.grouping[r.char]+=1);return}let r=n[0]?.char;if(r!==void 0){if(n.length>=2){t.grouping[r]+=1;return}if(e.length-(n[0]?.index??0)-1==3){t.ambiguous[r]+=1,t.ambiguousSamples.length<5&&t.ambiguousSamples.push(e);return}t.decimal[r]+=1}}function x(e){let n=_(),r=0,i=[];for(let t of e){let e=t.trim();e!==``&&g.test(e)&&(r+=1,i.push(e),b(e,n))}n.grouping[` `]+=n.decimal[` `]+n.ambiguous[` `],n.decimal[` `]=0,n.ambiguous[` `]=0;let a=n.decimal[`.`]+n.decimal[`,`],o=n.grouping[`.`]+n.grouping[`,`]+n.grouping[` `],s=n.ambiguous[`.`]+n.ambiguous[`,`],c={considered:e.length,numeric:r,separated:a+o},l=n.decimal[`.`],u=n.decimal[`,`];if(l>0&&u>0)return{format:null,confidence:`conflict`,evidence:`the file argues for BOTH decimal separators — ${l} value(s) put a non-3-digit run after "." and ${u} after ",". One feed is using two conventions, or the delimiter is wrong and two columns have been merged into one.`,unparsed:[],counts:c};let d=null;if(l>0?d=`.`:u>0&&(d=`,`),d===null){if(s>0)return{format:null,confidence:`ambiguous`,evidence:`every separated value has EXACTLY 3 digits after its separator (${h.filter(e=>n.ambiguous[e]>0).map(e=>`"${e}"`).join(`/`)}), e.g. ${n.ambiguousSamples.join(`, `)}. "1,234" is one-thousand-two-hundred-thirty-four or one-point-two-three-four and this file contains nothing that distinguishes them. Guessing here is a silent 1000x error, so state the format explicitly.`,unparsed:[],counts:c};let e=h.find(e=>n.grouping[e]>0);if(e!==void 0){let t=e===`.`?`,`:`.`;return{format:{decimal:t,thousands:e},confidence:`no-evidence`,evidence:`thousands "${e===` `?` `:e}" from ${o} grouped value(s), but NO value shows a decimal separator, so the decimal is a default ("${t}", chosen to differ from the grouping) rather than a reading. Check it against the supplier's own documentation before saving.`,unparsed:[],counts:c}}return{format:{decimal:`.`,thousands:``},confidence:`no-evidence`,evidence:r===0?`no numeric-shaped values were found in the sampled columns.`:`all ${r} numeric sample(s) were bare integers with no separator, so the file says nothing about its decimal separator. Defaulted to "." / none.`,unparsed:[],counts:c}}let f=h.filter(e=>e!==d&&n.grouping[e]>0),p=f[0]??``;if(f.length>1)return{format:null,confidence:`conflict`,evidence:`decimal is "${d}", but ${f.length} different characters (${f.map(e=>`"${e}"`).join(`, `)}) are each used as a thousands separator. A feed uses one.`,unparsed:[],counts:c};let m={decimal:d,thousands:p},v=[];for(let e of i)Number.isNaN(t(e,m))&&v.length<5&&v.push(e);return{format:m,confidence:`high`,evidence:`decimal "${d}" from ${d===`.`?l:u} value(s) with a non-3-digit run after it; thousands ${p===``?`none`:`"${p}"`} from ${o} grouped value(s). ${r} numeric sample(s) considered.`,unparsed:v,counts:c}}const S=1,C=new Set([`supersededno`,`supersededarticle`,`supersedednumber`,`ersatznummer`,`ersatznr`,`replacementcode`,`replacementno`]),w=[`superseding_code`,`replaces_code`];var T=class extends Error{hint;constructor(e,t){super(e),this.name=`FeedAnalysisError`,this.hint=t}};function E(e){return e===`number`?[{op:`trim`},{op:`number`}]:[{op:`trim`}]}function D(e){return e.toLowerCase().replace(/[^a-z0-9]/g,``)}async function O(e,t){let{open:n}=await import(`node:fs/promises`),r=await n(e,`r`);try{let e=Buffer.alloc(t),{bytesRead:n}=await r.read(e,0,t,0);return e.subarray(0,n).toString(`utf8`)}finally{await r.close()}}async function k(e){let{stat:t}=await import(`node:fs/promises`);return(await t(e)).size}function A(e,t){if(t===null)return{chosen:e,evidence:`no quote character appears anywhere in ${e.rows} scanned row(s), so quoting is immaterial; parsed ${e.rows} row(s) cleanly with quotes as data`,warning:null};if(e.ok&&!t.ok)return{chosen:e,evidence:`quotes are ordinary data: parsed ${e.rows} row(s) cleanly, while RFC 4180 failed after ${t.rows} row(s) — ${t.error}`,warning:null};if(!e.ok&&t.ok)return{chosen:t,evidence:`RFC 4180 quoting: parsed ${t.rows} row(s) cleanly, while treating quotes as data failed after ${e.rows} row(s) — ${e.error}`,warning:null};if(e.ok&&t.ok)return{chosen:t,evidence:`BOTH policies parse this file cleanly (${e.rows} row(s) each); chose RFC 4180 because every quote character sits at a field boundary`,warning:`Quoting is ambiguous — both policies parse this file. RFC 4180 was chosen, which STRIPS the enclosing quotes from field values; treating them as data would keep them. Check a sample row before saving, and set the policy explicitly if the other reading is right.`};throw new T(`neither quoting policy can parse this file.\n  quotes-as-data: ${e.error} (after ${e.rows} row(s))\n  RFC 4180:       ${t.error} (after ${t.rows} row(s))`,`Both failures usually mean the DELIMITER or the header verdict is wrong, not the quoting. Check the reported delimiter against the file, and re-run with an explicit --delimiter.`)}async function j(e){let{filePath:t}=e,o=e.targetSchema??`parts`,s=n(o);if(s===void 0)throw new T(`Unknown target schema "${o}".`,`Registered schemas: ${r().join(`, `)}`);let l=e.readSample??O,u=await(e.fileSize??k)(t),h=await l(t,c);if(h.trim()===``)throw new T(`${t} is empty.`);let g=d(h);if(g===null)throw new T(`Could not determine a field delimiter for ${t}.`,`No candidate (TAB, comma, semicolon, pipe) split the sample into more than one column consistently. If the feed uses another delimiter, or is a spreadsheet, map it in the admin wizard instead.`);let{delimiter:_}=g,v=_===`,`?`csv`:`tsv`,y=f(p(h,_),e=>a(e.filter(e=>e!==``),s.fields).size),{hasHeader:b}=y,S={filePath:t,format:v,delimiter:_,hasHeader:b,...e.maxRows!==void 0&&{maxRows:e.maxRows}},C=await m({...S,quote:!1}),{chosen:w,evidence:D,warning:j}=A(C,C.sawQuoteChar||!C.ok?await m({...S,quote:`"`}):null),N=[];j!==null&&N.push(j);let P=w.columns;if(P.length===0)throw new T(`${t} parsed to zero columns.`,`The file may be empty below its header.`);let F=a(P,s.fields),I=new Map(s.fields.map(e=>[e.key,e])),L=[],R=[];for(let e of P){let t=F.get(e);if(t!==void 0){let n=I.get(t);L.push({source:e,target:t,transforms:E(n?.type??`string`)});continue}R.push(M(e))}let z=L.filter(e=>I.get(e.target)?.type===`number`).map(e=>e.source),B=[];for(let e of z)for(let t of w.samples.get(e)??[])B.push(t);let V=x(B);V.format===null&&N.push(`The number format could NOT be decided (${V.confidence}): ${V.evidence} Supply { decimal, thousands } explicitly before saving.`),V.unparsed.length>0&&N.push(`${V.unparsed.length}+ sampled value(s) in the number columns do not parse under the inferred format: ${V.unparsed.join(`, `)}. Those rows will be row errors at import.`),z.length===0&&N.push(`No column mapped to a numeric target, so the number format was inferred from nothing. Check it before saving.`);let H=b?i({columns:[...P],rows:w.sampleRows}):null;return{version:1,file:{path:t,bytes:u},targetSchema:o,reader:{format:v,delimiter:_,hasHeader:b,quote:w.quote},numberFormat:V.format,numberFormatConfidence:V.confidence,columnMap:L,unmapped:R,headerSnapshot:H,dataRows:w.rows,truncated:e.maxRows!==void 0&&e.maxRows>0&&w.rows>=e.maxRows,evidence:{delimiter:g.evidence,hasHeader:y.evidence,quote:D,numberFormat:V.evidence},warnings:N}}function M(e){return C.has(D(e))?{source:e,reason:`"${e}" names a replacement code but not its DIRECTION, and the two readings are opposites: the column may hold the NEWER code that supersedes this row (superseding_code), or the OLDER code this row supersedes (replaces_code). This is deliberately not guessed — an inverted feed builds well-formed chains with valid heads and no warnings, and the merge is append-only, so re-importing the right way round does NOT repair it. Check a superseded code against the supplier's current catalogue and state the direction.`,candidates:[...w]}:{source:e,reason:`no target field matches "${e}".`,candidates:[]}}function N(e){return e.unmapped.map(e=>e.source)}export{s as CANDIDATE_DELIMITERS,c as DIALECT_SAMPLE_BYTES,l as DIALECT_SAMPLE_ROWS,S as FEED_ANALYSIS_VERSION,T as FeedAnalysisError,j as analyzeFeed,u as delimiterName,d as inferDelimiter,f as inferHasHeader,x as inferNumberFormat,p as sampleRecords,m as trialParse,N as unmappedSources};
 //# sourceMappingURL=index.mjs.map

package/dist/index.d.mts

@@ -230,14 +230,14 @@ declare const ImportMapping: import("@murumets-ee/entity").Entity<{
   id: import("@murumets-ee/entity").IdField;
 } & import("@murumets-ee/entity").AuditableFields & Record<never, never> & {
   /** FK to commerce `supplier` (by name string — no cross-package import). Required. */supplierId: import("@murumets-ee/entity").ReferenceField & {
-    cardinality: "one" | "many";
+    cardinality: "many" | "one";
   } & {
     readonly entity: "supplier";
     readonly required: true;
     readonly indexed: true;
   }; /** FK to commerce `brand`. NULL = supplier default (applies to all that supplier's brands). */
   brandId: import("@murumets-ee/entity").ReferenceField & {
-    cardinality: "one" | "many";
+    cardinality: "many" | "one";
   } & {
     readonly entity: "brand";
     readonly indexed: true;

package/dist/index.mjs

@@ -1,2 +1,2 @@
-import{a as e,c as t,i as n,l as r,n as i,o as a,r as o,s,t as c,u as l}from"./transform-Cu5YEoOl.mjs";import{a as u,n as d,o as f,r as p,t as m}from"./read-feed-DGjqejw3.mjs";import{n as h,t as g}from"./cleanup-DDljBe0Z.mjs";import{n as _,r as v}from"./engine-B0ExE4jL.mjs";import{a as y,c as b,d as x,f as S,i as C,l as w,n as T,o as E,r as D,s as O,t as k,u as A}from"./runner-C9GIe5Dr.mjs";import{n as j}from"./carmaker-feed-B-QYr3PI.mjs";import{DEFAULT_BOUNDED_CONCURRENCY as M,runBounded as N}from"@murumets-ee/core";const P=[`corrupt`,`clean`,`source_unavailable`,`undetectable`,`error`],F=[`succeeded`],I=5e3;function L(e){return e instanceof Error?e.message:String(e)}const R=new Set([`ENOENT`,`EACCES`,`EPERM`,`EISDIR`,`ENOTDIR`]),z=/FILE_ENDED/;function B(e,t){if(typeof e!=`object`||!e)return!1;if(e instanceof t)return!0;if(`code`in e){let{code:t}=e;if(typeof t==`string`&&R.has(t))return!0}return e instanceof Error&&z.test(e.message)}async function V(e){let{filePath:t,feed:n,resolveFilePath:r}=e;if(n.format!==`xlsx`)return{verdict:`undetectable`,detail:`not an xlsx run (format: ${n.format??`tsv (default)`}) — the defect is exceljs-specific`};if(n.hasHeader===!1)return{verdict:`undetectable`,detail:`run used hasHeader:false — no header row for the guard to test`};let{streamXlsx:i,XlsxEmptyHeaderError:a,XlsxNotAFileError:o}=await import(`./xlsx-reader-_ioPAqm5.mjs`),s;try{s=r?await r(t):{localPath:t}}catch(e){return{verdict:`source_unavailable`,detail:L(e)}}try{let e=!1;for await(let t of i({filePath:s.localPath,...n.headerRowIndex!==void 0&&{headerRowIndex:n.headerRowIndex},...n.sheet!==void 0&&{sheet:n.sheet}})){e=!0;break}return e?{verdict:`clean`}:{verdict:`undetectable`,detail:`file yielded no data rows — the header guard never ran`}}catch(e){return e instanceof a?{verdict:`corrupt`,detail:e.message}:B(e,o)?{verdict:`source_unavailable`,detail:L(e)}:{verdict:`error`,detail:L(e)}}finally{await s.cleanup?.().catch(()=>{})}}function H(e,t){let n=e[t];return typeof n==`number`&&Number.isFinite(n)?n:void 0}function U(e){if(typeof e!=`object`||!e||Array.isArray(e))return{};let t=e;return{rowsRead:H(t,`rowsRead`),rowsSucceeded:H(t,`rowsSucceeded`),rowsFailed:H(t,`rowsFailed`),rowsSkipped:H(t,`rowsSkipped`)}}function W(){return{corrupt:0,clean:0,source_unavailable:0,undetectable:0,error:0}}function G(e){return e.rowsRead===void 0||e.rowsSucceeded===void 0?`unknown`:e.rowsRead>0&&e.rowsSucceeded===0}function K(e){if(e instanceof Date)return e;if(typeof e==`string`){let t=new Date(e);return Number.isNaN(t.getTime())?null:t}return null}async function q(e){let{app:t,resolveFilePath:n,logger:r}=e,i=e.statuses??F;if(i.length===0)throw Error(`auditEmptyHeaderRuns: statuses must name at least one import_run status`);let a=e.limit??200;if(!Number.isInteger(a)||a<1||a>I)throw Error(`auditEmptyHeaderRuns: limit must be an integer in [1, ${I}] (got ${a})`);let o=e.concurrency??M;if(!Number.isInteger(o)||o<1||o>16)throw Error(`auditEmptyHeaderRuns: concurrency must be an integer in [1, 16] (got ${o})`);let s=t.getClient(l),{and:c,desc:u,getTableColumns:d,inArray:f,sql:m}=await import(`drizzle-orm`),h=d(s.getTable()),g=h.status,_=h.params,v=h.createdAt;if(!g||!_||!v)throw Error(`auditEmptyHeaderRuns: import_run schema missing expected columns (status/params/createdAt)`);let y=c(f(g,[...i]),m`${_}->'feed'->>'format' = ${`xlsx`}`),b=await s.count({where:y}),x=await s.findMany({where:y,orderBy:u(v),limit:a}),S=Math.max(0,b-x.length);if(S>0&&r.warn({candidates:b,examined:x.length,notExamined:S,limit:a},`imports audit: capped at the most recent runs — re-run with a higher limit to cover the rest`),x.length===0)return r.info({statuses:i},`imports audit: no xlsx runs to audit`),{candidates:b,examined:0,notExamined:S,counts:W(),runs:[]};r.info({candidates:b,examining:x.length,concurrency:o},`imports audit: re-reading source files for the empty-header defect`);let C=await N(x,async e=>{let t=e.filePath,r=U(e.totals),i={id:e.id,label:e.label,filePath:t,transformName:e.transformName,createdAt:K(e.createdAt),finishedAt:K(e.finishedAt),totals:r,zeroRowsIndexed:G(r)};try{let r=p(e.params);if(!r.success){let e=r.error.issues[0]?.message??`invalid feed reader config`;return{...i,verdict:`error`,detail:`invalid params.feed: ${e}`}}let a=await V({filePath:t,feed:r.data,...n!==void 0&&{resolveFilePath:n}});return{...i,verdict:a.verdict,...a.detail!==void 0&&{detail:a.detail}}}catch(e){return{...i,verdict:`error`,detail:L(e)}}},o),w=W();for(let e of C)w[e.verdict]+=1,e.verdict===`corrupt`&&r.warn({importRunId:e.id,label:e.label,filePath:e.filePath,transformName:e.transformName,createdAt:e.createdAt,totals:e.totals},`imports audit: SILENTLY CORRUPTED run — its source file has no readable header`);let T=C.filter(e=>e.zeroRowsIndexed===!0).length,E=C.filter(e=>e.zeroRowsIndexed===`unknown`).length;return n===void 0&&w.source_unavailable===C.length&&r.warn({examined:C.length},`imports audit: every source was unreadable and no resolveFilePath was given — if filePath holds storage keys rather than local paths, pass storageResolveFilePath`),r.info({candidates:b,examined:C.length,notExamined:S,...w,zeroRowsIndexed:T,totalsUnknown:E},`imports audit: complete`),{candidates:b,examined:C.length,notExamined:S,counts:w,runs:C}}const J=100,Y=20;async function X(e){let{maxRows:t,...n}=e,r=t??20,i=Math.min(100,Math.max(1,Number.isFinite(r)?Math.floor(r):20)),a=[],o=new Map;for await(let{row:e}of d({...n,relaxColumnCount:!0})){for(let t of Object.keys(e))o.set(t,!0);if(a.push(e),a.length>=i)break}return{columns:[...o.keys()],rows:a}}export{j as CARMAKER_MAPPING,k as DEFAULT_BATCH_SIZE,w as DEFAULT_MAX_PATTERNS,A as DEFAULT_MAX_SAMPLES_PER_PATTERN,Y as DEFAULT_PREVIEW_ROWS,P as EMPTY_HEADER_VERDICTS,D as EMPTY_PRICE_STATS,x as ErrorTracker,n as IMPORT_MAPPING_FORMATS,e as IMPORT_MAPPING_TARGET_SCHEMAS,r as IMPORT_RUN_STATUSES,a as ImportMapping,l as ImportRun,J as MAX_PREVIEW_ROWS,C as PRICE_ANOMALY_THRESHOLD_MULTIPLIER,y as PRICE_SAMPLE_CAP,E as PriceStatsCollector,c as TransformRegistry,s as assertValidMapping,q as auditEmptyHeaderRuns,g as cleanupOrphanedImports,h as cleanupProcessedImports,u as countDataRows,O as detectPriceAnomaly,m as feedReaderConfigSchema,i as getTransformRegistry,t as importMappingValidation,V as inspectImportRunSource,_ as mappingColumnSchema,v as mappingConfigSchema,S as mergeErrorTrackerSnapshots,b as parsePriceStats,X as previewFeed,d as readFeed,p as readPersistedFeedConfig,U as readRecordedTotals,o as registerImportTransform,T as runImport,f as streamFeed};
+import{a as e,c as t,i as n,l as r,n as i,o as a,r as o,s,t as c,u as l}from"./transform-Cu5YEoOl.mjs";import{a as u,n as d,o as f,r as p,t as m}from"./read-feed-DsLgRDNR.mjs";import{n as h,t as g}from"./cleanup-DDljBe0Z.mjs";import{n as _,r as v}from"./engine-B0ExE4jL.mjs";import{a as y,c as b,d as x,f as S,i as C,l as w,n as T,o as E,r as D,s as O,t as k,u as A}from"./runner-BnFvID1z.mjs";import{n as j}from"./carmaker-feed-B-QYr3PI.mjs";import{DEFAULT_BOUNDED_CONCURRENCY as M,runBounded as N}from"@murumets-ee/core";const P=[`corrupt`,`clean`,`source_unavailable`,`undetectable`,`error`],F=[`succeeded`],I=5e3;function L(e){return e instanceof Error?e.message:String(e)}const R=new Set([`ENOENT`,`EACCES`,`EPERM`,`EISDIR`,`ENOTDIR`]),z=/FILE_ENDED/;function B(e,t){if(typeof e!=`object`||!e)return!1;if(e instanceof t)return!0;if(`code`in e){let{code:t}=e;if(typeof t==`string`&&R.has(t))return!0}return e instanceof Error&&z.test(e.message)}async function V(e){let{filePath:t,feed:n,resolveFilePath:r}=e;if(n.format!==`xlsx`)return{verdict:`undetectable`,detail:`not an xlsx run (format: ${n.format??`tsv (default)`}) — the defect is exceljs-specific`};if(n.hasHeader===!1)return{verdict:`undetectable`,detail:`run used hasHeader:false — no header row for the guard to test`};let{streamXlsx:i,XlsxEmptyHeaderError:a,XlsxNotAFileError:o}=await import(`./xlsx-reader-D9Ntxm29.mjs`),s;try{s=r?await r(t):{localPath:t}}catch(e){return{verdict:`source_unavailable`,detail:L(e)}}try{let e=!1;for await(let t of i({filePath:s.localPath,...n.headerRowIndex!==void 0&&{headerRowIndex:n.headerRowIndex},...n.sheet!==void 0&&{sheet:n.sheet}})){e=!0;break}return e?{verdict:`clean`}:{verdict:`undetectable`,detail:`file yielded no data rows — the header guard never ran`}}catch(e){return e instanceof a?{verdict:`corrupt`,detail:e.message}:B(e,o)?{verdict:`source_unavailable`,detail:L(e)}:{verdict:`error`,detail:L(e)}}finally{await s.cleanup?.().catch(()=>{})}}function H(e,t){let n=e[t];return typeof n==`number`&&Number.isFinite(n)?n:void 0}function U(e){if(typeof e!=`object`||!e||Array.isArray(e))return{};let t=e;return{rowsRead:H(t,`rowsRead`),rowsSucceeded:H(t,`rowsSucceeded`),rowsFailed:H(t,`rowsFailed`),rowsSkipped:H(t,`rowsSkipped`)}}function W(){return{corrupt:0,clean:0,source_unavailable:0,undetectable:0,error:0}}function G(e){return e.rowsRead===void 0||e.rowsSucceeded===void 0?`unknown`:e.rowsRead>0&&e.rowsSucceeded===0}function K(e){if(e instanceof Date)return e;if(typeof e==`string`){let t=new Date(e);return Number.isNaN(t.getTime())?null:t}return null}async function q(e){let{app:t,resolveFilePath:n,logger:r}=e,i=e.statuses??F;if(i.length===0)throw Error(`auditEmptyHeaderRuns: statuses must name at least one import_run status`);let a=e.limit??200;if(!Number.isInteger(a)||a<1||a>I)throw Error(`auditEmptyHeaderRuns: limit must be an integer in [1, ${I}] (got ${a})`);let o=e.concurrency??M;if(!Number.isInteger(o)||o<1||o>16)throw Error(`auditEmptyHeaderRuns: concurrency must be an integer in [1, 16] (got ${o})`);let s=t.getClient(l),{and:c,desc:u,getTableColumns:d,inArray:f,sql:m}=await import(`drizzle-orm`),h=d(s.getTable()),g=h.status,_=h.params,v=h.createdAt;if(!g||!_||!v)throw Error(`auditEmptyHeaderRuns: import_run schema missing expected columns (status/params/createdAt)`);let y=c(f(g,[...i]),m`${_}->'feed'->>'format' = ${`xlsx`}`),b=await s.count({where:y}),x=await s.findMany({where:y,orderBy:u(v),limit:a}),S=Math.max(0,b-x.length);if(S>0&&r.warn({candidates:b,examined:x.length,notExamined:S,limit:a},`imports audit: capped at the most recent runs — re-run with a higher limit to cover the rest`),x.length===0)return r.info({statuses:i},`imports audit: no xlsx runs to audit`),{candidates:b,examined:0,notExamined:S,counts:W(),runs:[]};r.info({candidates:b,examining:x.length,concurrency:o},`imports audit: re-reading source files for the empty-header defect`);let C=await N(x,async e=>{let t=e.filePath,r=U(e.totals),i={id:e.id,label:e.label,filePath:t,transformName:e.transformName,createdAt:K(e.createdAt),finishedAt:K(e.finishedAt),totals:r,zeroRowsIndexed:G(r)};try{let r=p(e.params);if(!r.success){let e=r.error.issues[0]?.message??`invalid feed reader config`;return{...i,verdict:`error`,detail:`invalid params.feed: ${e}`}}let a=await V({filePath:t,feed:r.data,...n!==void 0&&{resolveFilePath:n}});return{...i,verdict:a.verdict,...a.detail!==void 0&&{detail:a.detail}}}catch(e){return{...i,verdict:`error`,detail:L(e)}}},o),w=W();for(let e of C)w[e.verdict]+=1,e.verdict===`corrupt`&&r.warn({importRunId:e.id,label:e.label,filePath:e.filePath,transformName:e.transformName,createdAt:e.createdAt,totals:e.totals},`imports audit: SILENTLY CORRUPTED run — its source file has no readable header`);let T=C.filter(e=>e.zeroRowsIndexed===!0).length,E=C.filter(e=>e.zeroRowsIndexed===`unknown`).length;return n===void 0&&w.source_unavailable===C.length&&r.warn({examined:C.length},`imports audit: every source was unreadable and no resolveFilePath was given — if filePath holds storage keys rather than local paths, pass storageResolveFilePath`),r.info({candidates:b,examined:C.length,notExamined:S,...w,zeroRowsIndexed:T,totalsUnknown:E},`imports audit: complete`),{candidates:b,examined:C.length,notExamined:S,counts:w,runs:C}}const J=100,Y=20;async function X(e){let{maxRows:t,...n}=e,r=t??20,i=Math.min(100,Math.max(1,Number.isFinite(r)?Math.floor(r):20)),a=[],o=new Map;for await(let{row:e}of d({...n,relaxColumnCount:!0})){for(let t of Object.keys(e))o.set(t,!0);if(a.push(e),a.length>=i)break}return{columns:[...o.keys()],rows:a}}export{j as CARMAKER_MAPPING,k as DEFAULT_BATCH_SIZE,w as DEFAULT_MAX_PATTERNS,A as DEFAULT_MAX_SAMPLES_PER_PATTERN,Y as DEFAULT_PREVIEW_ROWS,P as EMPTY_HEADER_VERDICTS,D as EMPTY_PRICE_STATS,x as ErrorTracker,n as IMPORT_MAPPING_FORMATS,e as IMPORT_MAPPING_TARGET_SCHEMAS,r as IMPORT_RUN_STATUSES,a as ImportMapping,l as ImportRun,J as MAX_PREVIEW_ROWS,C as PRICE_ANOMALY_THRESHOLD_MULTIPLIER,y as PRICE_SAMPLE_CAP,E as PriceStatsCollector,c as TransformRegistry,s as assertValidMapping,q as auditEmptyHeaderRuns,g as cleanupOrphanedImports,h as cleanupProcessedImports,u as countDataRows,O as detectPriceAnomaly,m as feedReaderConfigSchema,i as getTransformRegistry,t as importMappingValidation,V as inspectImportRunSource,_ as mappingColumnSchema,v as mappingConfigSchema,S as mergeErrorTrackerSnapshots,b as parsePriceStats,X as previewFeed,d as readFeed,p as readPersistedFeedConfig,U as readRecordedTotals,o as registerImportTransform,T as runImport,f as streamFeed};
 //# sourceMappingURL=index.mjs.map

package/dist/{read-feed-DGjqejw3.mjs → read-feed-DsLgRDNR.mjs}

@@ -1,2 +1,2 @@
-import{z as e}from"zod";import{createReadStream as t}from"node:fs";import{parse as n}from"csv-parse";function r(e){let t=new Set;for(let n of e)if(n!==``){if(t.has(n))throw Error(`Duplicate header "${n}" after trimming — column names must be unique`);t.add(n)}}async function i(e,n={}){let{hasHeader:r=!0}=n,i=t(e),a=0,o=0;for await(let e of i){for(let t=0;t<e.length;t+=1)e[t]===10&&(a+=1);e.length>0&&(o=e[e.length-1]??0)}return o!==0&&o!==10&&(a+=1),r?Math.max(0,a-1):a}async function*a(e){let{filePath:i,delimiter:a=`	`,hasHeader:o=!0,columns:s,relaxColumnCount:c=!1,quote:l=!1}=e,u=s?Array.from(s):o?e=>{let t=e.map(e=>e.trim());return r(t),t}:!1,d=t(i),f=d.pipe(n({delimiter:a,columns:u,bom:!0,skip_empty_lines:!0,relax_column_count:c,quote:l})),p=0;try{for await(let e of f){p+=1;let t={};if(Array.isArray(e)){let n=e;for(let e=0;e<n.length;e+=1)t[String(e)]=n[e]??``}else for(let[n,r]of Object.entries(e))t[n]=r??``;yield{rowNumber:p,row:t}}}finally{d.destroy()}}const o=e.object({format:e.enum([`csv`,`tsv`,`xlsx`]).optional(),delimiter:e.string().min(1).max(4).optional(),hasHeader:e.boolean().optional(),headerRowIndex:e.number().int().min(0).optional(),sheet:e.string().min(1).max(128).optional(),quote:e.union([e.string().length(1),e.literal(!1)]).optional()}).strict();function s(e){let t=e&&typeof e==`object`&&!Array.isArray(e)?e:{},n=t.feed===void 0||t.feed===null?{}:t.feed;return o.safeParse(n)}function c(e){let t=e.delimiter??(e.format===`csv`?`,`:`	`);return{filePath:e.filePath,delimiter:t,...e.hasHeader!==void 0&&{hasHeader:e.hasHeader},...e.columns!==void 0&&{columns:e.columns},...e.relaxColumnCount!==void 0&&{relaxColumnCount:e.relaxColumnCount},...e.quote!==void 0&&{quote:e.quote}}}async function*l(e){if(e.format===`xlsx`){let{streamXlsx:t}=await import(`./xlsx-reader-_ioPAqm5.mjs`);yield*t({filePath:e.filePath,...e.hasHeader!==void 0&&{hasHeader:e.hasHeader},...e.headerRowIndex!==void 0&&{headerRowIndex:e.headerRowIndex},...e.sheet!==void 0&&{sheet:e.sheet}});return}yield*a(c(e))}export{i as a,r as i,l as n,a as o,s as r,o as t};
-//# sourceMappingURL=read-feed-DGjqejw3.mjs.map
+import{z as e}from"zod";import{createReadStream as t}from"node:fs";import{parse as n}from"csv-parse";function r(e){let t=new Set;for(let n of e)if(n!==``){if(t.has(n))throw Error(`Duplicate header "${n}" after trimming — column names must be unique`);t.add(n)}}async function i(e,n={}){let{hasHeader:r=!0}=n,i=t(e),a=0,o=0;for await(let e of i){for(let t=0;t<e.length;t+=1)e[t]===10&&(a+=1);e.length>0&&(o=e[e.length-1]??0)}return o!==0&&o!==10&&(a+=1),r?Math.max(0,a-1):a}async function*a(e){let{filePath:i,delimiter:a=`	`,hasHeader:o=!0,columns:s,relaxColumnCount:c=!1,quote:l=!1}=e,u=s?Array.from(s):o?e=>{let t=e.map(e=>e.trim());return r(t),t}:!1,d=t(i),f=d.pipe(n({delimiter:a,columns:u,bom:!0,skip_empty_lines:!0,relax_column_count:c,quote:l})),p=0;try{for await(let e of f){p+=1;let t={};if(Array.isArray(e)){let n=e;for(let e=0;e<n.length;e+=1)t[String(e)]=n[e]??``}else for(let[n,r]of Object.entries(e))t[n]=r??``;yield{rowNumber:p,row:t}}}finally{d.destroy()}}const o=e.object({format:e.enum([`csv`,`tsv`,`xlsx`]).optional(),delimiter:e.string().min(1).max(4).optional(),hasHeader:e.boolean().optional(),headerRowIndex:e.number().int().min(0).optional(),sheet:e.string().min(1).max(128).optional(),quote:e.union([e.string().length(1),e.literal(!1)]).optional()}).strict();function s(e){let t=e&&typeof e==`object`&&!Array.isArray(e)?e:{},n=t.feed===void 0||t.feed===null?{}:t.feed;return o.safeParse(n)}function c(e){let t=e.delimiter??(e.format===`csv`?`,`:`	`);return{filePath:e.filePath,delimiter:t,...e.hasHeader!==void 0&&{hasHeader:e.hasHeader},...e.columns!==void 0&&{columns:e.columns},...e.relaxColumnCount!==void 0&&{relaxColumnCount:e.relaxColumnCount},...e.quote!==void 0&&{quote:e.quote}}}async function*l(e){if(e.format===`xlsx`){let{streamXlsx:t}=await import(`./xlsx-reader-D9Ntxm29.mjs`);yield*t({filePath:e.filePath,...e.hasHeader!==void 0&&{hasHeader:e.hasHeader},...e.headerRowIndex!==void 0&&{headerRowIndex:e.headerRowIndex},...e.sheet!==void 0&&{sheet:e.sheet}});return}yield*a(c(e))}export{i as a,r as i,l as n,a as o,s as r,o as t};
+//# sourceMappingURL=read-feed-DsLgRDNR.mjs.map

package/dist/{read-feed-DGjqejw3.mjs.map → read-feed-DsLgRDNR.mjs.map}

@@ -1 +1 @@
-{"version":3,"file":"read-feed-DGjqejw3.mjs","names":[],"sources":["../src/streaming.ts","../src/read-feed.ts"],"sourcesContent":["/**\n * Tab-delimited / CSV streaming reader. Generic over delimiter so the\n * same path handles `.txt` (tab), `.csv` (comma), and the rare `;`\n * European export dialect.\n *\n * Built on `csv-parse` per giga-test precedent — node-stream-based, low\n * memory, handles UTF-8 BOM (the carmaker feeds are Windows-exported\n * and ship with a BOM that breaks naive split-on-tab parsers).\n *\n * The reader yields `{ rowNumber, row }` pairs where:\n * - `rowNumber` is 1-based and counts the header as row 0.\n * - `row` is `Record<string, string>` keyed by header name. Empty cells\n *   are the empty string, NOT `undefined` — feed transforms test with\n *   `value === ''` consistently.\n *\n * Why this lives in a streaming reader and not inside the transform\n * itself: the transform sees one already-parsed row at a time, never\n * the file. That keeps transform implementations free of I/O concerns\n * and makes them trivially unit-testable with a fixture row map.\n */\n\nimport { createReadStream } from 'node:fs'\nimport { parse } from 'csv-parse'\n\n/**\n * Throw if two (non-empty) header names collide. Headers are the keys the mapping\n * engine matches a `source` against; after trimming, two columns differing only in\n * whitespace would collapse to one key and silently overwrite each other's data.\n * Fail fast instead — the operator fixes the feed or the mapping. Shared by the\n * csv/tsv reader (below) and the xlsx reader. Blank columns are ignored (unmapped).\n */\nexport function assertUniqueHeaders(header: readonly string[]): void {\n  const seen = new Set<string>()\n  for (const h of header) {\n    if (h === '') continue\n    if (seen.has(h)) {\n      throw new Error(`Duplicate header \"${h}\" after trimming — column names must be unique`)\n    }\n    seen.add(h)\n  }\n}\n\n/**\n * Pre-pass: count data rows in a delimited file (excluding the header\n * row when `hasHeader` is true; default `true` to match `streamFeed`).\n *\n * Implementation: stream the file as raw bytes and tally `\\n`. NOT a\n * full parse — escaped newlines inside quoted fields would over-count.\n * That was unconditionally fine while `quote: false` was the only\n * behaviour available; a mapping can now declare RFC 4180 (`quote`),\n * and an embedded newline in a quoted field then makes this an UPPER\n * BOUND rather than a count. It is used only as the progress\n * denominator (`totalRows`), so the visible consequence is a bar that\n * stops short of 100%, never a wrong import. The parsed row count from\n * `streamFeed` is authoritative and is what the run records.\n *\n * Why bytes-only and not csv-parse: a 16 MB feed counts in ~50 ms with\n * `\\n` tally vs ~500 ms with csv-parse parsing. For 500 MB the gap\n * matters — pre-pass needs to be a small fraction of the actual import\n * or the operator perceives the run as \"stuck on pre-pass\" with no UI.\n *\n * Returns the number of DATA rows (header excluded when present).\n * Empty trailing lines (no `\\n` after the final row, or a single\n * trailing `\\n`) are handled identically to how csv-parse treats them\n * with `skip_empty_lines: true`.\n */\nexport async function countDataRows(\n  filePath: string,\n  options: { hasHeader?: boolean } = {},\n): Promise<number> {\n  const { hasHeader = true } = options\n  const stream = createReadStream(filePath)\n  let count = 0\n  let lastByte = 0\n  for await (const chunk of stream as AsyncIterable<Buffer>) {\n    for (let i = 0; i < chunk.length; i += 1) {\n      if (chunk[i] === 0x0a /* \\n */) count += 1\n    }\n    if (chunk.length > 0) lastByte = chunk[chunk.length - 1] ?? 0\n  }\n  // If the file doesn't end with a newline, the final row didn't get\n  // counted by the `\\n` tally — add one. Skipping when the file is\n  // entirely empty (count=0 and lastByte=0).\n  if (lastByte !== 0 && lastByte !== 0x0a) count += 1\n  return hasHeader ? Math.max(0, count - 1) : count\n}\n\nexport interface StreamFeedOptions {\n  /** Path to the file on disk. The PoC uploads land on local disk; S3-keyed reads come later. */\n  filePath: string\n  /**\n   * Single-character field delimiter. Default `\\t` (the carmaker feed\n   * format). Pass `,` for CSV, `;` for some European dialects.\n   */\n  delimiter?: string\n  /**\n   * `true` (default): the first row is the header and column names come\n   * from it. `false`: rows are emitted as positional `{ \"0\": ..., \"1\": ... }`\n   * and the transform reads by index — useful for headerless feeds that\n   * commit to a documented column order.\n   */\n  hasHeader?: boolean\n  /**\n   * Optional explicit column-name list. When provided, takes precedence\n   * over `hasHeader` (header row, if present, is skipped but its values\n   * are ignored). Useful when the upstream header is unstable but the\n   * positional shape isn't.\n   */\n  columns?: ReadonlyArray<string>\n  /**\n   * Forward to `csv-parse` `relax_column_count`. Default `false` —\n   * a row whose column count doesn't match the header surfaces as a\n   * parser error so the transform isn't silently fed truncated data.\n   */\n  relaxColumnCount?: boolean\n  /**\n   * Forward to `csv-parse` `quote`. The character (or `false`) that\n   * encloses fields containing the delimiter. Default `false` — most\n   * supplier feeds in this package's target use case are tab-delimited\n   * exports that contain literal `\"` characters in description fields\n   * (e.g. `LIVRE \"INVERSION D'IMAGE\"` from the Mercedes carmaker feed)\n   * with NO quote-as-field-wrapper convention. csv-parse's library\n   * default `'\"'` then misreads the literal `\"` as an opening quote\n   * and rejects the row with `Invalid Opening Quote`.\n   *\n   * Pass `'\"'` (or any character) to opt back into RFC-4180-style\n   * CSV parsing for genuine quoted-CSV inputs. Pass `'\\''` for\n   * single-quoted dialects.\n   */\n  quote?: string | false\n}\n\nexport interface StreamFeedRow {\n  /** 1-based row number. Header (when present) is row 0; first data row is row 1. */\n  rowNumber: number\n  /**\n   * Cell values keyed by column name (or string-position when\n   * `hasHeader: false` AND no `columns`).\n   *\n   * **Cell-value invariants:**\n   * - Empty cells (`A\\t\\tC`) → `''` (empty string).\n   * - Missing TRAILING cells in `relaxColumnCount: true` mode → the\n   *   key is **absent** from the object, not present-with-`''`. csv-parse\n   *   does not emit keys for short rows. Transforms reading those\n   *   columns get `undefined` from `row['col']` and must handle it\n   *   (`row['col'] ?? ''` is the canonical idiom).\n   * - With the default `relaxColumnCount: false`, short rows reject at\n   *   the parser, so this case never reaches the transform.\n   */\n  row: Record<string, string>\n}\n\n/**\n * Async-iterable over the parsed rows of a delimited file. Use with\n * `for await (const { rowNumber, row } of streamFeed({ filePath, ... }))`.\n *\n * The iterator owns its file descriptor — the `for await` loop closes\n * the underlying stream when it returns or breaks. Aborting mid-stream\n * (`break`, `throw`, signal) is safe; csv-parse propagates the close.\n */\nexport async function* streamFeed(options: StreamFeedOptions): AsyncIterable<StreamFeedRow> {\n  const {\n    filePath,\n    delimiter = '\\t',\n    hasHeader = true,\n    columns,\n    relaxColumnCount = false,\n    quote = false,\n  } = options\n\n  // When the caller provides explicit `columns`, prefer them. When the file\n  // has a header but no explicit `columns`, derive the column names from the\n  // first row via a function so we can TRIM them — real supplier feeds ship\n  // leading/trailing-space headers (R03 Samples A/C: `\" JAEHIND KM-TA\"`), and\n  // an untrimmed key wouldn't match a mapping's `source`. The xlsx reader trims\n  // headers the same way. When neither is true, rows get string-position keys.\n  //\n  // csv-parse's `columns` accepts `ColumnOption[] | boolean | ((record) =>\n  // ColumnOption[])`; `string[]` satisfies `ColumnOption[]`.\n  const columnConfig: string[] | boolean | ((header: string[]) => string[]) = columns\n    ? Array.from(columns)\n    : hasHeader\n      ? (header: string[]) => {\n          const trimmed = header.map((h) => h.trim())\n          assertUniqueHeaders(trimmed)\n          return trimmed\n        }\n      : false\n\n  const stream = createReadStream(filePath)\n  const parser = stream.pipe(\n    parse({\n      delimiter,\n      columns: columnConfig,\n      bom: true,\n      skip_empty_lines: true,\n      // csv-parse defaults to strict column count; opt-in relaxation only.\n      relax_column_count: relaxColumnCount,\n      // Default `quote: false` — see StreamFeedOptions.quote JSDoc.\n      // Carmaker / parts-feed exports contain literal `\"` characters\n      // in description fields and would otherwise reject as\n      // `Invalid Opening Quote`. Genuine RFC-4180 CSV consumers opt\n      // back in by passing `quote: '\"'`.\n      quote,\n    }),\n  )\n\n  let rowNumber = 0\n  try {\n    for await (const rawRow of parser as AsyncIterable<\n      ReadonlyArray<string> | Record<string, string | undefined>\n    >) {\n      rowNumber += 1\n      // csv-parse emits records with string keys when `columns` is\n      // truthy, otherwise an Array. Normalise both shapes to\n      // `Record<string, string>` so transforms can rely on `value === ''`\n      // for missing cells (csv-parse leaves trailing missing cells as\n      // `undefined` when `relax_column_count: true`; this collapses\n      // them to `''` to keep the contract uniform).\n      const row: Record<string, string> = {}\n      if (Array.isArray(rawRow)) {\n        const arr = rawRow as ReadonlyArray<string | undefined>\n        for (let i = 0; i < arr.length; i += 1) {\n          row[String(i)] = arr[i] ?? ''\n        }\n      } else {\n        for (const [k, v] of Object.entries(rawRow)) {\n          row[k] = v ?? ''\n        }\n      }\n      yield { rowNumber, row }\n    }\n  } finally {\n    // Safety: ensure the underlying file descriptor closes even if the\n    // consumer breaks mid-iteration. Node closes streams on\n    // garbage-collection but this makes it deterministic under tests.\n    stream.destroy()\n  }\n}\n","/**\n * Feed reader dispatcher (commerce-demo phase 02, D005). Selects the reader by\n * `format` and yields the uniform `AsyncIterable<{ rowNumber, row }>` the runner\n * consumes — csv/tsv via the existing `streamFeed` (csv-parse), xlsx via the\n * exceljs-backed `streamXlsx`.\n *\n * **Back-compat**: with no `format` (the carmaker worker passes only `filePath`),\n * this defaults to tab-delimited — the original `streamFeed` behaviour — so the\n * existing import path is unchanged.\n *\n * **Bundle safety**: the xlsx reader (and its heavy, Node-only exceljs dep) is\n * `await import()`-ed ONLY for `format: 'xlsx'`, so the csv/tsv path — and anything\n * that statically imports the runner — never pulls exceljs into its module graph.\n */\n\nimport { z } from 'zod'\nimport { type StreamFeedOptions, type StreamFeedRow, streamFeed } from './streaming.js'\n\n/**\n * The reader-config subset safe to persist in `import_run.params.feed` and rehydrate\n * in the worker (commerce-demo PR3). It is exactly the {@link ReadFeedOptions} fields\n * that describe HOW to parse a feed — `filePath` is supplied by the worker after it\n * materialises the upload, never carried in params. Validated at BOTH ends: the\n * commerce run route builds it from a saved `import_mapping`, the worker re-parses it\n * before handing it to {@link readFeed} (defense-in-depth against a tampered row).\n */\nexport const feedReaderConfigSchema = z\n  .object({\n    format: z.enum(['csv', 'tsv', 'xlsx']).optional(),\n    delimiter: z.string().min(1).max(4).optional(),\n    hasHeader: z.boolean().optional(),\n    headerRowIndex: z.number().int().min(0).optional(),\n    sheet: z.string().min(1).max(128).optional(),\n    /**\n     * csv-parse's `quote`. Absent (or `false`) = quotes are ordinary data, which is this\n     * package's long-standing default and what the tab-delimited supplier dumps require. A\n     * single character opts into RFC 4180 for feeds that genuinely quote their fields.\n     *\n     * Persisted deliberately: the two shapes are mutually exclusive on real data (R027 —\n     * each of the three ground-truth feeds hard-fails under the other's setting), so this\n     * is a per-file fact a mapping has to carry. Deriving it from `format` would be the\n     * same class of mistake as a global decimal separator.\n     */\n    quote: z.union([z.string().length(1), z.literal(false)]).optional(),\n  })\n  .strict()\nexport type FeedReaderConfig = z.infer<typeof feedReaderConfigSchema>\n\n/**\n * Rehydrate the reader config a run persisted in `import_run.params.feed`.\n *\n * ONE implementation, because two would drift and drift here is silent: the\n * worker uses it to decide how to READ a feed, and the #842 audit uses it to\n * re-read the same file the same way. An audit that rehydrates differently from\n * the run it is auditing invents corruption, or misses it.\n *\n * ABSENT means \"the carmaker default\", i.e. tab-delimited, which is what an\n * empty config parses to — so a non-object `params`, or `params` with no\n * `feed`, is a real answer rather than a gap. A PRESENT `feed` is validated\n * whatever its shape: a scalar or an array there is a malformed persisted blob,\n * and reading it as \"no config\" would silently import an xlsx run as tsv.\n */\nexport function readPersistedFeedConfig(\n  params: unknown,\n): ReturnType<typeof feedReaderConfigSchema.safeParse> {\n  const record =\n    params && typeof params === 'object' && !Array.isArray(params)\n      ? (params as Record<string, unknown>)\n      : {}\n  // `null` is how \"absent\" round-trips through JSONB, so it defaults with\n  // `undefined`. Everything else present goes to the schema and is rejected on\n  // its merits rather than being flattened into the default.\n  const feed = record.feed === undefined || record.feed === null ? {} : record.feed\n  return feedReaderConfigSchema.safeParse(feed)\n}\n\nexport interface ReadFeedOptions {\n  /** Path to the file on local disk. */\n  filePath: string\n  /** Feed format. Omitted → tab-delimited (tsv), matching the legacy carmaker reader. */\n  format?: 'csv' | 'tsv' | 'xlsx'\n  /** Explicit delimiter override (csv/tsv). Defaults from `format` (`csv` → `,`, else `\\t`). */\n  delimiter?: string\n  /** First row is the header. Default `true`. */\n  hasHeader?: boolean\n  /** 0-based header row index (xlsx). */\n  headerRowIndex?: number\n  /** xlsx worksheet as a 1-based index string. Omit → first sheet. */\n  sheet?: string\n  /** Explicit column-name list (csv/tsv) — see `streamFeed`. */\n  columns?: ReadonlyArray<string>\n  /** Forward to csv-parse `relax_column_count` (csv/tsv). */\n  relaxColumnCount?: boolean\n  /** Forward to csv-parse `quote` (csv/tsv). */\n  quote?: string | false\n}\n\nfunction toStreamFeedOptions(o: ReadFeedOptions): StreamFeedOptions {\n  // No format → tab (the carmaker feed default). `csv` → comma; everything else → tab.\n  const delimiter = o.delimiter ?? (o.format === 'csv' ? ',' : '\\t')\n  return {\n    filePath: o.filePath,\n    delimiter,\n    ...(o.hasHeader !== undefined && { hasHeader: o.hasHeader }),\n    ...(o.columns !== undefined && { columns: o.columns }),\n    ...(o.relaxColumnCount !== undefined && { relaxColumnCount: o.relaxColumnCount }),\n    ...(o.quote !== undefined && { quote: o.quote }),\n  }\n}\n\n/** Stream a feed file's data rows, dispatching on `format`. */\nexport async function* readFeed(options: ReadFeedOptions): AsyncIterable<StreamFeedRow> {\n  if (options.format === 'xlsx') {\n    const { streamXlsx } = await import('./xlsx-reader.js')\n    yield* streamXlsx({\n      filePath: options.filePath,\n      ...(options.hasHeader !== undefined && { hasHeader: options.hasHeader }),\n      ...(options.headerRowIndex !== undefined && { headerRowIndex: options.headerRowIndex }),\n      ...(options.sheet !== undefined && { sheet: options.sheet }),\n    })\n    return\n  }\n  yield* streamFeed(toStreamFeedOptions(options))\n}\n"],"mappings":"qGA+BA,SAAgB,EAAoB,EAAiC,CACnE,IAAM,EAAO,IAAI,IACjB,IAAK,IAAM,KAAK,EACV,OAAM,GACV,IAAI,EAAK,IAAI,CAAC,EACZ,MAAU,MAAM,qBAAqB,EAAE,+CAA+C,EAExF,EAAK,IAAI,CAAC,CAF8E,CAI5F,CA0BA,eAAsB,EACpB,EACA,EAAmC,CAAC,EACnB,CACjB,GAAM,CAAE,YAAY,IAAS,EACvB,EAAS,EAAiB,CAAQ,EACpC,EAAQ,EACR,EAAW,EACf,UAAW,IAAM,KAAS,EAAiC,CACzD,IAAK,IAAI,EAAI,EAAG,EAAI,EAAM,OAAQ,GAAK,EACjC,EAAM,KAAO,KAAe,GAAS,GAEvC,EAAM,OAAS,IAAG,EAAW,EAAM,EAAM,OAAS,IAAM,EAC9D,CAKA,OADI,IAAa,GAAK,IAAa,KAAM,GAAS,GAC3C,EAAY,KAAK,IAAI,EAAG,EAAQ,CAAC,EAAI,CAC9C,CA2EA,eAAuB,EAAW,EAA0D,CAC1F,GAAM,CACJ,WACA,YAAY,IACZ,YAAY,GACZ,UACA,mBAAmB,GACnB,QAAQ,IACN,EAWE,EAAsE,EACxE,MAAM,KAAK,CAAO,EAClB,EACG,GAAqB,CACpB,IAAM,EAAU,EAAO,IAAK,GAAM,EAAE,KAAK,CAAC,EAE1C,OADA,EAAoB,CAAO,EACpB,CACT,EACA,GAEA,EAAS,EAAiB,CAAQ,EAClC,EAAS,EAAO,KACpB,EAAM,CACJ,YACA,QAAS,EACT,IAAK,GACL,iBAAkB,GAElB,mBAAoB,EAMpB,OACF,CAAC,CACH,EAEI,EAAY,EAChB,GAAI,CACF,UAAW,IAAM,KAAU,EAExB,CACD,GAAa,EAOb,IAAM,EAA8B,CAAC,EACrC,GAAI,MAAM,QAAQ,CAAM,EAAG,CACzB,IAAM,EAAM,EACZ,IAAK,IAAI,EAAI,EAAG,EAAI,EAAI,OAAQ,GAAK,EACnC,EAAI,OAAO,CAAC,GAAK,EAAI,IAAM,EAE/B,MACE,IAAK,GAAM,CAAC,EAAG,KAAM,OAAO,QAAQ,CAAM,EACxC,EAAI,GAAK,GAAK,GAGlB,KAAM,CAAE,YAAW,KAAI,CACzB,CACF,QAAU,CAIR,EAAO,QAAQ,CACjB,CACF,CCpNA,MAAa,EAAyB,EACnC,OAAO,CACN,OAAQ,EAAE,KAAK,CAAC,MAAO,MAAO,MAAM,CAAC,CAAC,CAAC,SAAS,EAChD,UAAW,EAAE,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,SAAS,EAC7C,UAAW,EAAE,QAAQ,CAAC,CAAC,SAAS,EAChC,eAAgB,EAAE,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,SAAS,EACjD,MAAO,EAAE,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,GAAG,CAAC,CAAC,SAAS,EAW3C,MAAO,EAAE,MAAM,CAAC,EAAE,OAAO,CAAC,CAAC,OAAO,CAAC,EAAG,EAAE,QAAQ,EAAK,CAAC,CAAC,CAAC,CAAC,SAAS,CACpE,CAAC,CAAC,CACD,OAAO,EAiBV,SAAgB,EACd,EACqD,CACrD,IAAM,EACJ,GAAU,OAAO,GAAW,UAAY,CAAC,MAAM,QAAQ,CAAM,EACxD,EACD,CAAC,EAID,EAAO,EAAO,OAAS,IAAA,IAAa,EAAO,OAAS,KAAO,CAAC,EAAI,EAAO,KAC7E,OAAO,EAAuB,UAAU,CAAI,CAC9C,CAuBA,SAAS,EAAoB,EAAuC,CAElE,IAAM,EAAY,EAAE,YAAc,EAAE,SAAW,MAAQ,IAAM,KAC7D,MAAO,CACL,SAAU,EAAE,SACZ,YACA,GAAI,EAAE,YAAc,IAAA,IAAa,CAAE,UAAW,EAAE,SAAU,EAC1D,GAAI,EAAE,UAAY,IAAA,IAAa,CAAE,QAAS,EAAE,OAAQ,EACpD,GAAI,EAAE,mBAAqB,IAAA,IAAa,CAAE,iBAAkB,EAAE,gBAAiB,EAC/E,GAAI,EAAE,QAAU,IAAA,IAAa,CAAE,MAAO,EAAE,KAAM,CAChD,CACF,CAGA,eAAuB,EAAS,EAAwD,CACtF,GAAI,EAAQ,SAAW,OAAQ,CAC7B,GAAM,CAAE,cAAe,MAAM,OAAO,8BACpC,MAAO,EAAW,CAChB,SAAU,EAAQ,SAClB,GAAI,EAAQ,YAAc,IAAA,IAAa,CAAE,UAAW,EAAQ,SAAU,EACtE,GAAI,EAAQ,iBAAmB,IAAA,IAAa,CAAE,eAAgB,EAAQ,cAAe,EACrF,GAAI,EAAQ,QAAU,IAAA,IAAa,CAAE,MAAO,EAAQ,KAAM,CAC5D,CAAC,EACD,MACF,CACA,MAAO,EAAW,EAAoB,CAAO,CAAC,CAChD"}
+{"version":3,"file":"read-feed-DsLgRDNR.mjs","names":[],"sources":["../src/streaming.ts","../src/read-feed.ts"],"sourcesContent":["/**\n * Tab-delimited / CSV streaming reader. Generic over delimiter so the\n * same path handles `.txt` (tab), `.csv` (comma), and the rare `;`\n * European export dialect.\n *\n * Built on `csv-parse` per giga-test precedent — node-stream-based, low\n * memory, handles UTF-8 BOM (the carmaker feeds are Windows-exported\n * and ship with a BOM that breaks naive split-on-tab parsers).\n *\n * The reader yields `{ rowNumber, row }` pairs where:\n * - `rowNumber` is 1-based and counts the header as row 0.\n * - `row` is `Record<string, string>` keyed by header name. Empty cells\n *   are the empty string, NOT `undefined` — feed transforms test with\n *   `value === ''` consistently.\n *\n * Why this lives in a streaming reader and not inside the transform\n * itself: the transform sees one already-parsed row at a time, never\n * the file. That keeps transform implementations free of I/O concerns\n * and makes them trivially unit-testable with a fixture row map.\n */\n\nimport { createReadStream } from 'node:fs'\nimport { parse } from 'csv-parse'\n\n/**\n * Throw if two (non-empty) header names collide. Headers are the keys the mapping\n * engine matches a `source` against; after trimming, two columns differing only in\n * whitespace would collapse to one key and silently overwrite each other's data.\n * Fail fast instead — the operator fixes the feed or the mapping. Shared by the\n * csv/tsv reader (below) and the xlsx reader. Blank columns are ignored (unmapped).\n */\nexport function assertUniqueHeaders(header: readonly string[]): void {\n  const seen = new Set<string>()\n  for (const h of header) {\n    if (h === '') continue\n    if (seen.has(h)) {\n      throw new Error(`Duplicate header \"${h}\" after trimming — column names must be unique`)\n    }\n    seen.add(h)\n  }\n}\n\n/**\n * Pre-pass: count data rows in a delimited file (excluding the header\n * row when `hasHeader` is true; default `true` to match `streamFeed`).\n *\n * Implementation: stream the file as raw bytes and tally `\\n`. NOT a\n * full parse — escaped newlines inside quoted fields would over-count.\n * That was unconditionally fine while `quote: false` was the only\n * behaviour available; a mapping can now declare RFC 4180 (`quote`),\n * and an embedded newline in a quoted field then makes this an UPPER\n * BOUND rather than a count. It is used only as the progress\n * denominator (`totalRows`), so the visible consequence is a bar that\n * stops short of 100%, never a wrong import. The parsed row count from\n * `streamFeed` is authoritative and is what the run records.\n *\n * Why bytes-only and not csv-parse: a 16 MB feed counts in ~50 ms with\n * `\\n` tally vs ~500 ms with csv-parse parsing. For 500 MB the gap\n * matters — pre-pass needs to be a small fraction of the actual import\n * or the operator perceives the run as \"stuck on pre-pass\" with no UI.\n *\n * Returns the number of DATA rows (header excluded when present).\n * Empty trailing lines (no `\\n` after the final row, or a single\n * trailing `\\n`) are handled identically to how csv-parse treats them\n * with `skip_empty_lines: true`.\n */\nexport async function countDataRows(\n  filePath: string,\n  options: { hasHeader?: boolean } = {},\n): Promise<number> {\n  const { hasHeader = true } = options\n  const stream = createReadStream(filePath)\n  let count = 0\n  let lastByte = 0\n  for await (const chunk of stream as AsyncIterable<Buffer>) {\n    for (let i = 0; i < chunk.length; i += 1) {\n      if (chunk[i] === 0x0a /* \\n */) count += 1\n    }\n    if (chunk.length > 0) lastByte = chunk[chunk.length - 1] ?? 0\n  }\n  // If the file doesn't end with a newline, the final row didn't get\n  // counted by the `\\n` tally — add one. Skipping when the file is\n  // entirely empty (count=0 and lastByte=0).\n  if (lastByte !== 0 && lastByte !== 0x0a) count += 1\n  return hasHeader ? Math.max(0, count - 1) : count\n}\n\nexport interface StreamFeedOptions {\n  /** Path to the file on disk. The PoC uploads land on local disk; S3-keyed reads come later. */\n  filePath: string\n  /**\n   * Single-character field delimiter. Default `\\t` (the carmaker feed\n   * format). Pass `,` for CSV, `;` for some European dialects.\n   */\n  delimiter?: string\n  /**\n   * `true` (default): the first row is the header and column names come\n   * from it. `false`: rows are emitted as positional `{ \"0\": ..., \"1\": ... }`\n   * and the transform reads by index — useful for headerless feeds that\n   * commit to a documented column order.\n   */\n  hasHeader?: boolean\n  /**\n   * Optional explicit column-name list. When provided, takes precedence\n   * over `hasHeader` (header row, if present, is skipped but its values\n   * are ignored). Useful when the upstream header is unstable but the\n   * positional shape isn't.\n   */\n  columns?: ReadonlyArray<string>\n  /**\n   * Forward to `csv-parse` `relax_column_count`. Default `false` —\n   * a row whose column count doesn't match the header surfaces as a\n   * parser error so the transform isn't silently fed truncated data.\n   */\n  relaxColumnCount?: boolean\n  /**\n   * Forward to `csv-parse` `quote`. The character (or `false`) that\n   * encloses fields containing the delimiter. Default `false` — most\n   * supplier feeds in this package's target use case are tab-delimited\n   * exports that contain literal `\"` characters in description fields\n   * (e.g. `LIVRE \"INVERSION D'IMAGE\"` from the Mercedes carmaker feed)\n   * with NO quote-as-field-wrapper convention. csv-parse's library\n   * default `'\"'` then misreads the literal `\"` as an opening quote\n   * and rejects the row with `Invalid Opening Quote`.\n   *\n   * Pass `'\"'` (or any character) to opt back into RFC-4180-style\n   * CSV parsing for genuine quoted-CSV inputs. Pass `'\\''` for\n   * single-quoted dialects.\n   */\n  quote?: string | false\n}\n\nexport interface StreamFeedRow {\n  /** 1-based row number. Header (when present) is row 0; first data row is row 1. */\n  rowNumber: number\n  /**\n   * Cell values keyed by column name (or string-position when\n   * `hasHeader: false` AND no `columns`).\n   *\n   * **Cell-value invariants:**\n   * - Empty cells (`A\\t\\tC`) → `''` (empty string).\n   * - Missing TRAILING cells in `relaxColumnCount: true` mode → the\n   *   key is **absent** from the object, not present-with-`''`. csv-parse\n   *   does not emit keys for short rows. Transforms reading those\n   *   columns get `undefined` from `row['col']` and must handle it\n   *   (`row['col'] ?? ''` is the canonical idiom).\n   * - With the default `relaxColumnCount: false`, short rows reject at\n   *   the parser, so this case never reaches the transform.\n   */\n  row: Record<string, string>\n}\n\n/**\n * Async-iterable over the parsed rows of a delimited file. Use with\n * `for await (const { rowNumber, row } of streamFeed({ filePath, ... }))`.\n *\n * The iterator owns its file descriptor — the `for await` loop closes\n * the underlying stream when it returns or breaks. Aborting mid-stream\n * (`break`, `throw`, signal) is safe; csv-parse propagates the close.\n */\nexport async function* streamFeed(options: StreamFeedOptions): AsyncIterable<StreamFeedRow> {\n  const {\n    filePath,\n    delimiter = '\\t',\n    hasHeader = true,\n    columns,\n    relaxColumnCount = false,\n    quote = false,\n  } = options\n\n  // When the caller provides explicit `columns`, prefer them. When the file\n  // has a header but no explicit `columns`, derive the column names from the\n  // first row via a function so we can TRIM them — real supplier feeds ship\n  // leading/trailing-space headers (R03 Samples A/C: `\" JAEHIND KM-TA\"`), and\n  // an untrimmed key wouldn't match a mapping's `source`. The xlsx reader trims\n  // headers the same way. When neither is true, rows get string-position keys.\n  //\n  // csv-parse's `columns` accepts `ColumnOption[] | boolean | ((record) =>\n  // ColumnOption[])`; `string[]` satisfies `ColumnOption[]`.\n  const columnConfig: string[] | boolean | ((header: string[]) => string[]) = columns\n    ? Array.from(columns)\n    : hasHeader\n      ? (header: string[]) => {\n          const trimmed = header.map((h) => h.trim())\n          assertUniqueHeaders(trimmed)\n          return trimmed\n        }\n      : false\n\n  const stream = createReadStream(filePath)\n  const parser = stream.pipe(\n    parse({\n      delimiter,\n      columns: columnConfig,\n      bom: true,\n      skip_empty_lines: true,\n      // csv-parse defaults to strict column count; opt-in relaxation only.\n      relax_column_count: relaxColumnCount,\n      // Default `quote: false` — see StreamFeedOptions.quote JSDoc.\n      // Carmaker / parts-feed exports contain literal `\"` characters\n      // in description fields and would otherwise reject as\n      // `Invalid Opening Quote`. Genuine RFC-4180 CSV consumers opt\n      // back in by passing `quote: '\"'`.\n      quote,\n    }),\n  )\n\n  let rowNumber = 0\n  try {\n    for await (const rawRow of parser as AsyncIterable<\n      ReadonlyArray<string> | Record<string, string | undefined>\n    >) {\n      rowNumber += 1\n      // csv-parse emits records with string keys when `columns` is\n      // truthy, otherwise an Array. Normalise both shapes to\n      // `Record<string, string>` so transforms can rely on `value === ''`\n      // for missing cells (csv-parse leaves trailing missing cells as\n      // `undefined` when `relax_column_count: true`; this collapses\n      // them to `''` to keep the contract uniform).\n      const row: Record<string, string> = {}\n      if (Array.isArray(rawRow)) {\n        const arr = rawRow as ReadonlyArray<string | undefined>\n        for (let i = 0; i < arr.length; i += 1) {\n          row[String(i)] = arr[i] ?? ''\n        }\n      } else {\n        for (const [k, v] of Object.entries(rawRow)) {\n          row[k] = v ?? ''\n        }\n      }\n      yield { rowNumber, row }\n    }\n  } finally {\n    // Safety: ensure the underlying file descriptor closes even if the\n    // consumer breaks mid-iteration. Node closes streams on\n    // garbage-collection but this makes it deterministic under tests.\n    stream.destroy()\n  }\n}\n","/**\n * Feed reader dispatcher (commerce-demo phase 02, D005). Selects the reader by\n * `format` and yields the uniform `AsyncIterable<{ rowNumber, row }>` the runner\n * consumes — csv/tsv via the existing `streamFeed` (csv-parse), xlsx via the\n * exceljs-backed `streamXlsx`.\n *\n * **Back-compat**: with no `format` (the carmaker worker passes only `filePath`),\n * this defaults to tab-delimited — the original `streamFeed` behaviour — so the\n * existing import path is unchanged.\n *\n * **Bundle safety**: the xlsx reader (and its heavy, Node-only exceljs dep) is\n * `await import()`-ed ONLY for `format: 'xlsx'`, so the csv/tsv path — and anything\n * that statically imports the runner — never pulls exceljs into its module graph.\n */\n\nimport { z } from 'zod'\nimport { type StreamFeedOptions, type StreamFeedRow, streamFeed } from './streaming.js'\n\n/**\n * The reader-config subset safe to persist in `import_run.params.feed` and rehydrate\n * in the worker (commerce-demo PR3). It is exactly the {@link ReadFeedOptions} fields\n * that describe HOW to parse a feed — `filePath` is supplied by the worker after it\n * materialises the upload, never carried in params. Validated at BOTH ends: the\n * commerce run route builds it from a saved `import_mapping`, the worker re-parses it\n * before handing it to {@link readFeed} (defense-in-depth against a tampered row).\n */\nexport const feedReaderConfigSchema = z\n  .object({\n    format: z.enum(['csv', 'tsv', 'xlsx']).optional(),\n    delimiter: z.string().min(1).max(4).optional(),\n    hasHeader: z.boolean().optional(),\n    headerRowIndex: z.number().int().min(0).optional(),\n    sheet: z.string().min(1).max(128).optional(),\n    /**\n     * csv-parse's `quote`. Absent (or `false`) = quotes are ordinary data, which is this\n     * package's long-standing default and what the tab-delimited supplier dumps require. A\n     * single character opts into RFC 4180 for feeds that genuinely quote their fields.\n     *\n     * Persisted deliberately: the two shapes are mutually exclusive on real data (R027 —\n     * each of the three ground-truth feeds hard-fails under the other's setting), so this\n     * is a per-file fact a mapping has to carry. Deriving it from `format` would be the\n     * same class of mistake as a global decimal separator.\n     */\n    quote: z.union([z.string().length(1), z.literal(false)]).optional(),\n  })\n  .strict()\nexport type FeedReaderConfig = z.infer<typeof feedReaderConfigSchema>\n\n/**\n * Rehydrate the reader config a run persisted in `import_run.params.feed`.\n *\n * ONE implementation, because two would drift and drift here is silent: the\n * worker uses it to decide how to READ a feed, and the #842 audit uses it to\n * re-read the same file the same way. An audit that rehydrates differently from\n * the run it is auditing invents corruption, or misses it.\n *\n * ABSENT means \"the carmaker default\", i.e. tab-delimited, which is what an\n * empty config parses to — so a non-object `params`, or `params` with no\n * `feed`, is a real answer rather than a gap. A PRESENT `feed` is validated\n * whatever its shape: a scalar or an array there is a malformed persisted blob,\n * and reading it as \"no config\" would silently import an xlsx run as tsv.\n */\nexport function readPersistedFeedConfig(\n  params: unknown,\n): ReturnType<typeof feedReaderConfigSchema.safeParse> {\n  const record =\n    params && typeof params === 'object' && !Array.isArray(params)\n      ? (params as Record<string, unknown>)\n      : {}\n  // `null` is how \"absent\" round-trips through JSONB, so it defaults with\n  // `undefined`. Everything else present goes to the schema and is rejected on\n  // its merits rather than being flattened into the default.\n  const feed = record.feed === undefined || record.feed === null ? {} : record.feed\n  return feedReaderConfigSchema.safeParse(feed)\n}\n\nexport interface ReadFeedOptions {\n  /** Path to the file on local disk. */\n  filePath: string\n  /** Feed format. Omitted → tab-delimited (tsv), matching the legacy carmaker reader. */\n  format?: 'csv' | 'tsv' | 'xlsx'\n  /** Explicit delimiter override (csv/tsv). Defaults from `format` (`csv` → `,`, else `\\t`). */\n  delimiter?: string\n  /** First row is the header. Default `true`. */\n  hasHeader?: boolean\n  /** 0-based header row index (xlsx). */\n  headerRowIndex?: number\n  /** xlsx worksheet as a 1-based index string. Omit → first sheet. */\n  sheet?: string\n  /** Explicit column-name list (csv/tsv) — see `streamFeed`. */\n  columns?: ReadonlyArray<string>\n  /** Forward to csv-parse `relax_column_count` (csv/tsv). */\n  relaxColumnCount?: boolean\n  /** Forward to csv-parse `quote` (csv/tsv). */\n  quote?: string | false\n}\n\nfunction toStreamFeedOptions(o: ReadFeedOptions): StreamFeedOptions {\n  // No format → tab (the carmaker feed default). `csv` → comma; everything else → tab.\n  const delimiter = o.delimiter ?? (o.format === 'csv' ? ',' : '\\t')\n  return {\n    filePath: o.filePath,\n    delimiter,\n    ...(o.hasHeader !== undefined && { hasHeader: o.hasHeader }),\n    ...(o.columns !== undefined && { columns: o.columns }),\n    ...(o.relaxColumnCount !== undefined && { relaxColumnCount: o.relaxColumnCount }),\n    ...(o.quote !== undefined && { quote: o.quote }),\n  }\n}\n\n/** Stream a feed file's data rows, dispatching on `format`. */\nexport async function* readFeed(options: ReadFeedOptions): AsyncIterable<StreamFeedRow> {\n  if (options.format === 'xlsx') {\n    const { streamXlsx } = await import('./xlsx-reader.js')\n    yield* streamXlsx({\n      filePath: options.filePath,\n      ...(options.hasHeader !== undefined && { hasHeader: options.hasHeader }),\n      ...(options.headerRowIndex !== undefined && { headerRowIndex: options.headerRowIndex }),\n      ...(options.sheet !== undefined && { sheet: options.sheet }),\n    })\n    return\n  }\n  yield* streamFeed(toStreamFeedOptions(options))\n}\n"],"mappings":"qGA+BA,SAAgB,EAAoB,EAAiC,CACnE,IAAM,EAAO,IAAI,IACjB,IAAK,IAAM,KAAK,EACV,OAAM,GACV,IAAI,EAAK,IAAI,CAAC,EACZ,MAAU,MAAM,qBAAqB,EAAE,+CAA+C,EAExF,EAAK,IAAI,CAAC,CAF8E,CAI5F,CA0BA,eAAsB,EACpB,EACA,EAAmC,CAAC,EACnB,CACjB,GAAM,CAAE,YAAY,IAAS,EACvB,EAAS,EAAiB,CAAQ,EACpC,EAAQ,EACR,EAAW,EACf,UAAW,IAAM,KAAS,EAAiC,CACzD,IAAK,IAAI,EAAI,EAAG,EAAI,EAAM,OAAQ,GAAK,EACjC,EAAM,KAAO,KAAe,GAAS,GAEvC,EAAM,OAAS,IAAG,EAAW,EAAM,EAAM,OAAS,IAAM,EAC9D,CAKA,OADI,IAAa,GAAK,IAAa,KAAM,GAAS,GAC3C,EAAY,KAAK,IAAI,EAAG,EAAQ,CAAC,EAAI,CAC9C,CA2EA,eAAuB,EAAW,EAA0D,CAC1F,GAAM,CACJ,WACA,YAAY,IACZ,YAAY,GACZ,UACA,mBAAmB,GACnB,QAAQ,IACN,EAWE,EAAsE,EACxE,MAAM,KAAK,CAAO,EAClB,EACG,GAAqB,CACpB,IAAM,EAAU,EAAO,IAAK,GAAM,EAAE,KAAK,CAAC,EAE1C,OADA,EAAoB,CAAO,EACpB,CACT,EACA,GAEA,EAAS,EAAiB,CAAQ,EAClC,EAAS,EAAO,KACpB,EAAM,CACJ,YACA,QAAS,EACT,IAAK,GACL,iBAAkB,GAElB,mBAAoB,EAMpB,OACF,CAAC,CACH,EAEI,EAAY,EAChB,GAAI,CACF,UAAW,IAAM,KAAU,EAExB,CACD,GAAa,EAOb,IAAM,EAA8B,CAAC,EACrC,GAAI,MAAM,QAAQ,CAAM,EAAG,CACzB,IAAM,EAAM,EACZ,IAAK,IAAI,EAAI,EAAG,EAAI,EAAI,OAAQ,GAAK,EACnC,EAAI,OAAO,CAAC,GAAK,EAAI,IAAM,EAE/B,MACE,IAAK,GAAM,CAAC,EAAG,KAAM,OAAO,QAAQ,CAAM,EACxC,EAAI,GAAK,GAAK,GAGlB,KAAM,CAAE,YAAW,KAAI,CACzB,CACF,QAAU,CAIR,EAAO,QAAQ,CACjB,CACF,CCpNA,MAAa,EAAyB,EACnC,OAAO,CACN,OAAQ,EAAE,KAAK,CAAC,MAAO,MAAO,MAAM,CAAC,CAAC,CAAC,SAAS,EAChD,UAAW,EAAE,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,SAAS,EAC7C,UAAW,EAAE,QAAQ,CAAC,CAAC,SAAS,EAChC,eAAgB,EAAE,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,SAAS,EACjD,MAAO,EAAE,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,GAAG,CAAC,CAAC,SAAS,EAW3C,MAAO,EAAE,MAAM,CAAC,EAAE,OAAO,CAAC,CAAC,OAAO,CAAC,EAAG,EAAE,QAAQ,EAAK,CAAC,CAAC,CAAC,CAAC,SAAS,CACpE,CAAC,CAAC,CACD,OAAO,EAiBV,SAAgB,EACd,EACqD,CACrD,IAAM,EACJ,GAAU,OAAO,GAAW,UAAY,CAAC,MAAM,QAAQ,CAAM,EACxD,EACD,CAAC,EAID,EAAO,EAAO,OAAS,IAAA,IAAa,EAAO,OAAS,KAAO,CAAC,EAAI,EAAO,KAC7E,OAAO,EAAuB,UAAU,CAAI,CAC9C,CAuBA,SAAS,EAAoB,EAAuC,CAElE,IAAM,EAAY,EAAE,YAAc,EAAE,SAAW,MAAQ,IAAM,KAC7D,MAAO,CACL,SAAU,EAAE,SACZ,YACA,GAAI,EAAE,YAAc,IAAA,IAAa,CAAE,UAAW,EAAE,SAAU,EAC1D,GAAI,EAAE,UAAY,IAAA,IAAa,CAAE,QAAS,EAAE,OAAQ,EACpD,GAAI,EAAE,mBAAqB,IAAA,IAAa,CAAE,iBAAkB,EAAE,gBAAiB,EAC/E,GAAI,EAAE,QAAU,IAAA,IAAa,CAAE,MAAO,EAAE,KAAM,CAChD,CACF,CAGA,eAAuB,EAAS,EAAwD,CACtF,GAAI,EAAQ,SAAW,OAAQ,CAC7B,GAAM,CAAE,cAAe,MAAM,OAAO,8BACpC,MAAO,EAAW,CAChB,SAAU,EAAQ,SAClB,GAAI,EAAQ,YAAc,IAAA,IAAa,CAAE,UAAW,EAAQ,SAAU,EACtE,GAAI,EAAQ,iBAAmB,IAAA,IAAa,CAAE,eAAgB,EAAQ,cAAe,EACrF,GAAI,EAAQ,QAAU,IAAA,IAAa,CAAE,MAAO,EAAQ,KAAM,CAC5D,CAAC,EACD,MACF,CACA,MAAO,EAAW,EAAoB,CAAO,CAAC,CAChD"}

package/dist/{runner-C9GIe5Dr.mjs → runner-BnFvID1z.mjs}

@@ -1,2 +1,2 @@
-import{n as e}from"./read-feed-DGjqejw3.mjs";import{bulkPartialUpdate as t,bulkUpsert as n}from"@murumets-ee/search-elasticsearch";const r=50,i=5;var a=class{patterns=new Map;maxPatterns;maxSamplesPerPattern;droppedSignatures=0;constructor(e={}){this.maxPatterns=e.maxPatterns??50,this.maxSamplesPerPattern=e.maxSamplesPerPattern??5}addError(e,t,n,r,i){let a=`${t}:${r??`GENERAL`}:${n}`,o=this.patterns.get(a);if(!o){if(this.patterns.size>=this.maxPatterns){this.droppedSignatures+=1;return}o={errorType:t,field:r??null,message:n,count:0,firstOccurrence:e,lastOccurrence:e,samples:[]},this.patterns.set(a,o)}o.count+=1,o.lastOccurrence=e,o.samples.length<this.maxSamplesPerPattern&&o.samples.push({rowNumber:e,rowData:i})}getTotalErrorCount(){let e=0;for(let t of this.patterns.values())e+=t.count;return e}getDistinctPatternCount(){return this.patterns.size}getDroppedSignatureCount(){return this.droppedSignatures}getTopPatterns(){let e=Array.from(this.patterns.values()).sort((e,t)=>t.count-e.count),t=e.reduce((e,t)=>e+t.count,0);return e.map(e=>({errorType:e.errorType,field:e.field,message:e.message,count:e.count,firstOccurrence:e.firstOccurrence,lastOccurrence:e.lastOccurrence,samples:e.samples.slice(),percentage:t>0?e.count/t*100:0}))}snapshot(){return{totalErrors:this.getTotalErrorCount(),distinctPatterns:this.getDistinctPatternCount(),droppedSignatures:this.droppedSignatures,patterns:this.getTopPatterns()}}};function o(e,t){let n=e.totalErrors+t.totalErrors,r=e=>({...e,percentage:n>0?e.count/n*100:0});return{totalErrors:n,distinctPatterns:e.distinctPatterns+t.distinctPatterns,droppedSignatures:e.droppedSignatures+t.droppedSignatures,patterns:[...e.patterns.map(r),...t.patterns.map(r)]}}const s={count:0,nullRate:0,min:null,median:null,max:null},c=5e4;var l=class{count=0;nullCount=0;nonNullSeen=0;samples=[];record(e){if(this.count+=1,e===void 0){this.nullCount+=1;return}if(this.nonNullSeen+=1,this.samples.length<5e4){this.samples.push(e);return}let t=Math.floor(Math.random()*this.nonNullSeen);t<5e4&&(this.samples[t]=e)}snapshot(){if(this.count===0)return{...s};let e=this.nullCount/this.count;if(this.samples.length===0)return{count:this.count,nullRate:e,min:null,median:null,max:null};let t=[...this.samples].sort((e,t)=>e-t),n=t[0]??null,r=t[t.length-1]??null,i=Math.floor(t.length/2),a;if(t.length%2==0){let e=t[i-1],n=t[i];a=e!==void 0&&n!==void 0?(e+n)/2:null}else a=t[i]??null;return{count:this.count,nullRate:e,min:n,median:a,max:r}}};const u=10;function d(e,t){if(e.median===null||t.median===null||e.median<=0||t.median<=0)return null;let n=t.median/e.median;return n>=10||n<=1/10?{ratio:n,prevMedian:e.median,currMedian:t.median}:null}function f(e){if(typeof e!=`object`||!e)return null;let t=e;return typeof t.count!=`number`||typeof t.nullRate!=`number`?null:{count:t.count,nullRate:t.nullRate,min:typeof t.min==`number`?t.min:null,median:typeof t.median==`number`?t.median:null,max:typeof t.max==`number`?t.max:null}}const p=1e3;async function m(r){let{importRunId:i,runLabel:o,params:s,transform:c,feed:u,esClient:d,esIndex:f,batchSize:m=p,onProgress:h,rowLimit:g,signal:_,errorTracker:v=new a,warningTracker:y=new a,priceStats:b=new l,logger:x}=r;if(m<1)throw Error(`batchSize must be >= 1 (got ${m})`);let S=Date.now();x?.info({importRunId:i,runLabel:o,batchSize:m,esIndex:f},`imports.runner: started`);let C=0,w=0,T=0,E=0,D=0,O=0,k=[],A=[],j=[],M=async()=>{if(j.length===0)return;let e=j;j=[];let t=Date.now(),r;try{r=await n(d,{index:f,docs:e.map(({id:e,doc:t})=>({id:e,doc:t})),..._!==void 0&&{signal:_}})}catch(t){let n=_?.aborted??(t instanceof Error&&(t.name===`AbortError`||/abort/i.test(t.message)))?`aborted`:`bulk_request_failed`,r=t instanceof Error?t.message:String(t);for(let{rowNumber:t}of e)v.addError(t,n,r,void 0,null);throw T+=e.length,t}w+=r.succeeded,T+=r.failures.length,O+=1;let a=Date.now()-t,o=(Date.now()-S)/1e3;if(x?.info({importRunId:i,batch:O,flushMs:a,submitted:e.length,succeeded:r.succeeded,failed:r.failures.length,rowsRead:C,rowsPerSecond:o>0?Math.round(C/o):0},`imports.runner: batch flushed`),r.failures.length>0){let t=new Map(e.map(e=>[e.id,e.rowNumber]));for(let e of r.failures){let n=t.get(e.id)??-1;v.addError(n,e.type,e.reason,void 0,{id:e.id})}}let s=new Set(r.failures.map(e=>e.id));for(let t of e)if(!(!t.warnings||t.warnings.length===0)&&!s.has(t.id)){D+=1;for(let e of t.warnings)y.addError(t.rowNumber,e.warningType,e.message,e.field,t.rowData)}for(let t of e)!t.pairs||t.pairs.length===0||s.has(t.id)||k.push(...t.pairs);for(let t of e)s.has(t.id)||b.record(t.priceSample);if(h){let e=(Date.now()-S)/1e3;h({rowsRead:C,rowsSucceeded:w,rowsFailed:T,rowsSkipped:E,rowsWarned:D,batchesCompleted:O,elapsedSeconds:e,rowsPerSecond:e>0?C/e:0,distinctErrorPatterns:v.getDistinctPatternCount()})}},N=m,P=new Map;for await(let{rowNumber:t,row:n}of e(u)){if(_?.aborted||g!==void 0&&C>=g)break;if(C+=1,C===1&&x?.info({importRunId:i,rowNumber:t,sampleKeys:Object.keys(n)},`imports.runner: STREAM_FIRST_ROW`),C%N===0){let e=(Date.now()-S)/1e3;x?.info({importRunId:i,rowsRead:C,rowsSucceeded:w,rowsFailed:T,rowsSkipped:E,pending:j.length,elapsedSec:Math.round(e*10)/10,rowsPerSecond:e>0?Math.round(C/e):0,distinctFirstErrorTypes:P.size},`imports.runner: STREAM_TICK`)}let e={importRunId:i,params:s,runLabel:o,rowNumber:t},r;try{r=await c(n,e)}catch(e){let r=e instanceof Error?e.message:String(e);v.addError(t,`transform_threw`,r,void 0,n),P.has(`transform_threw`)||(P.set(`transform_threw`,{rowNumber:t,message:r}),x?.warn({importRunId:i,rowNumber:t,errorType:`transform_threw`,message:r},`imports.runner: FIRST_ERROR transform_threw`)),T+=1;continue}if(r.kind===`skip`){E+=1,r.patches&&r.patches.length>0&&A.push(...r.patches);continue}if(r.kind===`error`){v.addError(t,r.error.errorType,r.error.message,r.error.field,n),P.has(r.error.errorType)||(P.set(r.error.errorType,{rowNumber:t,message:r.error.message}),x?.warn({importRunId:i,rowNumber:t,errorType:r.error.errorType,field:r.error.field,message:r.error.message},`imports.runner: FIRST_ERROR ${r.error.errorType}`)),T+=1;continue}j.push({id:r.id,doc:r.doc,rowNumber:t,rowData:n,...r.warnings&&r.warnings.length>0&&{warnings:r.warnings},...r.pairs&&r.pairs.length>0&&{pairs:r.pairs},...r.priceSample!==void 0&&{priceSample:r.priceSample}}),j.length>=m&&await M()}await M();let F=0,I=0;for(let e=0;e<A.length;e+=m){if(_?.aborted)throw Error(`imports: run aborted during core-charge attach (phase 2)`);let n=A.slice(e,e+m),r;try{r=await t(d,{index:f,updates:n.map(({id:e,doc:t})=>({id:e,doc:t})),..._!==void 0&&{signal:_}})}catch(e){if(_?.aborted??(e instanceof Error&&(e.name===`AbortError`||/abort/i.test(e.message))))throw e;let t=e instanceof Error?e.message:String(e);for(let e of n)y.addError(-1,`patch_attach_failed`,t,e.orphanWarning?.field,{target:e.id});continue}if(F+=r.succeeded,r.failures.length>0){let e=new Map(n.map(e=>[e.id,e]));for(let t of r.failures){let n=e.get(t.id)?.orphanWarning;t.status===404||t.type===`document_missing_exception`?(I+=1,y.addError(-1,n?.warningType??`orphan_patch`,n?.message??t.reason,n?.field,{target:t.id})):y.addError(-1,`patch_attach_failed`,t.reason,n?.field,{target:t.id})}}}let L=(Date.now()-S)/1e3;return x?.info({importRunId:i,rowsRead:C,rowsSucceeded:w,rowsFailed:T,rowsSkipped:E,batchesCompleted:O,patchesApplied:F,patchesOrphaned:I,totalSec:Math.round(L*10)/10,rowsPerSecond:L>0?Math.round(C/L):0},`imports.runner: finished`),{rowsRead:C,rowsSucceeded:w,rowsFailed:T,rowsSkipped:E,rowsWarned:D,batchesCompleted:O,errors:v.snapshot(),warnings:y.snapshot(),pairs:k,patchesApplied:F,patchesOrphaned:I,priceStats:b.snapshot()}}export{c as a,f as c,a as d,o as f,u as i,r as l,m as n,l as o,s as r,d as s,p as t,i as u};
-//# sourceMappingURL=runner-C9GIe5Dr.mjs.map
+import{n as e}from"./read-feed-DsLgRDNR.mjs";import{bulkPartialUpdate as t,bulkUpsert as n}from"@murumets-ee/search-elasticsearch";const r=50,i=5;var a=class{patterns=new Map;maxPatterns;maxSamplesPerPattern;droppedSignatures=0;constructor(e={}){this.maxPatterns=e.maxPatterns??50,this.maxSamplesPerPattern=e.maxSamplesPerPattern??5}addError(e,t,n,r,i){let a=`${t}:${r??`GENERAL`}:${n}`,o=this.patterns.get(a);if(!o){if(this.patterns.size>=this.maxPatterns){this.droppedSignatures+=1;return}o={errorType:t,field:r??null,message:n,count:0,firstOccurrence:e,lastOccurrence:e,samples:[]},this.patterns.set(a,o)}o.count+=1,o.lastOccurrence=e,o.samples.length<this.maxSamplesPerPattern&&o.samples.push({rowNumber:e,rowData:i})}getTotalErrorCount(){let e=0;for(let t of this.patterns.values())e+=t.count;return e}getDistinctPatternCount(){return this.patterns.size}getDroppedSignatureCount(){return this.droppedSignatures}getTopPatterns(){let e=Array.from(this.patterns.values()).sort((e,t)=>t.count-e.count),t=e.reduce((e,t)=>e+t.count,0);return e.map(e=>({errorType:e.errorType,field:e.field,message:e.message,count:e.count,firstOccurrence:e.firstOccurrence,lastOccurrence:e.lastOccurrence,samples:e.samples.slice(),percentage:t>0?e.count/t*100:0}))}snapshot(){return{totalErrors:this.getTotalErrorCount(),distinctPatterns:this.getDistinctPatternCount(),droppedSignatures:this.droppedSignatures,patterns:this.getTopPatterns()}}};function o(e,t){let n=e.totalErrors+t.totalErrors,r=e=>({...e,percentage:n>0?e.count/n*100:0});return{totalErrors:n,distinctPatterns:e.distinctPatterns+t.distinctPatterns,droppedSignatures:e.droppedSignatures+t.droppedSignatures,patterns:[...e.patterns.map(r),...t.patterns.map(r)]}}const s={count:0,nullRate:0,min:null,median:null,max:null},c=5e4;var l=class{count=0;nullCount=0;nonNullSeen=0;samples=[];record(e){if(this.count+=1,e===void 0){this.nullCount+=1;return}if(this.nonNullSeen+=1,this.samples.length<5e4){this.samples.push(e);return}let t=Math.floor(Math.random()*this.nonNullSeen);t<5e4&&(this.samples[t]=e)}snapshot(){if(this.count===0)return{...s};let e=this.nullCount/this.count;if(this.samples.length===0)return{count:this.count,nullRate:e,min:null,median:null,max:null};let t=[...this.samples].sort((e,t)=>e-t),n=t[0]??null,r=t[t.length-1]??null,i=Math.floor(t.length/2),a;if(t.length%2==0){let e=t[i-1],n=t[i];a=e!==void 0&&n!==void 0?(e+n)/2:null}else a=t[i]??null;return{count:this.count,nullRate:e,min:n,median:a,max:r}}};const u=10;function d(e,t){if(e.median===null||t.median===null||e.median<=0||t.median<=0)return null;let n=t.median/e.median;return n>=10||n<=1/10?{ratio:n,prevMedian:e.median,currMedian:t.median}:null}function f(e){if(typeof e!=`object`||!e)return null;let t=e;return typeof t.count!=`number`||typeof t.nullRate!=`number`?null:{count:t.count,nullRate:t.nullRate,min:typeof t.min==`number`?t.min:null,median:typeof t.median==`number`?t.median:null,max:typeof t.max==`number`?t.max:null}}const p=1e3;async function m(r){let{importRunId:i,runLabel:o,params:s,transform:c,feed:u,esClient:d,esIndex:f,batchSize:m=p,onProgress:h,rowLimit:g,signal:_,errorTracker:v=new a,warningTracker:y=new a,priceStats:b=new l,logger:x}=r;if(m<1)throw Error(`batchSize must be >= 1 (got ${m})`);let S=Date.now();x?.info({importRunId:i,runLabel:o,batchSize:m,esIndex:f},`imports.runner: started`);let C=0,w=0,T=0,E=0,D=0,O=0,k=[],A=[],j=[],M=async()=>{if(j.length===0)return;let e=j;j=[];let t=Date.now(),r;try{r=await n(d,{index:f,docs:e.map(({id:e,doc:t})=>({id:e,doc:t})),..._!==void 0&&{signal:_}})}catch(t){let n=_?.aborted??(t instanceof Error&&(t.name===`AbortError`||/abort/i.test(t.message)))?`aborted`:`bulk_request_failed`,r=t instanceof Error?t.message:String(t);for(let{rowNumber:t}of e)v.addError(t,n,r,void 0,null);throw T+=e.length,t}w+=r.succeeded,T+=r.failures.length,O+=1;let a=Date.now()-t,o=(Date.now()-S)/1e3;if(x?.info({importRunId:i,batch:O,flushMs:a,submitted:e.length,succeeded:r.succeeded,failed:r.failures.length,rowsRead:C,rowsPerSecond:o>0?Math.round(C/o):0},`imports.runner: batch flushed`),r.failures.length>0){let t=new Map(e.map(e=>[e.id,e.rowNumber]));for(let e of r.failures){let n=t.get(e.id)??-1;v.addError(n,e.type,e.reason,void 0,{id:e.id})}}let s=new Set(r.failures.map(e=>e.id));for(let t of e)if(!(!t.warnings||t.warnings.length===0)&&!s.has(t.id)){D+=1;for(let e of t.warnings)y.addError(t.rowNumber,e.warningType,e.message,e.field,t.rowData)}for(let t of e)!t.pairs||t.pairs.length===0||s.has(t.id)||k.push(...t.pairs);for(let t of e)s.has(t.id)||b.record(t.priceSample);if(h){let e=(Date.now()-S)/1e3;h({rowsRead:C,rowsSucceeded:w,rowsFailed:T,rowsSkipped:E,rowsWarned:D,batchesCompleted:O,elapsedSeconds:e,rowsPerSecond:e>0?C/e:0,distinctErrorPatterns:v.getDistinctPatternCount()})}},N=m,P=new Map;for await(let{rowNumber:t,row:n}of e(u)){if(_?.aborted||g!==void 0&&C>=g)break;if(C+=1,C===1&&x?.info({importRunId:i,rowNumber:t,sampleKeys:Object.keys(n)},`imports.runner: STREAM_FIRST_ROW`),C%N===0){let e=(Date.now()-S)/1e3;x?.info({importRunId:i,rowsRead:C,rowsSucceeded:w,rowsFailed:T,rowsSkipped:E,pending:j.length,elapsedSec:Math.round(e*10)/10,rowsPerSecond:e>0?Math.round(C/e):0,distinctFirstErrorTypes:P.size},`imports.runner: STREAM_TICK`)}let e={importRunId:i,params:s,runLabel:o,rowNumber:t},r;try{r=await c(n,e)}catch(e){let r=e instanceof Error?e.message:String(e);v.addError(t,`transform_threw`,r,void 0,n),P.has(`transform_threw`)||(P.set(`transform_threw`,{rowNumber:t,message:r}),x?.warn({importRunId:i,rowNumber:t,errorType:`transform_threw`,message:r},`imports.runner: FIRST_ERROR transform_threw`)),T+=1;continue}if(r.kind===`skip`){E+=1,r.patches&&r.patches.length>0&&A.push(...r.patches);continue}if(r.kind===`error`){v.addError(t,r.error.errorType,r.error.message,r.error.field,n),P.has(r.error.errorType)||(P.set(r.error.errorType,{rowNumber:t,message:r.error.message}),x?.warn({importRunId:i,rowNumber:t,errorType:r.error.errorType,field:r.error.field,message:r.error.message},`imports.runner: FIRST_ERROR ${r.error.errorType}`)),T+=1;continue}j.push({id:r.id,doc:r.doc,rowNumber:t,rowData:n,...r.warnings&&r.warnings.length>0&&{warnings:r.warnings},...r.pairs&&r.pairs.length>0&&{pairs:r.pairs},...r.priceSample!==void 0&&{priceSample:r.priceSample}}),j.length>=m&&await M()}await M();let F=0,I=0;for(let e=0;e<A.length;e+=m){if(_?.aborted)throw Error(`imports: run aborted during core-charge attach (phase 2)`);let n=A.slice(e,e+m),r;try{r=await t(d,{index:f,updates:n.map(({id:e,doc:t})=>({id:e,doc:t})),..._!==void 0&&{signal:_}})}catch(e){if(_?.aborted??(e instanceof Error&&(e.name===`AbortError`||/abort/i.test(e.message))))throw e;let t=e instanceof Error?e.message:String(e);for(let e of n)y.addError(-1,`patch_attach_failed`,t,e.orphanWarning?.field,{target:e.id});continue}if(F+=r.succeeded,r.failures.length>0){let e=new Map(n.map(e=>[e.id,e]));for(let t of r.failures){let n=e.get(t.id)?.orphanWarning;t.status===404||t.type===`document_missing_exception`?(I+=1,y.addError(-1,n?.warningType??`orphan_patch`,n?.message??t.reason,n?.field,{target:t.id})):y.addError(-1,`patch_attach_failed`,t.reason,n?.field,{target:t.id})}}}let L=(Date.now()-S)/1e3;return x?.info({importRunId:i,rowsRead:C,rowsSucceeded:w,rowsFailed:T,rowsSkipped:E,batchesCompleted:O,patchesApplied:F,patchesOrphaned:I,totalSec:Math.round(L*10)/10,rowsPerSecond:L>0?Math.round(C/L):0},`imports.runner: finished`),{rowsRead:C,rowsSucceeded:w,rowsFailed:T,rowsSkipped:E,rowsWarned:D,batchesCompleted:O,errors:v.snapshot(),warnings:y.snapshot(),pairs:k,patchesApplied:F,patchesOrphaned:I,priceStats:b.snapshot()}}export{c as a,f as c,a as d,o as f,u as i,r as l,m as n,l as o,s as r,d as s,p as t,i as u};
+//# sourceMappingURL=runner-BnFvID1z.mjs.map

package/dist/{runner-C9GIe5Dr.mjs.map → runner-BnFvID1z.mjs.map}

@@ -1 +1 @@
-{"version":3,"file":"runner-C9GIe5Dr.mjs","names":[],"sources":["../src/error-tracker.ts","../src/price-stats.ts","../src/runner.ts"],"sourcesContent":["/**\n * Aggregates per-row errors into top-N pattern buckets so a feed of 1M\n * malformed rows surfaces as a handful of actionable signatures rather\n * than a million identical strings.\n *\n * Lifted from giga-test (`backend/src/workers/csv-importer.ts`) and\n * generalized:\n * - Configurable caps so the importer can tune memory bounds per run.\n * - Pure data — no logging side effects, no I/O.\n * - JSON-serialisable output via {@link ErrorTracker.snapshot} for the\n *   `import_run.errorSummary` column.\n *\n * Pattern signature shape: `${errorType}:${field || 'GENERAL'}:${message}`.\n * Same `errorType + field + message` collapses to one bucket; differing\n * messages stay separate. This is intentional: a parser error on column\n * `NetPrice/Discount` (\"invalid number 'NA,5'\") and the same on column\n * `GrossPrice` are operationally distinct even if the parser is the same.\n *\n * Memory bounds: the patterns map is capped at `maxPatterns`. Once full,\n * additional NEW signatures are dropped — known patterns keep accumulating\n * counts. This is the \"top-N most common\" model: rare-but-novel errors\n * past the cap are invisible, but the cap protects against a runaway\n * adversarial feed exploding the map. Sample arrays are independently\n * capped at `maxSamplesPerPattern`.\n */\n\n/**\n * Recursively-defined JSON-serialisable value. Mirrors the shape of\n * `JsonValue` in `@murumets-ee/entity` without taking a dependency on\n * that package — error-tracker is otherwise standalone, and the\n * `ImportRun.errorSummary` JSONB column accepts anything in this shape.\n */\nexport type ImportJsonValue =\n  | string\n  | number\n  | boolean\n  | null\n  | ImportJsonValue[]\n  | { [key: string]: ImportJsonValue }\n\n/** One sample row attached to a pattern. `rowData` is the originally-parsed row. */\nexport interface ErrorSample {\n  rowNumber: number\n  rowData: ImportJsonValue\n}\n\n/** Public shape of an aggregated pattern as returned by {@link ErrorTracker.getTopPatterns}. */\nexport interface ErrorPattern {\n  errorType: string\n  field: string | null\n  message: string\n  count: number\n  firstOccurrence: number\n  lastOccurrence: number\n  samples: ReadonlyArray<ErrorSample>\n  /** `count / totalErrors`, scaled 0..100. `0` when there are zero errors total. */\n  percentage: number\n}\n\nexport interface ErrorTrackerConfig {\n  /** Hard cap on distinct signatures. Defaults to 50. New signatures past the cap are dropped. */\n  maxPatterns?: number\n  /** Hard cap on samples retained per pattern. Defaults to 5. Excess samples are dropped. */\n  maxSamplesPerPattern?: number\n}\n\n/** Default caps — match giga-test for compatibility. */\nexport const DEFAULT_MAX_PATTERNS = 50\nexport const DEFAULT_MAX_SAMPLES_PER_PATTERN = 5\n\ninterface InternalPattern {\n  errorType: string\n  field: string | null\n  message: string\n  count: number\n  firstOccurrence: number\n  lastOccurrence: number\n  samples: ErrorSample[]\n}\n\nexport class ErrorTracker {\n  private readonly patterns = new Map<string, InternalPattern>()\n  private readonly maxPatterns: number\n  private readonly maxSamplesPerPattern: number\n  private droppedSignatures = 0\n\n  constructor(config: ErrorTrackerConfig = {}) {\n    this.maxPatterns = config.maxPatterns ?? DEFAULT_MAX_PATTERNS\n    this.maxSamplesPerPattern = config.maxSamplesPerPattern ?? DEFAULT_MAX_SAMPLES_PER_PATTERN\n  }\n\n  /**\n   * Record one error. Same `(errorType, field, message)` triple bumps the\n   * existing bucket; a new triple opens a new one (subject to {@link maxPatterns}).\n   *\n   * `field` is optional — pass `undefined` for errors not tied to a single\n   * column (e.g. parse errors at row level). Internally normalised to the\n   * literal string `'GENERAL'` so it shares a bucket with other genericised\n   * errors of the same type+message.\n   */\n  addError(\n    rowNumber: number,\n    errorType: string,\n    message: string,\n    field: string | undefined,\n    rowData: ImportJsonValue,\n  ): void {\n    const fieldKey = field ?? 'GENERAL'\n    const signature = `${errorType}:${fieldKey}:${message}`\n\n    let pattern = this.patterns.get(signature)\n    if (!pattern) {\n      if (this.patterns.size >= this.maxPatterns) {\n        this.droppedSignatures += 1\n        return\n      }\n      pattern = {\n        errorType,\n        field: field ?? null,\n        message,\n        count: 0,\n        firstOccurrence: rowNumber,\n        lastOccurrence: rowNumber,\n        samples: [],\n      }\n      this.patterns.set(signature, pattern)\n    }\n\n    pattern.count += 1\n    pattern.lastOccurrence = rowNumber\n    if (pattern.samples.length < this.maxSamplesPerPattern) {\n      pattern.samples.push({ rowNumber, rowData })\n    }\n  }\n\n  /** Total count across every pattern. Counts errors, not patterns. */\n  getTotalErrorCount(): number {\n    let sum = 0\n    for (const p of this.patterns.values()) sum += p.count\n    return sum\n  }\n\n  /** Number of distinct signatures retained in the map (≤ `maxPatterns`). */\n  getDistinctPatternCount(): number {\n    return this.patterns.size\n  }\n\n  /**\n   * Number of NEW signatures dropped because the map was already at\n   * capacity. Surfacing this in the import_run summary tells the operator\n   * \"the top-N was saturated — there's a long tail you're not seeing\".\n   */\n  getDroppedSignatureCount(): number {\n    return this.droppedSignatures\n  }\n\n  /**\n   * Top patterns sorted by descending count, capped at `maxPatterns`.\n   * Stable secondary order is insertion order (Map iteration order is\n   * insertion order; Array.sort is stable in V8).\n   */\n  getTopPatterns(): ErrorPattern[] {\n    const all = Array.from(this.patterns.values()).sort((a, b) => b.count - a.count)\n    const totalErrors = all.reduce((sum, p) => sum + p.count, 0)\n    return all.map((p) => ({\n      errorType: p.errorType,\n      field: p.field,\n      message: p.message,\n      count: p.count,\n      firstOccurrence: p.firstOccurrence,\n      lastOccurrence: p.lastOccurrence,\n      samples: p.samples.slice(),\n      percentage: totalErrors > 0 ? (p.count / totalErrors) * 100 : 0,\n    }))\n  }\n\n  /**\n   * Compact JSON-serialisable snapshot for `import_run.errorSummary`.\n   * Aside from the patterns array, includes the totals so a reader of\n   * just this column doesn't have to re-derive them.\n   */\n  snapshot(): ErrorTrackerSnapshot {\n    return {\n      totalErrors: this.getTotalErrorCount(),\n      distinctPatterns: this.getDistinctPatternCount(),\n      droppedSignatures: this.droppedSignatures,\n      patterns: this.getTopPatterns(),\n    }\n  }\n}\n\nexport interface ErrorTrackerSnapshot {\n  totalErrors: number\n  distinctPatterns: number\n  droppedSignatures: number\n  patterns: ErrorPattern[]\n}\n\n/**\n * Merge two independently-produced `ErrorTrackerSnapshot`s into one — e.g.\n * the runner's per-row `warningSummary` plus a worker-level signal computed\n * AFTER the run completes (F010 / #602: the price-drift anomaly check runs\n * in `worker.ts`, after `runImport` returns, using a fresh single-pattern\n * tracker of its own). Patterns concatenate rather than re-dedup: the two\n * snapshots are assumed to use disjoint `errorType` signature spaces (a\n * worker-level tracker uses a type — `price_anomaly` — the runner's own\n * warning aggregation never emits), so counts sum directly with no risk of\n * double-counting one bucket as two.\n */\nexport function mergeErrorTrackerSnapshots(\n  a: ErrorTrackerSnapshot,\n  b: ErrorTrackerSnapshot,\n): ErrorTrackerSnapshot {\n  const totalErrors = a.totalErrors + b.totalErrors\n  // Each source snapshot computed `percentage` against its OWN `totalErrors`\n  // (see `getTopPatterns` above); the merged total is different, so every\n  // pattern's `percentage` must be recomputed against it — a straight\n  // concat would leave every value stale (e.g. `20/100=20%` staying `20`\n  // once the true merged denominator is `101`). `0` when the merged total\n  // is `0`, matching `getTopPatterns`' own zero-total convention.\n  const rescale = (p: ErrorPattern): ErrorPattern => ({\n    ...p,\n    percentage: totalErrors > 0 ? (p.count / totalErrors) * 100 : 0,\n  })\n  return {\n    totalErrors,\n    distinctPatterns: a.distinctPatterns + b.distinctPatterns,\n    droppedSignatures: a.droppedSignatures + b.droppedSignatures,\n    patterns: [...a.patterns.map(rescale), ...b.patterns.map(rescale)],\n  }\n}\n","/**\n * Price-distribution stats + prior-run anomaly detection (F010 / issue #602).\n *\n * The import pipeline's existing anomaly detection — register-handler.ts's\n * \"6b. Header-presence drift check\" — only catches STRUCTURAL drift: a\n * mapped column vanishing (or getting renamed) from the header row. It does\n * NOT catch VALUE drift: a supplier silently flipping their decimal\n * separator (e.g. `1.234,56` → `1234.56`) or thousands separator WITHOUT\n * touching a single column header. Since the saved\n * `import_mapping.numberFormat` (`{decimal, thousands}`) is unchanged,\n * every price in that run parses as a NUMERICALLY VALID but WRONG value —\n * silently corrupting the catalog's prices. Nothing upstream of this module\n * flags it.\n *\n * This module computes a per-run price-distribution snapshot (`PriceStats`)\n * over `PartsDocument.purchase_price_minor` — the DERIVED cost field,\n * rather than the raw net/list price columns, because both the direct-net\n * AND two-file derivation paths in `mapping/parts-schema.ts` (step 4) funnel\n * through it, so a decimal-format flip on EITHER source column surfaces\n * here regardless of which pricing path a supplier's feed uses.\n *\n * Collection (`PriceStatsCollector`) happens in the SAME streaming pass\n * `runner.ts` already uses to build `totals` — one `.record()` call per\n * ES-accepted row, no second pass over the file, no extra DB round-trip.\n * The prior-run comparison (`detectPriceAnomaly`) happens in `worker.ts`\n * AFTER the run completes, because it needs `AdminClient` to look the\n * prior run up — this module stays DB-agnostic, pure data + pure functions.\n */\n\n/** A run's price-distribution snapshot. Values are in MINOR units (cents), matching `PartsDocument.purchase_price_minor`'s own scale. */\nexport interface PriceStats {\n  /** ES-accepted rows considered — same accounting as `RunImportResult.rowsSucceeded`. */\n  count: number\n  /** Fraction (0..1) of `count` whose price sample was absent (e.g. a list-only row with no resolvable purchase cost — the same rows that carry a `no_purchase_price` warning). `0` when `count` is `0`. */\n  nullRate: number\n  /** Minimum of the non-null samples, in minor units. `null` when there are zero non-null samples. */\n  min: number | null\n  /** Median of the non-null samples, in minor units. `null` when there are zero non-null samples. */\n  median: number | null\n  /** Maximum of the non-null samples, in minor units. `null` when there are zero non-null samples. */\n  max: number | null\n}\n\n/** The all-zero/empty snapshot — a run that produced zero ES-accepted rows. */\nexport const EMPTY_PRICE_STATS: PriceStats = {\n  count: 0,\n  nullRate: 0,\n  min: null,\n  median: null,\n  max: null,\n}\n\n/**\n * Bounded sample cap for the min/median/max computation — the same\n * \"bounded, uniform sample\" philosophy `ErrorTracker` uses top-N patterns\n * for. 50,000 floats is ~400KB, trivial next to the ES bulk-write buffers\n * already in flight for a run this size, and far more than needed for a\n * stable median estimate. A feed with more ES-accepted rows than this cap\n * still gets an EXACT `count` / `nullRate` (those are running counters, not\n * sample-bounded) — only min/median/max become an approximation over a\n * UNIFORM RANDOM sample of `PRICE_SAMPLE_CAP` non-null values, not \"the\n * first N encountered\" (see `record`'s reservoir-sampling note for why\n * order-independence matters specifically for this module's job).\n */\nexport const PRICE_SAMPLE_CAP = 50_000\n\n/**\n * Accumulates price samples during a run's streaming pass. `record` is O(1)\n * (amortized array push below the cap, one RNG draw above it); `snapshot`\n * sorts the retained samples once at the end — cheap relative to the ES\n * bulk-write the run already performs every batch.\n */\nexport class PriceStatsCollector {\n  private count = 0\n  private nullCount = 0\n  private nonNullSeen = 0\n  private readonly samples: number[] = []\n\n  /**\n   * Record one ES-accepted row's price sample. Call exactly once per\n   * accepted row (mirrors the `rowsWarned` / `pairs` accounting in\n   * `runner.ts`'s flush loop) — `undefined` means \"this row had no price to\n   * sample\" and counts toward `nullRate` rather than being skipped.\n   *\n   * Reservoir sampling (Algorithm R) past the cap, not \"first N wins\": each\n   * new non-null value replaces a uniformly-random existing slot with\n   * probability `PRICE_SAMPLE_CAP / nonNullSeen`, giving every non-null\n   * sample seen so far an equal chance of surviving into the final\n   * reservoir. Order-independence matters here specifically — the anomaly\n   * this module exists to catch (a supplier's decimal-format flip) could\n   * start partway through a feed larger than the cap (e.g. a corrupted\n   * re-export appended to an otherwise-normal catalogue), and a first-N\n   * sample would never see a single row past the cutover point, silently\n   * missing exactly the drift F010 was built to detect.\n   */\n  record(value: number | undefined): void {\n    this.count += 1\n    if (value === undefined) {\n      this.nullCount += 1\n      return\n    }\n    this.nonNullSeen += 1\n    if (this.samples.length < PRICE_SAMPLE_CAP) {\n      this.samples.push(value)\n      return\n    }\n    const j = Math.floor(Math.random() * this.nonNullSeen)\n    if (j < PRICE_SAMPLE_CAP) this.samples[j] = value\n  }\n\n  snapshot(): PriceStats {\n    if (this.count === 0) return { ...EMPTY_PRICE_STATS }\n    const nullRate = this.nullCount / this.count\n    if (this.samples.length === 0) {\n      return { count: this.count, nullRate, min: null, median: null, max: null }\n    }\n    const sorted = [...this.samples].sort((a, b) => a - b)\n    const min = sorted[0] ?? null\n    const max = sorted[sorted.length - 1] ?? null\n    const mid = Math.floor(sorted.length / 2)\n    let median: number | null\n    if (sorted.length % 2 === 0) {\n      const lo = sorted[mid - 1]\n      const hi = sorted[mid]\n      median = lo !== undefined && hi !== undefined ? (lo + hi) / 2 : null\n    } else {\n      median = sorted[mid] ?? null\n    }\n    return { count: this.count, nullRate, min, median, max }\n  }\n}\n\n/**\n * Anomaly threshold: a run whose median purchase price is >= 10x higher OR\n * <= 1/10 (0.1x) of the prior successful run's median (for the same\n * transform + supplier + brand) is flagged.\n *\n * Chosen generously above the range a real repricing plausibly covers\n * (even an aggressive markup/discount renegotiation is a low-single-digit\n * multiplier — 2x-3x would already be an extreme outlier for a legitimate\n * price change) but comfortably BELOW the ~100x-1000x magnitude a\n * decimal/thousands-separator flip produces for typical EUR part prices:\n * `\"1.234,56\"` (a European-format €1234.56) misparsed with dot-decimal\n * conventions reads as `1.234,56` → `1234.56`, a ~1000x inflation if the\n * true value was e.g. €1.23; a thousands-separator flip on `\"1234,56\"`\n * (true €1234.56) misparsed as `123456` (dropping the decimal entirely) is\n * a ~100x inflation. 10x leaves a wide safety margin against false\n * positives on genuine price changes while still comfortably catching\n * every realistic decimal/thousands-separator flip.\n */\nexport const PRICE_ANOMALY_THRESHOLD_MULTIPLIER = 10\n\nexport interface PriceAnomaly {\n  /** `curr.median / prev.median`. */\n  ratio: number\n  /** The prior run's median (minor units). */\n  prevMedian: number\n  /** This run's median (minor units). */\n  currMedian: number\n}\n\n/**\n * Compare two runs' price stats for a magnitude-scale anomaly. Returns\n * `null` when there's nothing to compare (either side has no median — zero\n * non-null price samples, e.g. a schema/transform that never emits a price\n * sample, or a run with zero ES-accepted rows) or the ratio falls within\n * the safe band.\n */\nexport function detectPriceAnomaly(prev: PriceStats, curr: PriceStats): PriceAnomaly | null {\n  if (prev.median === null || curr.median === null) return null\n  // A zero/negative median can't produce a meaningful ratio. Shouldn't occur\n  // for real minor-unit prices (`toMinorUnits` on a positive price); this is\n  // defensive-only so a pathological value can't produce Infinity/NaN below.\n  if (prev.median <= 0 || curr.median <= 0) return null\n  const ratio = curr.median / prev.median\n  if (\n    ratio >= PRICE_ANOMALY_THRESHOLD_MULTIPLIER ||\n    ratio <= 1 / PRICE_ANOMALY_THRESHOLD_MULTIPLIER\n  ) {\n    return { ratio, prevMedian: prev.median, currMedian: curr.median }\n  }\n  return null\n}\n\n/**\n * Narrow an `unknown` value (a `field.json()` column read back through\n * `AdminClient`, typed `JsonValue | undefined` on the DTO) to `PriceStats`,\n * or `null` when it doesn't structurally match — e.g. a run written before\n * this field existed, or a run whose transform never emitted a price\n * sample so `min`/`median`/`max` serialised as `null`. Defensive, matching\n * the `params` narrowing pattern already used in `worker.ts`.\n */\nexport function parsePriceStats(value: unknown): PriceStats | null {\n  if (typeof value !== 'object' || value === null) return null\n  const v = value as Record<string, unknown>\n  if (typeof v.count !== 'number' || typeof v.nullRate !== 'number') return null\n  return {\n    count: v.count,\n    nullRate: v.nullRate,\n    min: typeof v.min === 'number' ? v.min : null,\n    median: typeof v.median === 'number' ? v.median : null,\n    max: typeof v.max === 'number' ? v.max : null,\n  }\n}\n","/**\n * One-shot importer: stream rows → transform → batched bulk-write to ES,\n * accumulating per-row errors into `ErrorTracker` and reporting progress\n * to the queue every batch.\n *\n * Per PLAN-ECOMMERCE.md PR 7 (PoC scope):\n * - **Batch size 1000.** Matches giga-test precedent. Configurable for\n *   integration tests that don't want a 1k row floor.\n * - **No resumability and no automatic retries.** The `imports:run` queue\n *   job is registered with `defaultRetries: 0` (see `worker.ts`) so a\n *   failed handler does NOT re-enqueue itself — re-running a multi-batch\n *   import against the same `import_run.id` while the previous attempt\n *   may still be writing is a footgun (duplicate batches, double-counted\n *   progress). Operator retries by creating a NEW `import_run` row.\n * - **No per-supplier transform plugin.** The runner takes a single\n *   `RowTransform<TDoc>` from the registry and applies it to every row;\n *   PR 8 may diverge but only by registering a different transform name.\n * - **Direct `bulkUpsert` into the live aliased index** (D6 alias is\n *   set up by PR 4's `ensureAliasedIndex`; the importer doesn't reindex).\n *\n * Per D21 (sanctioned bulk path): this runner intentionally bypasses\n * AdminClient and entity hooks. Per-batch audit / observability lives\n * on the surrounding `import_run` row + queue progress, NOT per-row.\n */\n\nimport {\n  type BulkIndexResult,\n  bulkPartialUpdate,\n  bulkUpsert,\n  type EsClientLike,\n} from '@murumets-ee/search-elasticsearch'\nimport { ErrorTracker } from './error-tracker.js'\nimport { type PriceStats, PriceStatsCollector } from './price-stats.js'\nimport { type ReadFeedOptions, readFeed } from './read-feed.js'\nimport type {\n  DirectionalPair,\n  DocPatch,\n  RowTransform,\n  RowWarning,\n  TransformContext,\n} from './transform.js'\n\n/** Soft default; chosen to match giga-test. ES bulk requests over ~5MB get split server-side anyway. */\nexport const DEFAULT_BATCH_SIZE = 1000\n\nexport interface RunImportOptions<TDoc> {\n  /** UUID of the `import_run` row driving this run. Forwarded to every transform invocation. */\n  importRunId: string\n  /** Operator-supplied label for the run. Forwarded to the transform context. */\n  runLabel: string\n  /** Opaque per-run params copied from `import_run.params`. */\n  params: Record<string, unknown>\n  /** Transform applied to every parsed row. */\n  transform: RowTransform<TDoc>\n  /** Feed reader options — file path + format (csv/tsv/xlsx) + delimiter/header config. */\n  feed: ReadFeedOptions\n  /** ES client (low-level shape from `@murumets-ee/search-elasticsearch`). */\n  esClient: EsClientLike\n  /** Index alias to write to. Per D6, callers always pass an alias, never a physical index. */\n  esIndex: string\n  /** Rows per `bulkUpsert` call. Default {@link DEFAULT_BATCH_SIZE}. */\n  batchSize?: number\n  /**\n   * Callback invoked after every batch. The handler in `worker.ts`\n   * forwards this to `ctx.updateProgress` for the queue UI; tests\n   * inspect it directly. Synchronous + cheap so a slow callback can't\n   * back-pressure the importer.\n   */\n  onProgress?: (progress: ImportRunProgress) => void\n  /** Optional: stop processing after this many rows. Tests use it; production passes `undefined`. */\n  rowLimit?: number\n  /** Abort signal threaded into the underlying ES client request — cooperative cancel. */\n  signal?: AbortSignal\n  /** Optional ErrorTracker config (caps). Default: top-50 patterns × 5 samples. */\n  errorTracker?: ErrorTracker\n  /**\n   * Optional separate tracker for NON-FATAL row warnings (commerce-demo PR4 —\n   * `no_purchase_price`). Distinct from {@link errorTracker} so warnings never\n   * inflate `totals.failed` / the error summary: a warned row is a SUCCESS whose\n   * doc is indexed. Default: a fresh tracker with the same caps.\n   */\n  warningTracker?: ErrorTracker\n  /**\n   * Optional price-stats collector (F010 / #602) — accumulates a\n   * min/median/max/nullRate distribution over ES-accepted rows' optional\n   * `priceSample`s in the SAME streaming pass that builds `totals`. Default:\n   * a fresh collector. Tests can inject their own to inspect it directly —\n   * mirrors the {@link errorTracker} / {@link warningTracker} DI shape.\n   */\n  priceStats?: PriceStatsCollector\n  /**\n   * Optional structural logger for per-batch telemetry. Receives\n   * `{ batch, ms, succeeded, failed, rowsRead, rowsPerSecond }` after\n   * every flush, plus a one-shot `started` and `finished` event. The\n   * worker forwards `app.logger.child({ pkg: 'imports' })`; tests\n   * leave it undefined.\n   *\n   * Why: without per-batch visibility, a stalled run looks identical\n   * to \"first batch in progress\" forever — the only feedback is the\n   * queue's `progress` jsonb (which writes after each batch) and the\n   * eventual `succeeded`/`failed` job event. Adding this lets the\n   * operator tail the worker log and immediately see whether ES is\n   * rate-limiting, the transform is rejecting every row, or the\n   * stream is actually progressing.\n   */\n  logger?: {\n    info: (data: Record<string, unknown>, msg: string) => void\n    warn: (data: Record<string, unknown>, msg: string) => void\n  }\n}\n\n/**\n * Progress payload written to `toolkit_jobs.progress` after every batch.\n * Caps + flush rules live on the queue's `updateProgress` debounce —\n * callers don't need to throttle.\n */\nexport interface ImportRunProgress {\n  rowsRead: number\n  rowsSucceeded: number\n  rowsFailed: number\n  rowsSkipped: number\n  /** Rows indexed WITH a non-fatal warning (commerce-demo PR4 — `no_purchase_price`). */\n  rowsWarned: number\n  batchesCompleted: number\n  /** Wall-clock seconds since the runner started. */\n  elapsedSeconds: number\n  /** Rows / second, computed at every batch. */\n  rowsPerSecond: number\n  /** Distinct error patterns currently held by the tracker. Saturates at the cap. */\n  distinctErrorPatterns: number\n  /**\n   * Total data-row count from a pre-pass scan, when the worker computed one.\n   * Lets the UI render `processed / total` percent. Absent when the worker\n   * skipped the count (file too large for the count budget, count failed,\n   * etc.) — UI must treat it as optional.\n   */\n  totalRows?: number\n}\n\n/**\n * Final result returned by {@link runImport}. The handler writes these\n * onto the `import_run` row alongside the ErrorTracker snapshot.\n */\nexport interface RunImportResult {\n  /** Total rows read from the file (excludes skipped empty lines). */\n  rowsRead: number\n  /** Rows the transform turned into a successful doc AND the ES cluster acknowledged. */\n  rowsSucceeded: number\n  /**\n   * Rows that the transform rejected (`{ kind: 'error' }`) OR that ES\n   * rejected on bulk-write (per-doc failure). Both are aggregated by\n   * `errorTracker` for the import_run summary.\n   */\n  rowsFailed: number\n  /** Rows that the transform skipped (`{ kind: 'skip' }`) — header noise, blank lines, intentional drop. */\n  rowsSkipped: number\n  /**\n   * Rows that succeeded (doc indexed) but carried ≥1 non-fatal warning —\n   * counted SEPARATELY from `rowsSucceeded`/`rowsFailed` (a warned row is also\n   * counted in `rowsSucceeded`). commerce-demo PR4: `no_purchase_price`.\n   */\n  rowsWarned: number\n  /** Number of `bulkUpsert` calls made. */\n  batchesCompleted: number\n  /** Final value of {@link ErrorTracker.snapshot}. */\n  errors: ReturnType<ErrorTracker['snapshot']>\n  /** Aggregated NON-FATAL warnings (commerce-demo PR4). Empty when no row warned. */\n  warnings: ReturnType<ErrorTracker['snapshot']>\n  /**\n   * Directional supersession pairs harvested as a side-output (commerce-demo PR6,\n   * SD007). Buffered in memory from ES-ACCEPTED rows only (same accounting as\n   * warnings — a row whose doc was rejected by ES contributes no pair). Empty unless\n   * a schema emits them (the `parts` schema, when a row's `superseding_code` is set).\n   * The worker hands these to an injected `pairSink` AFTER the run for the commerce\n   * replacement-merge engine; the runner itself never writes the consumer's tables.\n   */\n  pairs: readonly DirectionalPair[]\n  /**\n   * Doc-patches applied by the phase-2 pass (commerce-demo PR7): a `<code>*S`\n   * core-fee row is not a saleable doc — it emits a {@link DocPatch} that merges its\n   * deposit onto the PARENT part's `core_charge_minor`. The runner buffers these\n   * during the stream (sparse) and applies them via `bulkPartialUpdate` AFTER the ES\n   * pass. `patchesApplied` = targets the cluster merged; `patchesOrphaned` = patches\n   * whose parent genuinely wasn't in the index (a `document_missing`/404 → an\n   * `orphan_core_charge` warning). An INFRA failure (whole-batch transport error, or a\n   * non-404 per-doc failure) is NOT counted as orphaned — it lands in a distinct\n   * `patch_attach_failed` warning bucket so it can't masquerade as a data orphan. Zero\n   * unless a schema emits patches.\n   */\n  patchesApplied: number\n  patchesOrphaned: number\n  /**\n   * Price-distribution snapshot (F010 / #602) accumulated from ES-accepted\n   * rows' `priceSample`s during this same streaming pass. See `price-stats.ts`.\n   * `count` is `0` (all fields `null`/`0`) for a run whose schema never emits\n   * a price sample.\n   */\n  priceStats: PriceStats\n}\n\n/**\n * Apply the runner against a feed file. Stops on rowLimit OR end-of-file\n * OR if `signal` aborts. Throws if the streaming reader / ES client\n * throws — caller (the queue handler) catches that and writes\n * `import_run.status = 'failed'` with the error message in\n * `errorSummary.fatal`.\n */\nexport async function runImport<TDoc>(options: RunImportOptions<TDoc>): Promise<RunImportResult> {\n  const {\n    importRunId,\n    runLabel,\n    params,\n    transform,\n    feed,\n    esClient,\n    esIndex,\n    batchSize = DEFAULT_BATCH_SIZE,\n    onProgress,\n    rowLimit,\n    signal,\n    errorTracker = new ErrorTracker(),\n    warningTracker = new ErrorTracker(),\n    priceStats: priceStatsCollector = new PriceStatsCollector(),\n    logger,\n  } = options\n\n  if (batchSize < 1) {\n    throw new Error(`batchSize must be >= 1 (got ${batchSize})`)\n  }\n\n  const startedAt = Date.now()\n  logger?.info({ importRunId, runLabel, batchSize, esIndex }, 'imports.runner: started')\n  let rowsRead = 0\n  let rowsSucceeded = 0\n  let rowsFailed = 0\n  let rowsSkipped = 0\n  let rowsWarned = 0\n  let batchesCompleted = 0\n\n  // Side-output buffer (commerce-demo PR6, SD007): directional supersession pairs\n  // accumulated from ES-ACCEPTED rows across the whole run. An array append per\n  // sparse superseded row — it never blocks the `bulkUpsert` loop. Handed to the\n  // injected `pairSink` by the worker AFTER the run (read-once / merge-after).\n  const collectedPairs: DirectionalPair[] = []\n\n  // Side-output buffer (commerce-demo PR7): doc-patches from `*S` core-fee rows —\n  // each attaches a core charge onto a DIFFERENT (parent) doc. Collected off the\n  // `skip` result (a `*S` row is never a saleable doc), sparse, applied in a phase-2\n  // `bulkPartialUpdate` after the ES pass (below) — order-independent because ES\n  // `update` is realtime by `_id`.\n  const collectedPatches: DocPatch<TDoc>[] = []\n\n  // Warnings + pairs ride on the pending item (with the raw row for the warning\n  // sample) so they're accounted at FLUSH time — only for rows ES actually accepted.\n  // A warned row is by contract a SUCCESSFUL row, so counting it at push time (before\n  // the bulk response) would wrongly keep a later-ES-rejected row in rowsWarned /\n  // warnings; pairs follow the same accepted-rows-only rule.\n  let pending: Array<{\n    id: string\n    doc: TDoc\n    rowNumber: number\n    rowData: Record<string, string>\n    warnings?: readonly RowWarning[]\n    pairs?: readonly DirectionalPair[]\n    priceSample?: number\n  }> = []\n\n  const flush = async (): Promise<void> => {\n    if (pending.length === 0) return\n    const batch = pending\n    pending = []\n    const flushStartedAt = Date.now()\n    let result: BulkIndexResult\n    try {\n      result = await bulkUpsert<TDoc>(esClient, {\n        index: esIndex,\n        docs: batch.map(({ id, doc }) => ({ id, doc })),\n        ...(signal !== undefined && { signal }),\n      })\n    } catch (err) {\n      // Cluster- or transport-level failure — the whole batch is\n      // unaccounted for. Distinguish abort (operator-driven cancel) from\n      // a real cluster failure so the errorSummary doesn't mislabel a\n      // cancelled run as broken cluster connectivity.\n      const isAbort =\n        signal?.aborted ??\n        (err instanceof Error && (err.name === 'AbortError' || /abort/i.test(err.message)))\n      const errorType = isAbort ? 'aborted' : 'bulk_request_failed'\n      const reason = err instanceof Error ? err.message : String(err)\n      for (const { rowNumber } of batch) {\n        errorTracker.addError(rowNumber, errorType, reason, undefined, null)\n      }\n      rowsFailed += batch.length\n      throw err\n    }\n\n    rowsSucceeded += result.succeeded\n    rowsFailed += result.failures.length\n    batchesCompleted += 1\n\n    const flushMs = Date.now() - flushStartedAt\n    const elapsedSec = (Date.now() - startedAt) / 1000\n    logger?.info(\n      {\n        importRunId,\n        batch: batchesCompleted,\n        flushMs,\n        submitted: batch.length,\n        succeeded: result.succeeded,\n        failed: result.failures.length,\n        rowsRead,\n        rowsPerSecond: elapsedSec > 0 ? Math.round(rowsRead / elapsedSec) : 0,\n      },\n      'imports.runner: batch flushed',\n    )\n\n    if (result.failures.length > 0) {\n      // Map each ES failure back to its source row via `id`. The bulk\n      // response order matches the request order, but ES doesn't promise\n      // that; matching by `id` is the safe path. PoC volume is small\n      // enough that the O(failures × batch) cost is irrelevant.\n      const byId = new Map(batch.map((b) => [b.id, b.rowNumber]))\n      for (const fail of result.failures) {\n        const rowNumber = byId.get(fail.id) ?? -1\n        errorTracker.addError(rowNumber, fail.type, fail.reason, undefined, { id: fail.id })\n      }\n    }\n\n    // Account non-fatal warnings ONLY for rows ES accepted (commerce-demo PR4) — a\n    // warned row is by contract a SUCCESSFUL row, so a row that warned AND then failed\n    // the bulk write must NOT land in rowsWarned/warnings (it's already in rowsFailed).\n    const failedIds = new Set(result.failures.map((f) => f.id))\n    for (const item of batch) {\n      if (!item.warnings || item.warnings.length === 0) continue\n      if (failedIds.has(item.id)) continue\n      rowsWarned += 1\n      for (const w of item.warnings) {\n        warningTracker.addError(item.rowNumber, w.warningType, w.message, w.field, item.rowData)\n      }\n    }\n\n    // Buffer directional pairs for rows ES accepted (commerce-demo PR6, SD007) — same\n    // accepted-rows-only rule as warnings, so a pair from a doc ES rejected is dropped.\n    for (const item of batch) {\n      if (!item.pairs || item.pairs.length === 0) continue\n      if (failedIds.has(item.id)) continue\n      collectedPairs.push(...item.pairs)\n    }\n\n    // Price-distribution sampling (F010 / #602) — same accepted-rows-only rule as\n    // warnings/pairs: a row ES rejected contributes no sample. Every ES-accepted row\n    // records exactly once (`undefined` counts toward `nullRate` rather than being\n    // skipped), so `priceStats.count` tracks `rowsSucceeded` exactly.\n    for (const item of batch) {\n      if (failedIds.has(item.id)) continue\n      priceStatsCollector.record(item.priceSample)\n    }\n\n    if (onProgress) {\n      const elapsedSeconds = (Date.now() - startedAt) / 1000\n      onProgress({\n        rowsRead,\n        rowsSucceeded,\n        rowsFailed,\n        rowsSkipped,\n        rowsWarned,\n        batchesCompleted,\n        elapsedSeconds,\n        rowsPerSecond: elapsedSeconds > 0 ? rowsRead / elapsedSeconds : 0,\n        distinctErrorPatterns: errorTracker.getDistinctPatternCount(),\n      })\n    }\n  }\n\n  // Streaming heartbeat — sample every Nth row so the operator sees the\n  // streamer is alive even when a transform-rejected file would prevent\n  // any flush from firing. Without this, a feed that's 100% rejected\n  // looks identical to \"stuck\" until end-of-file. Sample frequency is\n  // batchSize so the cadence matches the natural batch rhythm.\n  const streamSampleEvery = batchSize\n\n  // First-error capture per error type — surface what's actually rejecting\n  // the rows when 100% of a feed is rejected. Without this, the only signal\n  // is `errorTracker.snapshot()` which only writes to import_run.errorSummary\n  // at the END of the run. Mid-run we want to know NOW.\n  const firstErrorByType = new Map<string, { rowNumber: number; message: string }>()\n\n  for await (const { rowNumber, row } of readFeed(feed)) {\n    if (signal?.aborted) break\n    if (rowLimit !== undefined && rowsRead >= rowLimit) break\n    rowsRead += 1\n\n    if (rowsRead === 1) {\n      logger?.info(\n        { importRunId, rowNumber, sampleKeys: Object.keys(row) },\n        'imports.runner: STREAM_FIRST_ROW',\n      )\n    }\n    if (rowsRead % streamSampleEvery === 0) {\n      const elapsedSec = (Date.now() - startedAt) / 1000\n      // Telemetry only — no row content, no error sample text.\n      // PR #262 review (M-5): the prior `firstErrors:\n      // Object.fromEntries(firstErrorByType)` payload included the\n      // raw error message captured from the transform, which for the\n      // carmaker feed embeds supplier codes + price fragments. The\n      // detailed first-error breakdown is already persisted on\n      // `import_run.errorSummary` via `errorTracker.snapshot()` —\n      // operators inspect it there, not in the worker log.\n      logger?.info(\n        {\n          importRunId,\n          rowsRead,\n          rowsSucceeded,\n          rowsFailed,\n          rowsSkipped,\n          pending: pending.length,\n          elapsedSec: Math.round(elapsedSec * 10) / 10,\n          rowsPerSecond: elapsedSec > 0 ? Math.round(rowsRead / elapsedSec) : 0,\n          distinctFirstErrorTypes: firstErrorByType.size,\n        },\n        'imports.runner: STREAM_TICK',\n      )\n    }\n\n    const ctx: TransformContext = { importRunId, params, runLabel, rowNumber }\n    let result: Awaited<ReturnType<typeof transform>>\n    try {\n      result = await transform(row, ctx)\n    } catch (err) {\n      // A throw from the transform is a programmer error — surface it as\n      // a row-level error so the run can continue. (If the bug is\n      // catastrophic, the operator sees the same message repeated and\n      // can stop the run.)\n      const reason = err instanceof Error ? err.message : String(err)\n      errorTracker.addError(rowNumber, 'transform_threw', reason, undefined, row)\n      if (!firstErrorByType.has('transform_threw')) {\n        firstErrorByType.set('transform_threw', { rowNumber, message: reason })\n        // PR #262 review (M-5): drop the `row` payload from the\n        // log line — supplier feeds carry pricing + internal codes.\n        // The full row is on `import_run.errorSummary.samples` for\n        // post-mortem review by operators with view permission.\n        logger?.warn(\n          { importRunId, rowNumber, errorType: 'transform_threw', message: reason },\n          'imports.runner: FIRST_ERROR transform_threw',\n        )\n      }\n      rowsFailed += 1\n      continue\n    }\n\n    if (result.kind === 'skip') {\n      rowsSkipped += 1\n      // A `*S` core-fee row skips the saleable index but may carry a doc-patch\n      // (commerce-demo PR7) — buffer it for the phase-2 attach. Unlike pairs, a\n      // patch has no ES write of its own to gate on, so it's collected here\n      // unconditionally; its success is decided by the phase-2 update.\n      if (result.patches && result.patches.length > 0) {\n        collectedPatches.push(...result.patches)\n      }\n      continue\n    }\n    if (result.kind === 'error') {\n      errorTracker.addError(\n        rowNumber,\n        result.error.errorType,\n        result.error.message,\n        result.error.field,\n        row,\n      )\n      if (!firstErrorByType.has(result.error.errorType)) {\n        firstErrorByType.set(result.error.errorType, {\n          rowNumber,\n          message: result.error.message,\n        })\n        logger?.warn(\n          {\n            importRunId,\n            rowNumber,\n            errorType: result.error.errorType,\n            field: result.error.field,\n            message: result.error.message,\n          },\n          `imports.runner: FIRST_ERROR ${result.error.errorType}`,\n        )\n      }\n      rowsFailed += 1\n      continue\n    }\n\n    // Non-fatal warnings ride on a SUCCESS — the doc is still indexed\n    // (commerce-demo PR4: `no_purchase_price`). Carry them (+ the raw row for the\n    // sample) on the pending item; the flush aggregates them into the separate\n    // `warningTracker` AFTER the bulk response confirms the row was accepted, so a\n    // warned-then-ES-rejected row never inflates rowsWarned / the warning summary.\n    pending.push({\n      id: result.id,\n      doc: result.doc,\n      rowNumber,\n      rowData: row,\n      ...(result.warnings && result.warnings.length > 0 && { warnings: result.warnings }),\n      ...(result.pairs && result.pairs.length > 0 && { pairs: result.pairs }),\n      ...(result.priceSample !== undefined && { priceSample: result.priceSample }),\n    })\n    if (pending.length >= batchSize) {\n      await flush()\n    }\n  }\n\n  await flush()\n\n  // Phase 2 (commerce-demo PR7): attach the buffered `*S` core-charge patches onto\n  // their PARENT docs. Runs AFTER the whole stream so every parent that exists in\n  // this run is already indexed; ES `update` is realtime by `_id`, so the merge\n  // lands regardless of whether the parent streamed before or after its `*S` row\n  // (order-independent). Patches are sparse (only reman parts carry `*S`), but batch\n  // anyway for a giant catalogue. A patch whose parent isn't in the index comes back\n  // as a per-doc failure (`document_missing_exception`) → an `orphan_core_charge`\n  // warning, never a phantom doc, never a silent drop (D004). A cluster failure of\n  // the whole batch is logged as a warning too — the saleable docs already indexed;\n  // only the deferred core-charge attach didn't happen — so we never fail the run\n  // over it.\n  let patchesApplied = 0\n  let patchesOrphaned = 0\n  for (let i = 0; i < collectedPatches.length; i += batchSize) {\n    // Operator-driven cancel BETWEEN batches: THROW rather than break, so a run\n    // cancelled after phase 1 is not marked `succeeded` with core charges left\n    // unattached. Mirrors phase-1's flush, which also re-throws on abort. The queue\n    // handler catches this and marks the run failed.\n    if (signal?.aborted) {\n      throw new Error('imports: run aborted during core-charge attach (phase 2)')\n    }\n    const slice = collectedPatches.slice(i, i + batchSize)\n    let updateResult: BulkIndexResult\n    try {\n      updateResult = await bulkPartialUpdate<TDoc>(esClient, {\n        index: esIndex,\n        updates: slice.map(({ id, doc }) => ({ id, doc })),\n        ...(signal !== undefined && { signal }),\n      })\n    } catch (err) {\n      // An abort surfacing as an AbortError from the ES client is a cancel, NOT an\n      // attach failure — re-throw so the run fails (same as phase 1), never a silent\n      // \"succeeded with patches unapplied\".\n      const isAbort =\n        signal?.aborted ??\n        (err instanceof Error && (err.name === 'AbortError' || /abort/i.test(err.message)))\n      if (isAbort) throw err\n      // Whole-batch transport/cluster failure (ES down, timeout). The parents may\n      // well exist — the attach simply didn't happen — so this is an INFRA failure,\n      // NOT a data orphan: it gets the distinct `patch_attach_failed` bucket and is\n      // NOT counted in patchesOrphaned. Best-effort: the saleable docs already\n      // indexed, and a re-import re-attaches, so we warn + continue rather than fail.\n      const reason = err instanceof Error ? err.message : String(err)\n      for (const patch of slice) {\n        warningTracker.addError(-1, 'patch_attach_failed', reason, patch.orphanWarning?.field, {\n          target: patch.id,\n        })\n      }\n      continue\n    }\n    patchesApplied += updateResult.succeeded\n    if (updateResult.failures.length > 0) {\n      const byId = new Map(slice.map((p) => [p.id, p]))\n      for (const fail of updateResult.failures) {\n        const w = byId.get(fail.id)?.orphanWarning\n        // A missing target (`document_missing_exception` / 404) is a genuine data\n        // orphan — the referenced parent isn't in this import (D004: warn, never a\n        // phantom doc). A stable message (from the patch's orphanWarning) aggregates\n        // orphans; the missing parent id rides in the sample. Any OTHER per-doc\n        // failure (mapper error, etc.) is an infra/attach failure kept in a distinct\n        // bucket so it can't masquerade as a data orphan.\n        const isOrphan = fail.status === 404 || fail.type === 'document_missing_exception'\n        if (isOrphan) {\n          patchesOrphaned += 1\n          warningTracker.addError(\n            -1,\n            w?.warningType ?? 'orphan_patch',\n            w?.message ?? fail.reason,\n            w?.field,\n            {\n              target: fail.id,\n            },\n          )\n        } else {\n          warningTracker.addError(-1, 'patch_attach_failed', fail.reason, w?.field, {\n            target: fail.id,\n          })\n        }\n      }\n    }\n  }\n\n  const totalSec = (Date.now() - startedAt) / 1000\n  logger?.info(\n    {\n      importRunId,\n      rowsRead,\n      rowsSucceeded,\n      rowsFailed,\n      rowsSkipped,\n      batchesCompleted,\n      patchesApplied,\n      patchesOrphaned,\n      totalSec: Math.round(totalSec * 10) / 10,\n      rowsPerSecond: totalSec > 0 ? Math.round(rowsRead / totalSec) : 0,\n    },\n    'imports.runner: finished',\n  )\n\n  return {\n    rowsRead,\n    rowsSucceeded,\n    rowsFailed,\n    rowsSkipped,\n    rowsWarned,\n    batchesCompleted,\n    errors: errorTracker.snapshot(),\n    warnings: warningTracker.snapshot(),\n    pairs: collectedPairs,\n    patchesApplied,\n    patchesOrphaned,\n    priceStats: priceStatsCollector.snapshot(),\n  }\n}\n"],"mappings":"mIAmEA,MAAa,EAAuB,GACvB,EAAkC,EAY/C,IAAa,EAAb,KAA0B,CACxB,SAA4B,IAAI,IAChC,YACA,qBACA,kBAA4B,EAE5B,YAAY,EAA6B,CAAC,EAAG,CAC3C,KAAK,YAAc,EAAO,aAAA,GAC1B,KAAK,qBAAuB,EAAO,sBAAA,CACrC,CAWA,SACE,EACA,EACA,EACA,EACA,EACM,CAEN,IAAM,EAAY,GAAG,EAAU,GADd,GAAS,UACiB,GAAG,IAE1C,EAAU,KAAK,SAAS,IAAI,CAAS,EACzC,GAAI,CAAC,EAAS,CACZ,GAAI,KAAK,SAAS,MAAQ,KAAK,YAAa,CAC1C,KAAK,mBAAqB,EAC1B,MACF,CACA,EAAU,CACR,YACA,MAAO,GAAS,KAChB,UACA,MAAO,EACP,gBAAiB,EACjB,eAAgB,EAChB,QAAS,CAAC,CACZ,EACA,KAAK,SAAS,IAAI,EAAW,CAAO,CACtC,CAEA,EAAQ,OAAS,EACjB,EAAQ,eAAiB,EACrB,EAAQ,QAAQ,OAAS,KAAK,sBAChC,EAAQ,QAAQ,KAAK,CAAE,YAAW,SAAQ,CAAC,CAE/C,CAGA,oBAA6B,CAC3B,IAAI,EAAM,EACV,IAAK,IAAM,KAAK,KAAK,SAAS,OAAO,EAAG,GAAO,EAAE,MACjD,OAAO,CACT,CAGA,yBAAkC,CAChC,OAAO,KAAK,SAAS,IACvB,CAOA,0BAAmC,CACjC,OAAO,KAAK,iBACd,CAOA,gBAAiC,CAC/B,IAAM,EAAM,MAAM,KAAK,KAAK,SAAS,OAAO,CAAC,CAAC,CAAC,MAAM,EAAG,IAAM,EAAE,MAAQ,EAAE,KAAK,EACzE,EAAc,EAAI,QAAQ,EAAK,IAAM,EAAM,EAAE,MAAO,CAAC,EAC3D,OAAO,EAAI,IAAK,IAAO,CACrB,UAAW,EAAE,UACb,MAAO,EAAE,MACT,QAAS,EAAE,QACX,MAAO,EAAE,MACT,gBAAiB,EAAE,gBACnB,eAAgB,EAAE,eAClB,QAAS,EAAE,QAAQ,MAAM,EACzB,WAAY,EAAc,EAAK,EAAE,MAAQ,EAAe,IAAM,CAChE,EAAE,CACJ,CAOA,UAAiC,CAC/B,MAAO,CACL,YAAa,KAAK,mBAAmB,EACrC,iBAAkB,KAAK,wBAAwB,EAC/C,kBAAmB,KAAK,kBACxB,SAAU,KAAK,eAAe,CAChC,CACF,CACF,EAoBA,SAAgB,EACd,EACA,EACsB,CACtB,IAAM,EAAc,EAAE,YAAc,EAAE,YAOhC,EAAW,IAAmC,CAClD,GAAG,EACH,WAAY,EAAc,EAAK,EAAE,MAAQ,EAAe,IAAM,CAChE,GACA,MAAO,CACL,cACA,iBAAkB,EAAE,iBAAmB,EAAE,iBACzC,kBAAmB,EAAE,kBAAoB,EAAE,kBAC3C,SAAU,CAAC,GAAG,EAAE,SAAS,IAAI,CAAO,EAAG,GAAG,EAAE,SAAS,IAAI,CAAO,CAAC,CACnE,CACF,CC1LA,MAAa,EAAgC,CAC3C,MAAO,EACP,SAAU,EACV,IAAK,KACL,OAAQ,KACR,IAAK,IACP,EAca,EAAmB,IAQhC,IAAa,EAAb,KAAiC,CAC/B,MAAgB,EAChB,UAAoB,EACpB,YAAsB,EACtB,QAAqC,CAAC,EAmBtC,OAAO,EAAiC,CAEtC,GADA,KAAK,OAAS,EACV,IAAU,IAAA,GAAW,CACvB,KAAK,WAAa,EAClB,MACF,CAEA,GADA,KAAK,aAAe,EAChB,KAAK,QAAQ,OAAA,IAA2B,CAC1C,KAAK,QAAQ,KAAK,CAAK,EACvB,MACF,CACA,IAAM,EAAI,KAAK,MAAM,KAAK,OAAO,EAAI,KAAK,WAAW,EACjD,EAAA,MAAsB,KAAK,QAAQ,GAAK,EAC9C,CAEA,UAAuB,CACrB,GAAI,KAAK,QAAU,EAAG,MAAO,CAAE,GAAG,CAAkB,EACpD,IAAM,EAAW,KAAK,UAAY,KAAK,MACvC,GAAI,KAAK,QAAQ,SAAW,EAC1B,MAAO,CAAE,MAAO,KAAK,MAAO,WAAU,IAAK,KAAM,OAAQ,KAAM,IAAK,IAAK,EAE3E,IAAM,EAAS,CAAC,GAAG,KAAK,OAAO,CAAC,CAAC,MAAM,EAAG,IAAM,EAAI,CAAC,EAC/C,EAAM,EAAO,IAAM,KACnB,EAAM,EAAO,EAAO,OAAS,IAAM,KACnC,EAAM,KAAK,MAAM,EAAO,OAAS,CAAC,EACpC,EACJ,GAAI,EAAO,OAAS,GAAM,EAAG,CAC3B,IAAM,EAAK,EAAO,EAAM,GAClB,EAAK,EAAO,GAClB,EAAS,IAAO,IAAA,IAAa,IAAO,IAAA,IAAa,EAAK,GAAM,EAAI,IAClE,KACE,GAAS,EAAO,IAAQ,KAE1B,MAAO,CAAE,MAAO,KAAK,MAAO,WAAU,MAAK,SAAQ,KAAI,CACzD,CACF,EAoBA,MAAa,EAAqC,GAkBlD,SAAgB,EAAmB,EAAkB,EAAuC,CAK1F,GAJI,EAAK,SAAW,MAAQ,EAAK,SAAW,MAIxC,EAAK,QAAU,GAAK,EAAK,QAAU,EAAG,OAAO,KACjD,IAAM,EAAQ,EAAK,OAAS,EAAK,OAOjC,OALE,GAAA,IACA,GAAS,EAAA,GAEF,CAAE,QAAO,WAAY,EAAK,OAAQ,WAAY,EAAK,MAAO,EAE5D,IACT,CAUA,SAAgB,EAAgB,EAAmC,CACjE,GAAI,OAAO,GAAU,WAAY,EAAgB,OAAO,KACxD,IAAM,EAAI,EAEV,OADI,OAAO,EAAE,OAAU,UAAY,OAAO,EAAE,UAAa,SAAiB,KACnE,CACL,MAAO,EAAE,MACT,SAAU,EAAE,SACZ,IAAK,OAAO,EAAE,KAAQ,SAAW,EAAE,IAAM,KACzC,OAAQ,OAAO,EAAE,QAAW,SAAW,EAAE,OAAS,KAClD,IAAK,OAAO,EAAE,KAAQ,SAAW,EAAE,IAAM,IAC3C,CACF,CChKA,MAAa,EAAqB,IAoKlC,eAAsB,EAAgB,EAA2D,CAC/F,GAAM,CACJ,cACA,WACA,SACA,YACA,OACA,WACA,UACA,YAAY,EACZ,aACA,WACA,SACA,eAAe,IAAI,EACnB,iBAAiB,IAAI,EACrB,WAAY,EAAsB,IAAI,EACtC,UACE,EAEJ,GAAI,EAAY,EACd,MAAU,MAAM,+BAA+B,EAAU,EAAE,EAG7D,IAAM,EAAY,KAAK,IAAI,EAC3B,GAAQ,KAAK,CAAE,cAAa,WAAU,YAAW,SAAQ,EAAG,yBAAyB,EACrF,IAAI,EAAW,EACX,EAAgB,EAChB,EAAa,EACb,EAAc,EACd,EAAa,EACb,EAAmB,EAMjB,EAAoC,CAAC,EAOrC,EAAqC,CAAC,EAOxC,EAQC,CAAC,EAEA,EAAQ,SAA2B,CACvC,GAAI,EAAQ,SAAW,EAAG,OAC1B,IAAM,EAAQ,EACd,EAAU,CAAC,EACX,IAAM,EAAiB,KAAK,IAAI,EAC5B,EACJ,GAAI,CACF,EAAS,MAAM,EAAiB,EAAU,CACxC,MAAO,EACP,KAAM,EAAM,KAAK,CAAE,KAAI,UAAW,CAAE,KAAI,KAAI,EAAE,EAC9C,GAAI,IAAW,IAAA,IAAa,CAAE,QAAO,CACvC,CAAC,CACH,OAAS,EAAK,CAQZ,IAAM,EAFJ,GAAQ,UACP,aAAe,QAAU,EAAI,OAAS,cAAgB,SAAS,KAAK,EAAI,OAAO,IACtD,UAAY,sBAClC,EAAS,aAAe,MAAQ,EAAI,QAAU,OAAO,CAAG,EAC9D,IAAK,GAAM,CAAE,eAAe,EAC1B,EAAa,SAAS,EAAW,EAAW,EAAQ,IAAA,GAAW,IAAI,EAGrE,KADA,IAAc,EAAM,OACd,CACR,CAEA,GAAiB,EAAO,UACxB,GAAc,EAAO,SAAS,OAC9B,GAAoB,EAEpB,IAAM,EAAU,KAAK,IAAI,EAAI,EACvB,GAAc,KAAK,IAAI,EAAI,GAAa,IAe9C,GAdA,GAAQ,KACN,CACE,cACA,MAAO,EACP,UACA,UAAW,EAAM,OACjB,UAAW,EAAO,UAClB,OAAQ,EAAO,SAAS,OACxB,WACA,cAAe,EAAa,EAAI,KAAK,MAAM,EAAW,CAAU,EAAI,CACtE,EACA,+BACF,EAEI,EAAO,SAAS,OAAS,EAAG,CAK9B,IAAM,EAAO,IAAI,IAAI,EAAM,IAAK,GAAM,CAAC,EAAE,GAAI,EAAE,SAAS,CAAC,CAAC,EAC1D,IAAK,IAAM,KAAQ,EAAO,SAAU,CAClC,IAAM,EAAY,EAAK,IAAI,EAAK,EAAE,GAAK,GACvC,EAAa,SAAS,EAAW,EAAK,KAAM,EAAK,OAAQ,IAAA,GAAW,CAAE,GAAI,EAAK,EAAG,CAAC,CACrF,CACF,CAKA,IAAM,EAAY,IAAI,IAAI,EAAO,SAAS,IAAK,GAAM,EAAE,EAAE,CAAC,EAC1D,IAAK,IAAM,KAAQ,EACb,MAAC,EAAK,UAAY,EAAK,SAAS,SAAW,IAC3C,GAAU,IAAI,EAAK,EAAE,EACzB,IAAc,EACd,IAAK,IAAM,KAAK,EAAK,SACnB,EAAe,SAAS,EAAK,UAAW,EAAE,YAAa,EAAE,QAAS,EAAE,MAAO,EAAK,OAAO,CAF3E,CAQhB,IAAK,IAAM,KAAQ,EACb,CAAC,EAAK,OAAS,EAAK,MAAM,SAAW,GACrC,EAAU,IAAI,EAAK,EAAE,GACzB,EAAe,KAAK,GAAG,EAAK,KAAK,EAOnC,IAAK,IAAM,KAAQ,EACb,EAAU,IAAI,EAAK,EAAE,GACzB,EAAoB,OAAO,EAAK,WAAW,EAG7C,GAAI,EAAY,CACd,IAAM,GAAkB,KAAK,IAAI,EAAI,GAAa,IAClD,EAAW,CACT,WACA,gBACA,aACA,cACA,aACA,mBACA,iBACA,cAAe,EAAiB,EAAI,EAAW,EAAiB,EAChE,sBAAuB,EAAa,wBAAwB,CAC9D,CAAC,CACH,CACF,EAOM,EAAoB,EAMpB,EAAmB,IAAI,IAE7B,UAAW,GAAM,CAAE,YAAW,SAAS,EAAS,CAAI,EAAG,CAErD,GADI,GAAQ,SACR,IAAa,IAAA,IAAa,GAAY,EAAU,MASpD,GARA,GAAY,EAER,IAAa,GACf,GAAQ,KACN,CAAE,cAAa,YAAW,WAAY,OAAO,KAAK,CAAG,CAAE,EACvD,kCACF,EAEE,EAAW,IAAsB,EAAG,CACtC,IAAM,GAAc,KAAK,IAAI,EAAI,GAAa,IAS9C,GAAQ,KACN,CACE,cACA,WACA,gBACA,aACA,cACA,QAAS,EAAQ,OACjB,WAAY,KAAK,MAAM,EAAa,EAAE,EAAI,GAC1C,cAAe,EAAa,EAAI,KAAK,MAAM,EAAW,CAAU,EAAI,EACpE,wBAAyB,EAAiB,IAC5C,EACA,6BACF,CACF,CAEA,IAAM,EAAwB,CAAE,cAAa,SAAQ,WAAU,WAAU,EACrE,EACJ,GAAI,CACF,EAAS,MAAM,EAAU,EAAK,CAAG,CACnC,OAAS,EAAK,CAKZ,IAAM,EAAS,aAAe,MAAQ,EAAI,QAAU,OAAO,CAAG,EAC9D,EAAa,SAAS,EAAW,kBAAmB,EAAQ,IAAA,GAAW,CAAG,EACrE,EAAiB,IAAI,iBAAiB,IACzC,EAAiB,IAAI,kBAAmB,CAAE,YAAW,QAAS,CAAO,CAAC,EAKtE,GAAQ,KACN,CAAE,cAAa,YAAW,UAAW,kBAAmB,QAAS,CAAO,EACxE,6CACF,GAEF,GAAc,EACd,QACF,CAEA,GAAI,EAAO,OAAS,OAAQ,CAC1B,GAAe,EAKX,EAAO,SAAW,EAAO,QAAQ,OAAS,GAC5C,EAAiB,KAAK,GAAG,EAAO,OAAO,EAEzC,QACF,CACA,GAAI,EAAO,OAAS,QAAS,CAC3B,EAAa,SACX,EACA,EAAO,MAAM,UACb,EAAO,MAAM,QACb,EAAO,MAAM,MACb,CACF,EACK,EAAiB,IAAI,EAAO,MAAM,SAAS,IAC9C,EAAiB,IAAI,EAAO,MAAM,UAAW,CAC3C,YACA,QAAS,EAAO,MAAM,OACxB,CAAC,EACD,GAAQ,KACN,CACE,cACA,YACA,UAAW,EAAO,MAAM,UACxB,MAAO,EAAO,MAAM,MACpB,QAAS,EAAO,MAAM,OACxB,EACA,+BAA+B,EAAO,MAAM,WAC9C,GAEF,GAAc,EACd,QACF,CAOA,EAAQ,KAAK,CACX,GAAI,EAAO,GACX,IAAK,EAAO,IACZ,YACA,QAAS,EACT,GAAI,EAAO,UAAY,EAAO,SAAS,OAAS,GAAK,CAAE,SAAU,EAAO,QAAS,EACjF,GAAI,EAAO,OAAS,EAAO,MAAM,OAAS,GAAK,CAAE,MAAO,EAAO,KAAM,EACrE,GAAI,EAAO,cAAgB,IAAA,IAAa,CAAE,YAAa,EAAO,WAAY,CAC5E,CAAC,EACG,EAAQ,QAAU,GACpB,MAAM,EAAM,CAEhB,CAEA,MAAM,EAAM,EAaZ,IAAI,EAAiB,EACjB,EAAkB,EACtB,IAAK,IAAI,EAAI,EAAG,EAAI,EAAiB,OAAQ,GAAK,EAAW,CAK3D,GAAI,GAAQ,QACV,MAAU,MAAM,0DAA0D,EAE5E,IAAM,EAAQ,EAAiB,MAAM,EAAG,EAAI,CAAS,EACjD,EACJ,GAAI,CACF,EAAe,MAAM,EAAwB,EAAU,CACrD,MAAO,EACP,QAAS,EAAM,KAAK,CAAE,KAAI,UAAW,CAAE,KAAI,KAAI,EAAE,EACjD,GAAI,IAAW,IAAA,IAAa,CAAE,QAAO,CACvC,CAAC,CACH,OAAS,EAAK,CAOZ,GAFE,GAAQ,UACP,aAAe,QAAU,EAAI,OAAS,cAAgB,SAAS,KAAK,EAAI,OAAO,IACrE,MAAM,EAMnB,IAAM,EAAS,aAAe,MAAQ,EAAI,QAAU,OAAO,CAAG,EAC9D,IAAK,IAAM,KAAS,EAClB,EAAe,SAAS,GAAI,sBAAuB,EAAQ,EAAM,eAAe,MAAO,CACrF,OAAQ,EAAM,EAChB,CAAC,EAEH,QACF,CAEA,GADA,GAAkB,EAAa,UAC3B,EAAa,SAAS,OAAS,EAAG,CACpC,IAAM,EAAO,IAAI,IAAI,EAAM,IAAK,GAAM,CAAC,EAAE,GAAI,CAAC,CAAC,CAAC,EAChD,IAAK,IAAM,KAAQ,EAAa,SAAU,CACxC,IAAM,EAAI,EAAK,IAAI,EAAK,EAAE,CAAC,EAAE,cAOZ,EAAK,SAAW,KAAO,EAAK,OAAS,8BAEpD,GAAmB,EACnB,EAAe,SACb,GACA,GAAG,aAAe,eAClB,GAAG,SAAW,EAAK,OACnB,GAAG,MACH,CACE,OAAQ,EAAK,EACf,CACF,GAEA,EAAe,SAAS,GAAI,sBAAuB,EAAK,OAAQ,GAAG,MAAO,CACxE,OAAQ,EAAK,EACf,CAAC,CAEL,CACF,CACF,CAEA,IAAM,GAAY,KAAK,IAAI,EAAI,GAAa,IAiB5C,OAhBA,GAAQ,KACN,CACE,cACA,WACA,gBACA,aACA,cACA,mBACA,iBACA,kBACA,SAAU,KAAK,MAAM,EAAW,EAAE,EAAI,GACtC,cAAe,EAAW,EAAI,KAAK,MAAM,EAAW,CAAQ,EAAI,CAClE,EACA,0BACF,EAEO,CACL,WACA,gBACA,aACA,cACA,aACA,mBACA,OAAQ,EAAa,SAAS,EAC9B,SAAU,EAAe,SAAS,EAClC,MAAO,EACP,iBACA,kBACA,WAAY,EAAoB,SAAS,CAC3C,CACF"}
+{"version":3,"file":"runner-BnFvID1z.mjs","names":[],"sources":["../src/error-tracker.ts","../src/price-stats.ts","../src/runner.ts"],"sourcesContent":["/**\n * Aggregates per-row errors into top-N pattern buckets so a feed of 1M\n * malformed rows surfaces as a handful of actionable signatures rather\n * than a million identical strings.\n *\n * Lifted from giga-test (`backend/src/workers/csv-importer.ts`) and\n * generalized:\n * - Configurable caps so the importer can tune memory bounds per run.\n * - Pure data — no logging side effects, no I/O.\n * - JSON-serialisable output via {@link ErrorTracker.snapshot} for the\n *   `import_run.errorSummary` column.\n *\n * Pattern signature shape: `${errorType}:${field || 'GENERAL'}:${message}`.\n * Same `errorType + field + message` collapses to one bucket; differing\n * messages stay separate. This is intentional: a parser error on column\n * `NetPrice/Discount` (\"invalid number 'NA,5'\") and the same on column\n * `GrossPrice` are operationally distinct even if the parser is the same.\n *\n * Memory bounds: the patterns map is capped at `maxPatterns`. Once full,\n * additional NEW signatures are dropped — known patterns keep accumulating\n * counts. This is the \"top-N most common\" model: rare-but-novel errors\n * past the cap are invisible, but the cap protects against a runaway\n * adversarial feed exploding the map. Sample arrays are independently\n * capped at `maxSamplesPerPattern`.\n */\n\n/**\n * Recursively-defined JSON-serialisable value. Mirrors the shape of\n * `JsonValue` in `@murumets-ee/entity` without taking a dependency on\n * that package — error-tracker is otherwise standalone, and the\n * `ImportRun.errorSummary` JSONB column accepts anything in this shape.\n */\nexport type ImportJsonValue =\n  | string\n  | number\n  | boolean\n  | null\n  | ImportJsonValue[]\n  | { [key: string]: ImportJsonValue }\n\n/** One sample row attached to a pattern. `rowData` is the originally-parsed row. */\nexport interface ErrorSample {\n  rowNumber: number\n  rowData: ImportJsonValue\n}\n\n/** Public shape of an aggregated pattern as returned by {@link ErrorTracker.getTopPatterns}. */\nexport interface ErrorPattern {\n  errorType: string\n  field: string | null\n  message: string\n  count: number\n  firstOccurrence: number\n  lastOccurrence: number\n  samples: ReadonlyArray<ErrorSample>\n  /** `count / totalErrors`, scaled 0..100. `0` when there are zero errors total. */\n  percentage: number\n}\n\nexport interface ErrorTrackerConfig {\n  /** Hard cap on distinct signatures. Defaults to 50. New signatures past the cap are dropped. */\n  maxPatterns?: number\n  /** Hard cap on samples retained per pattern. Defaults to 5. Excess samples are dropped. */\n  maxSamplesPerPattern?: number\n}\n\n/** Default caps — match giga-test for compatibility. */\nexport const DEFAULT_MAX_PATTERNS = 50\nexport const DEFAULT_MAX_SAMPLES_PER_PATTERN = 5\n\ninterface InternalPattern {\n  errorType: string\n  field: string | null\n  message: string\n  count: number\n  firstOccurrence: number\n  lastOccurrence: number\n  samples: ErrorSample[]\n}\n\nexport class ErrorTracker {\n  private readonly patterns = new Map<string, InternalPattern>()\n  private readonly maxPatterns: number\n  private readonly maxSamplesPerPattern: number\n  private droppedSignatures = 0\n\n  constructor(config: ErrorTrackerConfig = {}) {\n    this.maxPatterns = config.maxPatterns ?? DEFAULT_MAX_PATTERNS\n    this.maxSamplesPerPattern = config.maxSamplesPerPattern ?? DEFAULT_MAX_SAMPLES_PER_PATTERN\n  }\n\n  /**\n   * Record one error. Same `(errorType, field, message)` triple bumps the\n   * existing bucket; a new triple opens a new one (subject to {@link maxPatterns}).\n   *\n   * `field` is optional — pass `undefined` for errors not tied to a single\n   * column (e.g. parse errors at row level). Internally normalised to the\n   * literal string `'GENERAL'` so it shares a bucket with other genericised\n   * errors of the same type+message.\n   */\n  addError(\n    rowNumber: number,\n    errorType: string,\n    message: string,\n    field: string | undefined,\n    rowData: ImportJsonValue,\n  ): void {\n    const fieldKey = field ?? 'GENERAL'\n    const signature = `${errorType}:${fieldKey}:${message}`\n\n    let pattern = this.patterns.get(signature)\n    if (!pattern) {\n      if (this.patterns.size >= this.maxPatterns) {\n        this.droppedSignatures += 1\n        return\n      }\n      pattern = {\n        errorType,\n        field: field ?? null,\n        message,\n        count: 0,\n        firstOccurrence: rowNumber,\n        lastOccurrence: rowNumber,\n        samples: [],\n      }\n      this.patterns.set(signature, pattern)\n    }\n\n    pattern.count += 1\n    pattern.lastOccurrence = rowNumber\n    if (pattern.samples.length < this.maxSamplesPerPattern) {\n      pattern.samples.push({ rowNumber, rowData })\n    }\n  }\n\n  /** Total count across every pattern. Counts errors, not patterns. */\n  getTotalErrorCount(): number {\n    let sum = 0\n    for (const p of this.patterns.values()) sum += p.count\n    return sum\n  }\n\n  /** Number of distinct signatures retained in the map (≤ `maxPatterns`). */\n  getDistinctPatternCount(): number {\n    return this.patterns.size\n  }\n\n  /**\n   * Number of NEW signatures dropped because the map was already at\n   * capacity. Surfacing this in the import_run summary tells the operator\n   * \"the top-N was saturated — there's a long tail you're not seeing\".\n   */\n  getDroppedSignatureCount(): number {\n    return this.droppedSignatures\n  }\n\n  /**\n   * Top patterns sorted by descending count, capped at `maxPatterns`.\n   * Stable secondary order is insertion order (Map iteration order is\n   * insertion order; Array.sort is stable in V8).\n   */\n  getTopPatterns(): ErrorPattern[] {\n    const all = Array.from(this.patterns.values()).sort((a, b) => b.count - a.count)\n    const totalErrors = all.reduce((sum, p) => sum + p.count, 0)\n    return all.map((p) => ({\n      errorType: p.errorType,\n      field: p.field,\n      message: p.message,\n      count: p.count,\n      firstOccurrence: p.firstOccurrence,\n      lastOccurrence: p.lastOccurrence,\n      samples: p.samples.slice(),\n      percentage: totalErrors > 0 ? (p.count / totalErrors) * 100 : 0,\n    }))\n  }\n\n  /**\n   * Compact JSON-serialisable snapshot for `import_run.errorSummary`.\n   * Aside from the patterns array, includes the totals so a reader of\n   * just this column doesn't have to re-derive them.\n   */\n  snapshot(): ErrorTrackerSnapshot {\n    return {\n      totalErrors: this.getTotalErrorCount(),\n      distinctPatterns: this.getDistinctPatternCount(),\n      droppedSignatures: this.droppedSignatures,\n      patterns: this.getTopPatterns(),\n    }\n  }\n}\n\nexport interface ErrorTrackerSnapshot {\n  totalErrors: number\n  distinctPatterns: number\n  droppedSignatures: number\n  patterns: ErrorPattern[]\n}\n\n/**\n * Merge two independently-produced `ErrorTrackerSnapshot`s into one — e.g.\n * the runner's per-row `warningSummary` plus a worker-level signal computed\n * AFTER the run completes (F010 / #602: the price-drift anomaly check runs\n * in `worker.ts`, after `runImport` returns, using a fresh single-pattern\n * tracker of its own). Patterns concatenate rather than re-dedup: the two\n * snapshots are assumed to use disjoint `errorType` signature spaces (a\n * worker-level tracker uses a type — `price_anomaly` — the runner's own\n * warning aggregation never emits), so counts sum directly with no risk of\n * double-counting one bucket as two.\n */\nexport function mergeErrorTrackerSnapshots(\n  a: ErrorTrackerSnapshot,\n  b: ErrorTrackerSnapshot,\n): ErrorTrackerSnapshot {\n  const totalErrors = a.totalErrors + b.totalErrors\n  // Each source snapshot computed `percentage` against its OWN `totalErrors`\n  // (see `getTopPatterns` above); the merged total is different, so every\n  // pattern's `percentage` must be recomputed against it — a straight\n  // concat would leave every value stale (e.g. `20/100=20%` staying `20`\n  // once the true merged denominator is `101`). `0` when the merged total\n  // is `0`, matching `getTopPatterns`' own zero-total convention.\n  const rescale = (p: ErrorPattern): ErrorPattern => ({\n    ...p,\n    percentage: totalErrors > 0 ? (p.count / totalErrors) * 100 : 0,\n  })\n  return {\n    totalErrors,\n    distinctPatterns: a.distinctPatterns + b.distinctPatterns,\n    droppedSignatures: a.droppedSignatures + b.droppedSignatures,\n    patterns: [...a.patterns.map(rescale), ...b.patterns.map(rescale)],\n  }\n}\n","/**\n * Price-distribution stats + prior-run anomaly detection (F010 / issue #602).\n *\n * The import pipeline's existing anomaly detection — register-handler.ts's\n * \"6b. Header-presence drift check\" — only catches STRUCTURAL drift: a\n * mapped column vanishing (or getting renamed) from the header row. It does\n * NOT catch VALUE drift: a supplier silently flipping their decimal\n * separator (e.g. `1.234,56` → `1234.56`) or thousands separator WITHOUT\n * touching a single column header. Since the saved\n * `import_mapping.numberFormat` (`{decimal, thousands}`) is unchanged,\n * every price in that run parses as a NUMERICALLY VALID but WRONG value —\n * silently corrupting the catalog's prices. Nothing upstream of this module\n * flags it.\n *\n * This module computes a per-run price-distribution snapshot (`PriceStats`)\n * over `PartsDocument.purchase_price_minor` — the DERIVED cost field,\n * rather than the raw net/list price columns, because both the direct-net\n * AND two-file derivation paths in `mapping/parts-schema.ts` (step 4) funnel\n * through it, so a decimal-format flip on EITHER source column surfaces\n * here regardless of which pricing path a supplier's feed uses.\n *\n * Collection (`PriceStatsCollector`) happens in the SAME streaming pass\n * `runner.ts` already uses to build `totals` — one `.record()` call per\n * ES-accepted row, no second pass over the file, no extra DB round-trip.\n * The prior-run comparison (`detectPriceAnomaly`) happens in `worker.ts`\n * AFTER the run completes, because it needs `AdminClient` to look the\n * prior run up — this module stays DB-agnostic, pure data + pure functions.\n */\n\n/** A run's price-distribution snapshot. Values are in MINOR units (cents), matching `PartsDocument.purchase_price_minor`'s own scale. */\nexport interface PriceStats {\n  /** ES-accepted rows considered — same accounting as `RunImportResult.rowsSucceeded`. */\n  count: number\n  /** Fraction (0..1) of `count` whose price sample was absent (e.g. a list-only row with no resolvable purchase cost — the same rows that carry a `no_purchase_price` warning). `0` when `count` is `0`. */\n  nullRate: number\n  /** Minimum of the non-null samples, in minor units. `null` when there are zero non-null samples. */\n  min: number | null\n  /** Median of the non-null samples, in minor units. `null` when there are zero non-null samples. */\n  median: number | null\n  /** Maximum of the non-null samples, in minor units. `null` when there are zero non-null samples. */\n  max: number | null\n}\n\n/** The all-zero/empty snapshot — a run that produced zero ES-accepted rows. */\nexport const EMPTY_PRICE_STATS: PriceStats = {\n  count: 0,\n  nullRate: 0,\n  min: null,\n  median: null,\n  max: null,\n}\n\n/**\n * Bounded sample cap for the min/median/max computation — the same\n * \"bounded, uniform sample\" philosophy `ErrorTracker` uses top-N patterns\n * for. 50,000 floats is ~400KB, trivial next to the ES bulk-write buffers\n * already in flight for a run this size, and far more than needed for a\n * stable median estimate. A feed with more ES-accepted rows than this cap\n * still gets an EXACT `count` / `nullRate` (those are running counters, not\n * sample-bounded) — only min/median/max become an approximation over a\n * UNIFORM RANDOM sample of `PRICE_SAMPLE_CAP` non-null values, not \"the\n * first N encountered\" (see `record`'s reservoir-sampling note for why\n * order-independence matters specifically for this module's job).\n */\nexport const PRICE_SAMPLE_CAP = 50_000\n\n/**\n * Accumulates price samples during a run's streaming pass. `record` is O(1)\n * (amortized array push below the cap, one RNG draw above it); `snapshot`\n * sorts the retained samples once at the end — cheap relative to the ES\n * bulk-write the run already performs every batch.\n */\nexport class PriceStatsCollector {\n  private count = 0\n  private nullCount = 0\n  private nonNullSeen = 0\n  private readonly samples: number[] = []\n\n  /**\n   * Record one ES-accepted row's price sample. Call exactly once per\n   * accepted row (mirrors the `rowsWarned` / `pairs` accounting in\n   * `runner.ts`'s flush loop) — `undefined` means \"this row had no price to\n   * sample\" and counts toward `nullRate` rather than being skipped.\n   *\n   * Reservoir sampling (Algorithm R) past the cap, not \"first N wins\": each\n   * new non-null value replaces a uniformly-random existing slot with\n   * probability `PRICE_SAMPLE_CAP / nonNullSeen`, giving every non-null\n   * sample seen so far an equal chance of surviving into the final\n   * reservoir. Order-independence matters here specifically — the anomaly\n   * this module exists to catch (a supplier's decimal-format flip) could\n   * start partway through a feed larger than the cap (e.g. a corrupted\n   * re-export appended to an otherwise-normal catalogue), and a first-N\n   * sample would never see a single row past the cutover point, silently\n   * missing exactly the drift F010 was built to detect.\n   */\n  record(value: number | undefined): void {\n    this.count += 1\n    if (value === undefined) {\n      this.nullCount += 1\n      return\n    }\n    this.nonNullSeen += 1\n    if (this.samples.length < PRICE_SAMPLE_CAP) {\n      this.samples.push(value)\n      return\n    }\n    const j = Math.floor(Math.random() * this.nonNullSeen)\n    if (j < PRICE_SAMPLE_CAP) this.samples[j] = value\n  }\n\n  snapshot(): PriceStats {\n    if (this.count === 0) return { ...EMPTY_PRICE_STATS }\n    const nullRate = this.nullCount / this.count\n    if (this.samples.length === 0) {\n      return { count: this.count, nullRate, min: null, median: null, max: null }\n    }\n    const sorted = [...this.samples].sort((a, b) => a - b)\n    const min = sorted[0] ?? null\n    const max = sorted[sorted.length - 1] ?? null\n    const mid = Math.floor(sorted.length / 2)\n    let median: number | null\n    if (sorted.length % 2 === 0) {\n      const lo = sorted[mid - 1]\n      const hi = sorted[mid]\n      median = lo !== undefined && hi !== undefined ? (lo + hi) / 2 : null\n    } else {\n      median = sorted[mid] ?? null\n    }\n    return { count: this.count, nullRate, min, median, max }\n  }\n}\n\n/**\n * Anomaly threshold: a run whose median purchase price is >= 10x higher OR\n * <= 1/10 (0.1x) of the prior successful run's median (for the same\n * transform + supplier + brand) is flagged.\n *\n * Chosen generously above the range a real repricing plausibly covers\n * (even an aggressive markup/discount renegotiation is a low-single-digit\n * multiplier — 2x-3x would already be an extreme outlier for a legitimate\n * price change) but comfortably BELOW the ~100x-1000x magnitude a\n * decimal/thousands-separator flip produces for typical EUR part prices:\n * `\"1.234,56\"` (a European-format €1234.56) misparsed with dot-decimal\n * conventions reads as `1.234,56` → `1234.56`, a ~1000x inflation if the\n * true value was e.g. €1.23; a thousands-separator flip on `\"1234,56\"`\n * (true €1234.56) misparsed as `123456` (dropping the decimal entirely) is\n * a ~100x inflation. 10x leaves a wide safety margin against false\n * positives on genuine price changes while still comfortably catching\n * every realistic decimal/thousands-separator flip.\n */\nexport const PRICE_ANOMALY_THRESHOLD_MULTIPLIER = 10\n\nexport interface PriceAnomaly {\n  /** `curr.median / prev.median`. */\n  ratio: number\n  /** The prior run's median (minor units). */\n  prevMedian: number\n  /** This run's median (minor units). */\n  currMedian: number\n}\n\n/**\n * Compare two runs' price stats for a magnitude-scale anomaly. Returns\n * `null` when there's nothing to compare (either side has no median — zero\n * non-null price samples, e.g. a schema/transform that never emits a price\n * sample, or a run with zero ES-accepted rows) or the ratio falls within\n * the safe band.\n */\nexport function detectPriceAnomaly(prev: PriceStats, curr: PriceStats): PriceAnomaly | null {\n  if (prev.median === null || curr.median === null) return null\n  // A zero/negative median can't produce a meaningful ratio. Shouldn't occur\n  // for real minor-unit prices (`toMinorUnits` on a positive price); this is\n  // defensive-only so a pathological value can't produce Infinity/NaN below.\n  if (prev.median <= 0 || curr.median <= 0) return null\n  const ratio = curr.median / prev.median\n  if (\n    ratio >= PRICE_ANOMALY_THRESHOLD_MULTIPLIER ||\n    ratio <= 1 / PRICE_ANOMALY_THRESHOLD_MULTIPLIER\n  ) {\n    return { ratio, prevMedian: prev.median, currMedian: curr.median }\n  }\n  return null\n}\n\n/**\n * Narrow an `unknown` value (a `field.json()` column read back through\n * `AdminClient`, typed `JsonValue | undefined` on the DTO) to `PriceStats`,\n * or `null` when it doesn't structurally match — e.g. a run written before\n * this field existed, or a run whose transform never emitted a price\n * sample so `min`/`median`/`max` serialised as `null`. Defensive, matching\n * the `params` narrowing pattern already used in `worker.ts`.\n */\nexport function parsePriceStats(value: unknown): PriceStats | null {\n  if (typeof value !== 'object' || value === null) return null\n  const v = value as Record<string, unknown>\n  if (typeof v.count !== 'number' || typeof v.nullRate !== 'number') return null\n  return {\n    count: v.count,\n    nullRate: v.nullRate,\n    min: typeof v.min === 'number' ? v.min : null,\n    median: typeof v.median === 'number' ? v.median : null,\n    max: typeof v.max === 'number' ? v.max : null,\n  }\n}\n","/**\n * One-shot importer: stream rows → transform → batched bulk-write to ES,\n * accumulating per-row errors into `ErrorTracker` and reporting progress\n * to the queue every batch.\n *\n * Per PLAN-ECOMMERCE.md PR 7 (PoC scope):\n * - **Batch size 1000.** Matches giga-test precedent. Configurable for\n *   integration tests that don't want a 1k row floor.\n * - **No resumability and no automatic retries.** The `imports:run` queue\n *   job is registered with `defaultRetries: 0` (see `worker.ts`) so a\n *   failed handler does NOT re-enqueue itself — re-running a multi-batch\n *   import against the same `import_run.id` while the previous attempt\n *   may still be writing is a footgun (duplicate batches, double-counted\n *   progress). Operator retries by creating a NEW `import_run` row.\n * - **No per-supplier transform plugin.** The runner takes a single\n *   `RowTransform<TDoc>` from the registry and applies it to every row;\n *   PR 8 may diverge but only by registering a different transform name.\n * - **Direct `bulkUpsert` into the live aliased index** (D6 alias is\n *   set up by PR 4's `ensureAliasedIndex`; the importer doesn't reindex).\n *\n * Per D21 (sanctioned bulk path): this runner intentionally bypasses\n * AdminClient and entity hooks. Per-batch audit / observability lives\n * on the surrounding `import_run` row + queue progress, NOT per-row.\n */\n\nimport {\n  type BulkIndexResult,\n  bulkPartialUpdate,\n  bulkUpsert,\n  type EsClientLike,\n} from '@murumets-ee/search-elasticsearch'\nimport { ErrorTracker } from './error-tracker.js'\nimport { type PriceStats, PriceStatsCollector } from './price-stats.js'\nimport { type ReadFeedOptions, readFeed } from './read-feed.js'\nimport type {\n  DirectionalPair,\n  DocPatch,\n  RowTransform,\n  RowWarning,\n  TransformContext,\n} from './transform.js'\n\n/** Soft default; chosen to match giga-test. ES bulk requests over ~5MB get split server-side anyway. */\nexport const DEFAULT_BATCH_SIZE = 1000\n\nexport interface RunImportOptions<TDoc> {\n  /** UUID of the `import_run` row driving this run. Forwarded to every transform invocation. */\n  importRunId: string\n  /** Operator-supplied label for the run. Forwarded to the transform context. */\n  runLabel: string\n  /** Opaque per-run params copied from `import_run.params`. */\n  params: Record<string, unknown>\n  /** Transform applied to every parsed row. */\n  transform: RowTransform<TDoc>\n  /** Feed reader options — file path + format (csv/tsv/xlsx) + delimiter/header config. */\n  feed: ReadFeedOptions\n  /** ES client (low-level shape from `@murumets-ee/search-elasticsearch`). */\n  esClient: EsClientLike\n  /** Index alias to write to. Per D6, callers always pass an alias, never a physical index. */\n  esIndex: string\n  /** Rows per `bulkUpsert` call. Default {@link DEFAULT_BATCH_SIZE}. */\n  batchSize?: number\n  /**\n   * Callback invoked after every batch. The handler in `worker.ts`\n   * forwards this to `ctx.updateProgress` for the queue UI; tests\n   * inspect it directly. Synchronous + cheap so a slow callback can't\n   * back-pressure the importer.\n   */\n  onProgress?: (progress: ImportRunProgress) => void\n  /** Optional: stop processing after this many rows. Tests use it; production passes `undefined`. */\n  rowLimit?: number\n  /** Abort signal threaded into the underlying ES client request — cooperative cancel. */\n  signal?: AbortSignal\n  /** Optional ErrorTracker config (caps). Default: top-50 patterns × 5 samples. */\n  errorTracker?: ErrorTracker\n  /**\n   * Optional separate tracker for NON-FATAL row warnings (commerce-demo PR4 —\n   * `no_purchase_price`). Distinct from {@link errorTracker} so warnings never\n   * inflate `totals.failed` / the error summary: a warned row is a SUCCESS whose\n   * doc is indexed. Default: a fresh tracker with the same caps.\n   */\n  warningTracker?: ErrorTracker\n  /**\n   * Optional price-stats collector (F010 / #602) — accumulates a\n   * min/median/max/nullRate distribution over ES-accepted rows' optional\n   * `priceSample`s in the SAME streaming pass that builds `totals`. Default:\n   * a fresh collector. Tests can inject their own to inspect it directly —\n   * mirrors the {@link errorTracker} / {@link warningTracker} DI shape.\n   */\n  priceStats?: PriceStatsCollector\n  /**\n   * Optional structural logger for per-batch telemetry. Receives\n   * `{ batch, ms, succeeded, failed, rowsRead, rowsPerSecond }` after\n   * every flush, plus a one-shot `started` and `finished` event. The\n   * worker forwards `app.logger.child({ pkg: 'imports' })`; tests\n   * leave it undefined.\n   *\n   * Why: without per-batch visibility, a stalled run looks identical\n   * to \"first batch in progress\" forever — the only feedback is the\n   * queue's `progress` jsonb (which writes after each batch) and the\n   * eventual `succeeded`/`failed` job event. Adding this lets the\n   * operator tail the worker log and immediately see whether ES is\n   * rate-limiting, the transform is rejecting every row, or the\n   * stream is actually progressing.\n   */\n  logger?: {\n    info: (data: Record<string, unknown>, msg: string) => void\n    warn: (data: Record<string, unknown>, msg: string) => void\n  }\n}\n\n/**\n * Progress payload written to `toolkit_jobs.progress` after every batch.\n * Caps + flush rules live on the queue's `updateProgress` debounce —\n * callers don't need to throttle.\n */\nexport interface ImportRunProgress {\n  rowsRead: number\n  rowsSucceeded: number\n  rowsFailed: number\n  rowsSkipped: number\n  /** Rows indexed WITH a non-fatal warning (commerce-demo PR4 — `no_purchase_price`). */\n  rowsWarned: number\n  batchesCompleted: number\n  /** Wall-clock seconds since the runner started. */\n  elapsedSeconds: number\n  /** Rows / second, computed at every batch. */\n  rowsPerSecond: number\n  /** Distinct error patterns currently held by the tracker. Saturates at the cap. */\n  distinctErrorPatterns: number\n  /**\n   * Total data-row count from a pre-pass scan, when the worker computed one.\n   * Lets the UI render `processed / total` percent. Absent when the worker\n   * skipped the count (file too large for the count budget, count failed,\n   * etc.) — UI must treat it as optional.\n   */\n  totalRows?: number\n}\n\n/**\n * Final result returned by {@link runImport}. The handler writes these\n * onto the `import_run` row alongside the ErrorTracker snapshot.\n */\nexport interface RunImportResult {\n  /** Total rows read from the file (excludes skipped empty lines). */\n  rowsRead: number\n  /** Rows the transform turned into a successful doc AND the ES cluster acknowledged. */\n  rowsSucceeded: number\n  /**\n   * Rows that the transform rejected (`{ kind: 'error' }`) OR that ES\n   * rejected on bulk-write (per-doc failure). Both are aggregated by\n   * `errorTracker` for the import_run summary.\n   */\n  rowsFailed: number\n  /** Rows that the transform skipped (`{ kind: 'skip' }`) — header noise, blank lines, intentional drop. */\n  rowsSkipped: number\n  /**\n   * Rows that succeeded (doc indexed) but carried ≥1 non-fatal warning —\n   * counted SEPARATELY from `rowsSucceeded`/`rowsFailed` (a warned row is also\n   * counted in `rowsSucceeded`). commerce-demo PR4: `no_purchase_price`.\n   */\n  rowsWarned: number\n  /** Number of `bulkUpsert` calls made. */\n  batchesCompleted: number\n  /** Final value of {@link ErrorTracker.snapshot}. */\n  errors: ReturnType<ErrorTracker['snapshot']>\n  /** Aggregated NON-FATAL warnings (commerce-demo PR4). Empty when no row warned. */\n  warnings: ReturnType<ErrorTracker['snapshot']>\n  /**\n   * Directional supersession pairs harvested as a side-output (commerce-demo PR6,\n   * SD007). Buffered in memory from ES-ACCEPTED rows only (same accounting as\n   * warnings — a row whose doc was rejected by ES contributes no pair). Empty unless\n   * a schema emits them (the `parts` schema, when a row's `superseding_code` is set).\n   * The worker hands these to an injected `pairSink` AFTER the run for the commerce\n   * replacement-merge engine; the runner itself never writes the consumer's tables.\n   */\n  pairs: readonly DirectionalPair[]\n  /**\n   * Doc-patches applied by the phase-2 pass (commerce-demo PR7): a `<code>*S`\n   * core-fee row is not a saleable doc — it emits a {@link DocPatch} that merges its\n   * deposit onto the PARENT part's `core_charge_minor`. The runner buffers these\n   * during the stream (sparse) and applies them via `bulkPartialUpdate` AFTER the ES\n   * pass. `patchesApplied` = targets the cluster merged; `patchesOrphaned` = patches\n   * whose parent genuinely wasn't in the index (a `document_missing`/404 → an\n   * `orphan_core_charge` warning). An INFRA failure (whole-batch transport error, or a\n   * non-404 per-doc failure) is NOT counted as orphaned — it lands in a distinct\n   * `patch_attach_failed` warning bucket so it can't masquerade as a data orphan. Zero\n   * unless a schema emits patches.\n   */\n  patchesApplied: number\n  patchesOrphaned: number\n  /**\n   * Price-distribution snapshot (F010 / #602) accumulated from ES-accepted\n   * rows' `priceSample`s during this same streaming pass. See `price-stats.ts`.\n   * `count` is `0` (all fields `null`/`0`) for a run whose schema never emits\n   * a price sample.\n   */\n  priceStats: PriceStats\n}\n\n/**\n * Apply the runner against a feed file. Stops on rowLimit OR end-of-file\n * OR if `signal` aborts. Throws if the streaming reader / ES client\n * throws — caller (the queue handler) catches that and writes\n * `import_run.status = 'failed'` with the error message in\n * `errorSummary.fatal`.\n */\nexport async function runImport<TDoc>(options: RunImportOptions<TDoc>): Promise<RunImportResult> {\n  const {\n    importRunId,\n    runLabel,\n    params,\n    transform,\n    feed,\n    esClient,\n    esIndex,\n    batchSize = DEFAULT_BATCH_SIZE,\n    onProgress,\n    rowLimit,\n    signal,\n    errorTracker = new ErrorTracker(),\n    warningTracker = new ErrorTracker(),\n    priceStats: priceStatsCollector = new PriceStatsCollector(),\n    logger,\n  } = options\n\n  if (batchSize < 1) {\n    throw new Error(`batchSize must be >= 1 (got ${batchSize})`)\n  }\n\n  const startedAt = Date.now()\n  logger?.info({ importRunId, runLabel, batchSize, esIndex }, 'imports.runner: started')\n  let rowsRead = 0\n  let rowsSucceeded = 0\n  let rowsFailed = 0\n  let rowsSkipped = 0\n  let rowsWarned = 0\n  let batchesCompleted = 0\n\n  // Side-output buffer (commerce-demo PR6, SD007): directional supersession pairs\n  // accumulated from ES-ACCEPTED rows across the whole run. An array append per\n  // sparse superseded row — it never blocks the `bulkUpsert` loop. Handed to the\n  // injected `pairSink` by the worker AFTER the run (read-once / merge-after).\n  const collectedPairs: DirectionalPair[] = []\n\n  // Side-output buffer (commerce-demo PR7): doc-patches from `*S` core-fee rows —\n  // each attaches a core charge onto a DIFFERENT (parent) doc. Collected off the\n  // `skip` result (a `*S` row is never a saleable doc), sparse, applied in a phase-2\n  // `bulkPartialUpdate` after the ES pass (below) — order-independent because ES\n  // `update` is realtime by `_id`.\n  const collectedPatches: DocPatch<TDoc>[] = []\n\n  // Warnings + pairs ride on the pending item (with the raw row for the warning\n  // sample) so they're accounted at FLUSH time — only for rows ES actually accepted.\n  // A warned row is by contract a SUCCESSFUL row, so counting it at push time (before\n  // the bulk response) would wrongly keep a later-ES-rejected row in rowsWarned /\n  // warnings; pairs follow the same accepted-rows-only rule.\n  let pending: Array<{\n    id: string\n    doc: TDoc\n    rowNumber: number\n    rowData: Record<string, string>\n    warnings?: readonly RowWarning[]\n    pairs?: readonly DirectionalPair[]\n    priceSample?: number\n  }> = []\n\n  const flush = async (): Promise<void> => {\n    if (pending.length === 0) return\n    const batch = pending\n    pending = []\n    const flushStartedAt = Date.now()\n    let result: BulkIndexResult\n    try {\n      result = await bulkUpsert<TDoc>(esClient, {\n        index: esIndex,\n        docs: batch.map(({ id, doc }) => ({ id, doc })),\n        ...(signal !== undefined && { signal }),\n      })\n    } catch (err) {\n      // Cluster- or transport-level failure — the whole batch is\n      // unaccounted for. Distinguish abort (operator-driven cancel) from\n      // a real cluster failure so the errorSummary doesn't mislabel a\n      // cancelled run as broken cluster connectivity.\n      const isAbort =\n        signal?.aborted ??\n        (err instanceof Error && (err.name === 'AbortError' || /abort/i.test(err.message)))\n      const errorType = isAbort ? 'aborted' : 'bulk_request_failed'\n      const reason = err instanceof Error ? err.message : String(err)\n      for (const { rowNumber } of batch) {\n        errorTracker.addError(rowNumber, errorType, reason, undefined, null)\n      }\n      rowsFailed += batch.length\n      throw err\n    }\n\n    rowsSucceeded += result.succeeded\n    rowsFailed += result.failures.length\n    batchesCompleted += 1\n\n    const flushMs = Date.now() - flushStartedAt\n    const elapsedSec = (Date.now() - startedAt) / 1000\n    logger?.info(\n      {\n        importRunId,\n        batch: batchesCompleted,\n        flushMs,\n        submitted: batch.length,\n        succeeded: result.succeeded,\n        failed: result.failures.length,\n        rowsRead,\n        rowsPerSecond: elapsedSec > 0 ? Math.round(rowsRead / elapsedSec) : 0,\n      },\n      'imports.runner: batch flushed',\n    )\n\n    if (result.failures.length > 0) {\n      // Map each ES failure back to its source row via `id`. The bulk\n      // response order matches the request order, but ES doesn't promise\n      // that; matching by `id` is the safe path. PoC volume is small\n      // enough that the O(failures × batch) cost is irrelevant.\n      const byId = new Map(batch.map((b) => [b.id, b.rowNumber]))\n      for (const fail of result.failures) {\n        const rowNumber = byId.get(fail.id) ?? -1\n        errorTracker.addError(rowNumber, fail.type, fail.reason, undefined, { id: fail.id })\n      }\n    }\n\n    // Account non-fatal warnings ONLY for rows ES accepted (commerce-demo PR4) — a\n    // warned row is by contract a SUCCESSFUL row, so a row that warned AND then failed\n    // the bulk write must NOT land in rowsWarned/warnings (it's already in rowsFailed).\n    const failedIds = new Set(result.failures.map((f) => f.id))\n    for (const item of batch) {\n      if (!item.warnings || item.warnings.length === 0) continue\n      if (failedIds.has(item.id)) continue\n      rowsWarned += 1\n      for (const w of item.warnings) {\n        warningTracker.addError(item.rowNumber, w.warningType, w.message, w.field, item.rowData)\n      }\n    }\n\n    // Buffer directional pairs for rows ES accepted (commerce-demo PR6, SD007) — same\n    // accepted-rows-only rule as warnings, so a pair from a doc ES rejected is dropped.\n    for (const item of batch) {\n      if (!item.pairs || item.pairs.length === 0) continue\n      if (failedIds.has(item.id)) continue\n      collectedPairs.push(...item.pairs)\n    }\n\n    // Price-distribution sampling (F010 / #602) — same accepted-rows-only rule as\n    // warnings/pairs: a row ES rejected contributes no sample. Every ES-accepted row\n    // records exactly once (`undefined` counts toward `nullRate` rather than being\n    // skipped), so `priceStats.count` tracks `rowsSucceeded` exactly.\n    for (const item of batch) {\n      if (failedIds.has(item.id)) continue\n      priceStatsCollector.record(item.priceSample)\n    }\n\n    if (onProgress) {\n      const elapsedSeconds = (Date.now() - startedAt) / 1000\n      onProgress({\n        rowsRead,\n        rowsSucceeded,\n        rowsFailed,\n        rowsSkipped,\n        rowsWarned,\n        batchesCompleted,\n        elapsedSeconds,\n        rowsPerSecond: elapsedSeconds > 0 ? rowsRead / elapsedSeconds : 0,\n        distinctErrorPatterns: errorTracker.getDistinctPatternCount(),\n      })\n    }\n  }\n\n  // Streaming heartbeat — sample every Nth row so the operator sees the\n  // streamer is alive even when a transform-rejected file would prevent\n  // any flush from firing. Without this, a feed that's 100% rejected\n  // looks identical to \"stuck\" until end-of-file. Sample frequency is\n  // batchSize so the cadence matches the natural batch rhythm.\n  const streamSampleEvery = batchSize\n\n  // First-error capture per error type — surface what's actually rejecting\n  // the rows when 100% of a feed is rejected. Without this, the only signal\n  // is `errorTracker.snapshot()` which only writes to import_run.errorSummary\n  // at the END of the run. Mid-run we want to know NOW.\n  const firstErrorByType = new Map<string, { rowNumber: number; message: string }>()\n\n  for await (const { rowNumber, row } of readFeed(feed)) {\n    if (signal?.aborted) break\n    if (rowLimit !== undefined && rowsRead >= rowLimit) break\n    rowsRead += 1\n\n    if (rowsRead === 1) {\n      logger?.info(\n        { importRunId, rowNumber, sampleKeys: Object.keys(row) },\n        'imports.runner: STREAM_FIRST_ROW',\n      )\n    }\n    if (rowsRead % streamSampleEvery === 0) {\n      const elapsedSec = (Date.now() - startedAt) / 1000\n      // Telemetry only — no row content, no error sample text.\n      // PR #262 review (M-5): the prior `firstErrors:\n      // Object.fromEntries(firstErrorByType)` payload included the\n      // raw error message captured from the transform, which for the\n      // carmaker feed embeds supplier codes + price fragments. The\n      // detailed first-error breakdown is already persisted on\n      // `import_run.errorSummary` via `errorTracker.snapshot()` —\n      // operators inspect it there, not in the worker log.\n      logger?.info(\n        {\n          importRunId,\n          rowsRead,\n          rowsSucceeded,\n          rowsFailed,\n          rowsSkipped,\n          pending: pending.length,\n          elapsedSec: Math.round(elapsedSec * 10) / 10,\n          rowsPerSecond: elapsedSec > 0 ? Math.round(rowsRead / elapsedSec) : 0,\n          distinctFirstErrorTypes: firstErrorByType.size,\n        },\n        'imports.runner: STREAM_TICK',\n      )\n    }\n\n    const ctx: TransformContext = { importRunId, params, runLabel, rowNumber }\n    let result: Awaited<ReturnType<typeof transform>>\n    try {\n      result = await transform(row, ctx)\n    } catch (err) {\n      // A throw from the transform is a programmer error — surface it as\n      // a row-level error so the run can continue. (If the bug is\n      // catastrophic, the operator sees the same message repeated and\n      // can stop the run.)\n      const reason = err instanceof Error ? err.message : String(err)\n      errorTracker.addError(rowNumber, 'transform_threw', reason, undefined, row)\n      if (!firstErrorByType.has('transform_threw')) {\n        firstErrorByType.set('transform_threw', { rowNumber, message: reason })\n        // PR #262 review (M-5): drop the `row` payload from the\n        // log line — supplier feeds carry pricing + internal codes.\n        // The full row is on `import_run.errorSummary.samples` for\n        // post-mortem review by operators with view permission.\n        logger?.warn(\n          { importRunId, rowNumber, errorType: 'transform_threw', message: reason },\n          'imports.runner: FIRST_ERROR transform_threw',\n        )\n      }\n      rowsFailed += 1\n      continue\n    }\n\n    if (result.kind === 'skip') {\n      rowsSkipped += 1\n      // A `*S` core-fee row skips the saleable index but may carry a doc-patch\n      // (commerce-demo PR7) — buffer it for the phase-2 attach. Unlike pairs, a\n      // patch has no ES write of its own to gate on, so it's collected here\n      // unconditionally; its success is decided by the phase-2 update.\n      if (result.patches && result.patches.length > 0) {\n        collectedPatches.push(...result.patches)\n      }\n      continue\n    }\n    if (result.kind === 'error') {\n      errorTracker.addError(\n        rowNumber,\n        result.error.errorType,\n        result.error.message,\n        result.error.field,\n        row,\n      )\n      if (!firstErrorByType.has(result.error.errorType)) {\n        firstErrorByType.set(result.error.errorType, {\n          rowNumber,\n          message: result.error.message,\n        })\n        logger?.warn(\n          {\n            importRunId,\n            rowNumber,\n            errorType: result.error.errorType,\n            field: result.error.field,\n            message: result.error.message,\n          },\n          `imports.runner: FIRST_ERROR ${result.error.errorType}`,\n        )\n      }\n      rowsFailed += 1\n      continue\n    }\n\n    // Non-fatal warnings ride on a SUCCESS — the doc is still indexed\n    // (commerce-demo PR4: `no_purchase_price`). Carry them (+ the raw row for the\n    // sample) on the pending item; the flush aggregates them into the separate\n    // `warningTracker` AFTER the bulk response confirms the row was accepted, so a\n    // warned-then-ES-rejected row never inflates rowsWarned / the warning summary.\n    pending.push({\n      id: result.id,\n      doc: result.doc,\n      rowNumber,\n      rowData: row,\n      ...(result.warnings && result.warnings.length > 0 && { warnings: result.warnings }),\n      ...(result.pairs && result.pairs.length > 0 && { pairs: result.pairs }),\n      ...(result.priceSample !== undefined && { priceSample: result.priceSample }),\n    })\n    if (pending.length >= batchSize) {\n      await flush()\n    }\n  }\n\n  await flush()\n\n  // Phase 2 (commerce-demo PR7): attach the buffered `*S` core-charge patches onto\n  // their PARENT docs. Runs AFTER the whole stream so every parent that exists in\n  // this run is already indexed; ES `update` is realtime by `_id`, so the merge\n  // lands regardless of whether the parent streamed before or after its `*S` row\n  // (order-independent). Patches are sparse (only reman parts carry `*S`), but batch\n  // anyway for a giant catalogue. A patch whose parent isn't in the index comes back\n  // as a per-doc failure (`document_missing_exception`) → an `orphan_core_charge`\n  // warning, never a phantom doc, never a silent drop (D004). A cluster failure of\n  // the whole batch is logged as a warning too — the saleable docs already indexed;\n  // only the deferred core-charge attach didn't happen — so we never fail the run\n  // over it.\n  let patchesApplied = 0\n  let patchesOrphaned = 0\n  for (let i = 0; i < collectedPatches.length; i += batchSize) {\n    // Operator-driven cancel BETWEEN batches: THROW rather than break, so a run\n    // cancelled after phase 1 is not marked `succeeded` with core charges left\n    // unattached. Mirrors phase-1's flush, which also re-throws on abort. The queue\n    // handler catches this and marks the run failed.\n    if (signal?.aborted) {\n      throw new Error('imports: run aborted during core-charge attach (phase 2)')\n    }\n    const slice = collectedPatches.slice(i, i + batchSize)\n    let updateResult: BulkIndexResult\n    try {\n      updateResult = await bulkPartialUpdate<TDoc>(esClient, {\n        index: esIndex,\n        updates: slice.map(({ id, doc }) => ({ id, doc })),\n        ...(signal !== undefined && { signal }),\n      })\n    } catch (err) {\n      // An abort surfacing as an AbortError from the ES client is a cancel, NOT an\n      // attach failure — re-throw so the run fails (same as phase 1), never a silent\n      // \"succeeded with patches unapplied\".\n      const isAbort =\n        signal?.aborted ??\n        (err instanceof Error && (err.name === 'AbortError' || /abort/i.test(err.message)))\n      if (isAbort) throw err\n      // Whole-batch transport/cluster failure (ES down, timeout). The parents may\n      // well exist — the attach simply didn't happen — so this is an INFRA failure,\n      // NOT a data orphan: it gets the distinct `patch_attach_failed` bucket and is\n      // NOT counted in patchesOrphaned. Best-effort: the saleable docs already\n      // indexed, and a re-import re-attaches, so we warn + continue rather than fail.\n      const reason = err instanceof Error ? err.message : String(err)\n      for (const patch of slice) {\n        warningTracker.addError(-1, 'patch_attach_failed', reason, patch.orphanWarning?.field, {\n          target: patch.id,\n        })\n      }\n      continue\n    }\n    patchesApplied += updateResult.succeeded\n    if (updateResult.failures.length > 0) {\n      const byId = new Map(slice.map((p) => [p.id, p]))\n      for (const fail of updateResult.failures) {\n        const w = byId.get(fail.id)?.orphanWarning\n        // A missing target (`document_missing_exception` / 404) is a genuine data\n        // orphan — the referenced parent isn't in this import (D004: warn, never a\n        // phantom doc). A stable message (from the patch's orphanWarning) aggregates\n        // orphans; the missing parent id rides in the sample. Any OTHER per-doc\n        // failure (mapper error, etc.) is an infra/attach failure kept in a distinct\n        // bucket so it can't masquerade as a data orphan.\n        const isOrphan = fail.status === 404 || fail.type === 'document_missing_exception'\n        if (isOrphan) {\n          patchesOrphaned += 1\n          warningTracker.addError(\n            -1,\n            w?.warningType ?? 'orphan_patch',\n            w?.message ?? fail.reason,\n            w?.field,\n            {\n              target: fail.id,\n            },\n          )\n        } else {\n          warningTracker.addError(-1, 'patch_attach_failed', fail.reason, w?.field, {\n            target: fail.id,\n          })\n        }\n      }\n    }\n  }\n\n  const totalSec = (Date.now() - startedAt) / 1000\n  logger?.info(\n    {\n      importRunId,\n      rowsRead,\n      rowsSucceeded,\n      rowsFailed,\n      rowsSkipped,\n      batchesCompleted,\n      patchesApplied,\n      patchesOrphaned,\n      totalSec: Math.round(totalSec * 10) / 10,\n      rowsPerSecond: totalSec > 0 ? Math.round(rowsRead / totalSec) : 0,\n    },\n    'imports.runner: finished',\n  )\n\n  return {\n    rowsRead,\n    rowsSucceeded,\n    rowsFailed,\n    rowsSkipped,\n    rowsWarned,\n    batchesCompleted,\n    errors: errorTracker.snapshot(),\n    warnings: warningTracker.snapshot(),\n    pairs: collectedPairs,\n    patchesApplied,\n    patchesOrphaned,\n    priceStats: priceStatsCollector.snapshot(),\n  }\n}\n"],"mappings":"mIAmEA,MAAa,EAAuB,GACvB,EAAkC,EAY/C,IAAa,EAAb,KAA0B,CACxB,SAA4B,IAAI,IAChC,YACA,qBACA,kBAA4B,EAE5B,YAAY,EAA6B,CAAC,EAAG,CAC3C,KAAK,YAAc,EAAO,aAAA,GAC1B,KAAK,qBAAuB,EAAO,sBAAA,CACrC,CAWA,SACE,EACA,EACA,EACA,EACA,EACM,CAEN,IAAM,EAAY,GAAG,EAAU,GADd,GAAS,UACiB,GAAG,IAE1C,EAAU,KAAK,SAAS,IAAI,CAAS,EACzC,GAAI,CAAC,EAAS,CACZ,GAAI,KAAK,SAAS,MAAQ,KAAK,YAAa,CAC1C,KAAK,mBAAqB,EAC1B,MACF,CACA,EAAU,CACR,YACA,MAAO,GAAS,KAChB,UACA,MAAO,EACP,gBAAiB,EACjB,eAAgB,EAChB,QAAS,CAAC,CACZ,EACA,KAAK,SAAS,IAAI,EAAW,CAAO,CACtC,CAEA,EAAQ,OAAS,EACjB,EAAQ,eAAiB,EACrB,EAAQ,QAAQ,OAAS,KAAK,sBAChC,EAAQ,QAAQ,KAAK,CAAE,YAAW,SAAQ,CAAC,CAE/C,CAGA,oBAA6B,CAC3B,IAAI,EAAM,EACV,IAAK,IAAM,KAAK,KAAK,SAAS,OAAO,EAAG,GAAO,EAAE,MACjD,OAAO,CACT,CAGA,yBAAkC,CAChC,OAAO,KAAK,SAAS,IACvB,CAOA,0BAAmC,CACjC,OAAO,KAAK,iBACd,CAOA,gBAAiC,CAC/B,IAAM,EAAM,MAAM,KAAK,KAAK,SAAS,OAAO,CAAC,CAAC,CAAC,MAAM,EAAG,IAAM,EAAE,MAAQ,EAAE,KAAK,EACzE,EAAc,EAAI,QAAQ,EAAK,IAAM,EAAM,EAAE,MAAO,CAAC,EAC3D,OAAO,EAAI,IAAK,IAAO,CACrB,UAAW,EAAE,UACb,MAAO,EAAE,MACT,QAAS,EAAE,QACX,MAAO,EAAE,MACT,gBAAiB,EAAE,gBACnB,eAAgB,EAAE,eAClB,QAAS,EAAE,QAAQ,MAAM,EACzB,WAAY,EAAc,EAAK,EAAE,MAAQ,EAAe,IAAM,CAChE,EAAE,CACJ,CAOA,UAAiC,CAC/B,MAAO,CACL,YAAa,KAAK,mBAAmB,EACrC,iBAAkB,KAAK,wBAAwB,EAC/C,kBAAmB,KAAK,kBACxB,SAAU,KAAK,eAAe,CAChC,CACF,CACF,EAoBA,SAAgB,EACd,EACA,EACsB,CACtB,IAAM,EAAc,EAAE,YAAc,EAAE,YAOhC,EAAW,IAAmC,CAClD,GAAG,EACH,WAAY,EAAc,EAAK,EAAE,MAAQ,EAAe,IAAM,CAChE,GACA,MAAO,CACL,cACA,iBAAkB,EAAE,iBAAmB,EAAE,iBACzC,kBAAmB,EAAE,kBAAoB,EAAE,kBAC3C,SAAU,CAAC,GAAG,EAAE,SAAS,IAAI,CAAO,EAAG,GAAG,EAAE,SAAS,IAAI,CAAO,CAAC,CACnE,CACF,CC1LA,MAAa,EAAgC,CAC3C,MAAO,EACP,SAAU,EACV,IAAK,KACL,OAAQ,KACR,IAAK,IACP,EAca,EAAmB,IAQhC,IAAa,EAAb,KAAiC,CAC/B,MAAgB,EAChB,UAAoB,EACpB,YAAsB,EACtB,QAAqC,CAAC,EAmBtC,OAAO,EAAiC,CAEtC,GADA,KAAK,OAAS,EACV,IAAU,IAAA,GAAW,CACvB,KAAK,WAAa,EAClB,MACF,CAEA,GADA,KAAK,aAAe,EAChB,KAAK,QAAQ,OAAA,IAA2B,CAC1C,KAAK,QAAQ,KAAK,CAAK,EACvB,MACF,CACA,IAAM,EAAI,KAAK,MAAM,KAAK,OAAO,EAAI,KAAK,WAAW,EACjD,EAAA,MAAsB,KAAK,QAAQ,GAAK,EAC9C,CAEA,UAAuB,CACrB,GAAI,KAAK,QAAU,EAAG,MAAO,CAAE,GAAG,CAAkB,EACpD,IAAM,EAAW,KAAK,UAAY,KAAK,MACvC,GAAI,KAAK,QAAQ,SAAW,EAC1B,MAAO,CAAE,MAAO,KAAK,MAAO,WAAU,IAAK,KAAM,OAAQ,KAAM,IAAK,IAAK,EAE3E,IAAM,EAAS,CAAC,GAAG,KAAK,OAAO,CAAC,CAAC,MAAM,EAAG,IAAM,EAAI,CAAC,EAC/C,EAAM,EAAO,IAAM,KACnB,EAAM,EAAO,EAAO,OAAS,IAAM,KACnC,EAAM,KAAK,MAAM,EAAO,OAAS,CAAC,EACpC,EACJ,GAAI,EAAO,OAAS,GAAM,EAAG,CAC3B,IAAM,EAAK,EAAO,EAAM,GAClB,EAAK,EAAO,GAClB,EAAS,IAAO,IAAA,IAAa,IAAO,IAAA,IAAa,EAAK,GAAM,EAAI,IAClE,KACE,GAAS,EAAO,IAAQ,KAE1B,MAAO,CAAE,MAAO,KAAK,MAAO,WAAU,MAAK,SAAQ,KAAI,CACzD,CACF,EAoBA,MAAa,EAAqC,GAkBlD,SAAgB,EAAmB,EAAkB,EAAuC,CAK1F,GAJI,EAAK,SAAW,MAAQ,EAAK,SAAW,MAIxC,EAAK,QAAU,GAAK,EAAK,QAAU,EAAG,OAAO,KACjD,IAAM,EAAQ,EAAK,OAAS,EAAK,OAOjC,OALE,GAAA,IACA,GAAS,EAAA,GAEF,CAAE,QAAO,WAAY,EAAK,OAAQ,WAAY,EAAK,MAAO,EAE5D,IACT,CAUA,SAAgB,EAAgB,EAAmC,CACjE,GAAI,OAAO,GAAU,WAAY,EAAgB,OAAO,KACxD,IAAM,EAAI,EAEV,OADI,OAAO,EAAE,OAAU,UAAY,OAAO,EAAE,UAAa,SAAiB,KACnE,CACL,MAAO,EAAE,MACT,SAAU,EAAE,SACZ,IAAK,OAAO,EAAE,KAAQ,SAAW,EAAE,IAAM,KACzC,OAAQ,OAAO,EAAE,QAAW,SAAW,EAAE,OAAS,KAClD,IAAK,OAAO,EAAE,KAAQ,SAAW,EAAE,IAAM,IAC3C,CACF,CChKA,MAAa,EAAqB,IAoKlC,eAAsB,EAAgB,EAA2D,CAC/F,GAAM,CACJ,cACA,WACA,SACA,YACA,OACA,WACA,UACA,YAAY,EACZ,aACA,WACA,SACA,eAAe,IAAI,EACnB,iBAAiB,IAAI,EACrB,WAAY,EAAsB,IAAI,EACtC,UACE,EAEJ,GAAI,EAAY,EACd,MAAU,MAAM,+BAA+B,EAAU,EAAE,EAG7D,IAAM,EAAY,KAAK,IAAI,EAC3B,GAAQ,KAAK,CAAE,cAAa,WAAU,YAAW,SAAQ,EAAG,yBAAyB,EACrF,IAAI,EAAW,EACX,EAAgB,EAChB,EAAa,EACb,EAAc,EACd,EAAa,EACb,EAAmB,EAMjB,EAAoC,CAAC,EAOrC,EAAqC,CAAC,EAOxC,EAQC,CAAC,EAEA,EAAQ,SAA2B,CACvC,GAAI,EAAQ,SAAW,EAAG,OAC1B,IAAM,EAAQ,EACd,EAAU,CAAC,EACX,IAAM,EAAiB,KAAK,IAAI,EAC5B,EACJ,GAAI,CACF,EAAS,MAAM,EAAiB,EAAU,CACxC,MAAO,EACP,KAAM,EAAM,KAAK,CAAE,KAAI,UAAW,CAAE,KAAI,KAAI,EAAE,EAC9C,GAAI,IAAW,IAAA,IAAa,CAAE,QAAO,CACvC,CAAC,CACH,OAAS,EAAK,CAQZ,IAAM,EAFJ,GAAQ,UACP,aAAe,QAAU,EAAI,OAAS,cAAgB,SAAS,KAAK,EAAI,OAAO,IACtD,UAAY,sBAClC,EAAS,aAAe,MAAQ,EAAI,QAAU,OAAO,CAAG,EAC9D,IAAK,GAAM,CAAE,eAAe,EAC1B,EAAa,SAAS,EAAW,EAAW,EAAQ,IAAA,GAAW,IAAI,EAGrE,KADA,IAAc,EAAM,OACd,CACR,CAEA,GAAiB,EAAO,UACxB,GAAc,EAAO,SAAS,OAC9B,GAAoB,EAEpB,IAAM,EAAU,KAAK,IAAI,EAAI,EACvB,GAAc,KAAK,IAAI,EAAI,GAAa,IAe9C,GAdA,GAAQ,KACN,CACE,cACA,MAAO,EACP,UACA,UAAW,EAAM,OACjB,UAAW,EAAO,UAClB,OAAQ,EAAO,SAAS,OACxB,WACA,cAAe,EAAa,EAAI,KAAK,MAAM,EAAW,CAAU,EAAI,CACtE,EACA,+BACF,EAEI,EAAO,SAAS,OAAS,EAAG,CAK9B,IAAM,EAAO,IAAI,IAAI,EAAM,IAAK,GAAM,CAAC,EAAE,GAAI,EAAE,SAAS,CAAC,CAAC,EAC1D,IAAK,IAAM,KAAQ,EAAO,SAAU,CAClC,IAAM,EAAY,EAAK,IAAI,EAAK,EAAE,GAAK,GACvC,EAAa,SAAS,EAAW,EAAK,KAAM,EAAK,OAAQ,IAAA,GAAW,CAAE,GAAI,EAAK,EAAG,CAAC,CACrF,CACF,CAKA,IAAM,EAAY,IAAI,IAAI,EAAO,SAAS,IAAK,GAAM,EAAE,EAAE,CAAC,EAC1D,IAAK,IAAM,KAAQ,EACb,MAAC,EAAK,UAAY,EAAK,SAAS,SAAW,IAC3C,GAAU,IAAI,EAAK,EAAE,EACzB,IAAc,EACd,IAAK,IAAM,KAAK,EAAK,SACnB,EAAe,SAAS,EAAK,UAAW,EAAE,YAAa,EAAE,QAAS,EAAE,MAAO,EAAK,OAAO,CAF3E,CAQhB,IAAK,IAAM,KAAQ,EACb,CAAC,EAAK,OAAS,EAAK,MAAM,SAAW,GACrC,EAAU,IAAI,EAAK,EAAE,GACzB,EAAe,KAAK,GAAG,EAAK,KAAK,EAOnC,IAAK,IAAM,KAAQ,EACb,EAAU,IAAI,EAAK,EAAE,GACzB,EAAoB,OAAO,EAAK,WAAW,EAG7C,GAAI,EAAY,CACd,IAAM,GAAkB,KAAK,IAAI,EAAI,GAAa,IAClD,EAAW,CACT,WACA,gBACA,aACA,cACA,aACA,mBACA,iBACA,cAAe,EAAiB,EAAI,EAAW,EAAiB,EAChE,sBAAuB,EAAa,wBAAwB,CAC9D,CAAC,CACH,CACF,EAOM,EAAoB,EAMpB,EAAmB,IAAI,IAE7B,UAAW,GAAM,CAAE,YAAW,SAAS,EAAS,CAAI,EAAG,CAErD,GADI,GAAQ,SACR,IAAa,IAAA,IAAa,GAAY,EAAU,MASpD,GARA,GAAY,EAER,IAAa,GACf,GAAQ,KACN,CAAE,cAAa,YAAW,WAAY,OAAO,KAAK,CAAG,CAAE,EACvD,kCACF,EAEE,EAAW,IAAsB,EAAG,CACtC,IAAM,GAAc,KAAK,IAAI,EAAI,GAAa,IAS9C,GAAQ,KACN,CACE,cACA,WACA,gBACA,aACA,cACA,QAAS,EAAQ,OACjB,WAAY,KAAK,MAAM,EAAa,EAAE,EAAI,GAC1C,cAAe,EAAa,EAAI,KAAK,MAAM,EAAW,CAAU,EAAI,EACpE,wBAAyB,EAAiB,IAC5C,EACA,6BACF,CACF,CAEA,IAAM,EAAwB,CAAE,cAAa,SAAQ,WAAU,WAAU,EACrE,EACJ,GAAI,CACF,EAAS,MAAM,EAAU,EAAK,CAAG,CACnC,OAAS,EAAK,CAKZ,IAAM,EAAS,aAAe,MAAQ,EAAI,QAAU,OAAO,CAAG,EAC9D,EAAa,SAAS,EAAW,kBAAmB,EAAQ,IAAA,GAAW,CAAG,EACrE,EAAiB,IAAI,iBAAiB,IACzC,EAAiB,IAAI,kBAAmB,CAAE,YAAW,QAAS,CAAO,CAAC,EAKtE,GAAQ,KACN,CAAE,cAAa,YAAW,UAAW,kBAAmB,QAAS,CAAO,EACxE,6CACF,GAEF,GAAc,EACd,QACF,CAEA,GAAI,EAAO,OAAS,OAAQ,CAC1B,GAAe,EAKX,EAAO,SAAW,EAAO,QAAQ,OAAS,GAC5C,EAAiB,KAAK,GAAG,EAAO,OAAO,EAEzC,QACF,CACA,GAAI,EAAO,OAAS,QAAS,CAC3B,EAAa,SACX,EACA,EAAO,MAAM,UACb,EAAO,MAAM,QACb,EAAO,MAAM,MACb,CACF,EACK,EAAiB,IAAI,EAAO,MAAM,SAAS,IAC9C,EAAiB,IAAI,EAAO,MAAM,UAAW,CAC3C,YACA,QAAS,EAAO,MAAM,OACxB,CAAC,EACD,GAAQ,KACN,CACE,cACA,YACA,UAAW,EAAO,MAAM,UACxB,MAAO,EAAO,MAAM,MACpB,QAAS,EAAO,MAAM,OACxB,EACA,+BAA+B,EAAO,MAAM,WAC9C,GAEF,GAAc,EACd,QACF,CAOA,EAAQ,KAAK,CACX,GAAI,EAAO,GACX,IAAK,EAAO,IACZ,YACA,QAAS,EACT,GAAI,EAAO,UAAY,EAAO,SAAS,OAAS,GAAK,CAAE,SAAU,EAAO,QAAS,EACjF,GAAI,EAAO,OAAS,EAAO,MAAM,OAAS,GAAK,CAAE,MAAO,EAAO,KAAM,EACrE,GAAI,EAAO,cAAgB,IAAA,IAAa,CAAE,YAAa,EAAO,WAAY,CAC5E,CAAC,EACG,EAAQ,QAAU,GACpB,MAAM,EAAM,CAEhB,CAEA,MAAM,EAAM,EAaZ,IAAI,EAAiB,EACjB,EAAkB,EACtB,IAAK,IAAI,EAAI,EAAG,EAAI,EAAiB,OAAQ,GAAK,EAAW,CAK3D,GAAI,GAAQ,QACV,MAAU,MAAM,0DAA0D,EAE5E,IAAM,EAAQ,EAAiB,MAAM,EAAG,EAAI,CAAS,EACjD,EACJ,GAAI,CACF,EAAe,MAAM,EAAwB,EAAU,CACrD,MAAO,EACP,QAAS,EAAM,KAAK,CAAE,KAAI,UAAW,CAAE,KAAI,KAAI,EAAE,EACjD,GAAI,IAAW,IAAA,IAAa,CAAE,QAAO,CACvC,CAAC,CACH,OAAS,EAAK,CAOZ,GAFE,GAAQ,UACP,aAAe,QAAU,EAAI,OAAS,cAAgB,SAAS,KAAK,EAAI,OAAO,IACrE,MAAM,EAMnB,IAAM,EAAS,aAAe,MAAQ,EAAI,QAAU,OAAO,CAAG,EAC9D,IAAK,IAAM,KAAS,EAClB,EAAe,SAAS,GAAI,sBAAuB,EAAQ,EAAM,eAAe,MAAO,CACrF,OAAQ,EAAM,EAChB,CAAC,EAEH,QACF,CAEA,GADA,GAAkB,EAAa,UAC3B,EAAa,SAAS,OAAS,EAAG,CACpC,IAAM,EAAO,IAAI,IAAI,EAAM,IAAK,GAAM,CAAC,EAAE,GAAI,CAAC,CAAC,CAAC,EAChD,IAAK,IAAM,KAAQ,EAAa,SAAU,CACxC,IAAM,EAAI,EAAK,IAAI,EAAK,EAAE,CAAC,EAAE,cAOZ,EAAK,SAAW,KAAO,EAAK,OAAS,8BAEpD,GAAmB,EACnB,EAAe,SACb,GACA,GAAG,aAAe,eAClB,GAAG,SAAW,EAAK,OACnB,GAAG,MACH,CACE,OAAQ,EAAK,EACf,CACF,GAEA,EAAe,SAAS,GAAI,sBAAuB,EAAK,OAAQ,GAAG,MAAO,CACxE,OAAQ,EAAK,EACf,CAAC,CAEL,CACF,CACF,CAEA,IAAM,GAAY,KAAK,IAAI,EAAI,GAAa,IAiB5C,OAhBA,GAAQ,KACN,CACE,cACA,WACA,gBACA,aACA,cACA,mBACA,iBACA,kBACA,SAAU,KAAK,MAAM,EAAW,EAAE,EAAI,GACtC,cAAe,EAAW,EAAI,KAAK,MAAM,EAAW,CAAQ,EAAI,CAClE,EACA,0BACF,EAEO,CACL,WACA,gBACA,aACA,cACA,aACA,mBACA,OAAQ,EAAa,SAAS,EAC9B,SAAU,EAAe,SAAS,EAClC,MAAO,EACP,iBACA,kBACA,WAAY,EAAoB,SAAS,CAC3C,CACF"}

package/dist/worker.mjs

@@ -1,2 +1,2 @@
-import{a as e,r as t}from"./read-feed-DGjqejw3.mjs";import{c as n,d as r,f as i,n as a,s as o}from"./runner-C9GIe5Dr.mjs";import{a as s,i as c,r as l}from"./publish-DCXIPwTE.mjs";import{runAsCli as u}from"@murumets-ee/core";import{z as d}from"zod";import{and as f,desc as p,eq as m}from"drizzle-orm";import{defineJob as h}from"@murumets-ee/queue/client";const g=d.object({importRunId:d.string().uuid()}),_=h({name:`imports:run`,description:`Stream a feed file, transform rows, bulk-write to Elasticsearch.`,schema:g,defaultRetries:0});function v(l){let{importRuns:d,transforms:h,esClient:g,esIndex:_,resolveFilePath:v,ensureIndex:b,pairSink:x,logger:S}=l;return async l=>u(async()=>{let{importRunId:u}=l.payload,C=S?.child({jobId:l.id,importRunId:u,type:`imports:run`}),w=await d.findById(u);if(!w){C?.warn(`import_run row not found — skipping`);return}w.status===`running`&&C?.warn({previousQueueJobId:w.queueJobId,previousStartedAt:w.startedAt},`imports: re-claiming a row left in running by a prior worker — likely a crash recovered by queue lock-recovery`);let T=h.get(w.transformName);if(!T)throw await y(d,u,`Unknown transform "${w.transformName}"`),Error(`No transform registered for name "${w.transformName}"`);let E=w.params&&typeof w.params==`object`&&!Array.isArray(w.params)?w.params:{},D=t(E);if(!D.success){let e=D.error.issues[0]?.message??`invalid feed reader config`;throw await y(d,u,`Invalid params.feed: ${e}`),Error(`Invalid params.feed for import_run ${u}: ${e}`)}let O=D.data;await d.update(u,{status:`running`,queueJobId:l.id,startedAt:new Date}),c(u);let k;try{k=await g()}catch(e){throw await y(d,u,`Could not resolve ES client: ${e instanceof Error?e.message:String(e)}`),e}if(b)try{await b()}catch(e){throw await y(d,u,`Could not ensure ES index/alias: ${e instanceof Error?e.message:String(e)}`),e}let A,j;if(v)try{let e=await v(w.filePath);A=e.localPath,j=e.cleanup}catch(e){throw await y(d,u,`Could not resolve filePath: ${e instanceof Error?e.message:String(e)}`),e}else A=w.filePath;let M;if(O.format!==`xlsx`)try{M=await e(A,O.hasHeader===void 0?{}:{hasHeader:O.hasHeader}),C?.info({importRunId:u,totalRows:M},`imports: pre-pass row count`)}catch(e){C?.warn({err:e},`imports: pre-pass row count failed — continuing without total`)}let N;try{let e={filePath:A,...O.format!==void 0&&{format:O.format},...O.delimiter!==void 0&&{delimiter:O.delimiter},...O.hasHeader!==void 0&&{hasHeader:O.hasHeader},...O.headerRowIndex!==void 0&&{headerRowIndex:O.headerRowIndex},...O.sheet!==void 0&&{sheet:O.sheet},...O.quote!==void 0&&{quote:O.quote}};N=await a({importRunId:u,runLabel:w.label,params:E,transform:T,feed:e,esClient:k,esIndex:_,onProgress:e=>l.updateProgress(M===void 0?e:{...e,totalRows:M}),...C!==void 0&&{logger:C}})}catch(e){throw await y(d,u,e instanceof Error?e.message:String(e)),j&&await j().catch(e=>{C?.warn({err:e},`filePath cleanup after failed run threw — ignoring`)}),e}j&&await j().catch(e=>{C?.warn({err:e},`filePath cleanup after successful run threw — ignoring`)});let P=N.warnings,F=typeof E.brandId==`string`?E.brandId:void 0,I=typeof E.supplierId==`string`?E.supplierId:void 0;if(F&&I&&N.priceStats.median!==null)try{let e=d.getTable(),t=await d.findMany({where:f(m(e.transformName,w.transformName),m(e.status,`succeeded`)),orderBy:p(e.createdAt),limit:50}),a=t.find(e=>{if(e.id===u)return!1;let t=e.params;if(!t||typeof t!=`object`||Array.isArray(t))return!1;let n=t;return n.brandId===F&&n.supplierId===I});!a&&t.length>=50&&C?.warn({brandId:F,supplierId:I,transformName:w.transformName},`imports: price-drift check found no prior run for this (brand, supplier) within the 50 most recent succeeded runs for this transform — a real prior run may exist beyond the bound, not necessarily absent`);let s=a?n(a.priceStats):null,c=s?o(s,N.priceStats):null;if(c){let e=c.ratio>=1,t=new r;t.addError(-1,`price_anomaly`,`Median purchase price is ${(e?c.ratio:1/c.ratio).toFixed(1)}x ${e?`higher`:`lower`} than the previous successful run for this supplier/brand (previous run ${a?.id??`unknown`}: median ${c.prevMedian}; this run: median ${c.currMedian}, minor units). Check for a decimal or thousands-separator format change in the source feed.`,`priceStats.median`,{previousImportRunId:a?.id??null,previousMedian:c.prevMedian,currentMedian:c.currMedian,ratio:c.ratio}),P=i(N.warnings,t.snapshot()),C?.warn({previousImportRunId:a?.id,ratio:c.ratio},`imports: price_anomaly — median purchase price shifted by a magnitude vs the prior run`)}}catch(e){C?.warn({err:e},`imports: price-drift comparison failed — continuing without it`)}if(await d.update(u,{status:`succeeded`,finishedAt:new Date,totals:{rowsRead:N.rowsRead,rowsSucceeded:N.rowsSucceeded,rowsFailed:N.rowsFailed,rowsSkipped:N.rowsSkipped,rowsWarned:N.rowsWarned,batchesCompleted:N.batchesCompleted,patchesApplied:N.patchesApplied,patchesOrphaned:N.patchesOrphaned},errorSummary:JSON.parse(JSON.stringify(N.errors)),warningSummary:JSON.parse(JSON.stringify(P)),priceStats:JSON.parse(JSON.stringify(N.priceStats))}),s(u,{rowsRead:N.rowsRead,rowsSucceeded:N.rowsSucceeded,rowsFailed:N.rowsFailed}),x&&N.pairs.length>0)try{await x(N.pairs,{importRunId:u,params:E,...C!==void 0&&{logger:C}}),C?.info({importRunId:u,pairs:N.pairs.length},`imports: pairSink invoked`)}catch(e){C?.warn({importRunId:u,pairs:N.pairs.length,err:e},`imports: pairSink threw — replacement ingestion incomplete, catalogue import unaffected`)}C?.info({rowsRead:N.rowsRead,rowsSucceeded:N.rowsSucceeded,rowsFailed:N.rowsFailed,rowsSkipped:N.rowsSkipped,batches:N.batchesCompleted},`import_run completed`)})}async function y(e,t,n){try{await e.update(t,{status:`failed`,finishedAt:new Date,errorSummary:{fatal:n}}),l(t,n)}catch{}}function b(e){let{importRuns:t,logger:n}=e;return{onJobDead:async e=>{e.jobType===`imports:run`&&await u(async()=>{let r=n?.child({event:`imports.run.dead`,jobId:e.jobId,type:`imports:run`}),i=t.getTable(),a=(await t.findMany({where:m(i.queueJobId,e.jobId),limit:1}))[0];if(!a){r?.warn(`queue.job.dead with no matching import_run — skipping`);return}if(a.status!==`running`&&a.status!==`pending`){r?.info({importRunId:a.id,status:a.status},`queue.job.dead arrived for a row already in a terminal state — leaving as-is`);return}a.status===`pending`&&r?.warn({importRunId:a.id,attempts:e.attempts},`queue.job.dead arrived for an import_run still in pending — worker crashed BEFORE claim/lookup`),await y(t,a.id,e.error),r?.info({importRunId:a.id,attempts:e.attempts,priorStatus:a.status},`queue.job.dead → import_run marked failed`)})}}}export{b as createImportsRunDeadListener,v as createRunImportHandler,_ as importsRunJob,g as importsRunJobPayloadSchema};
+import{a as e,r as t}from"./read-feed-DsLgRDNR.mjs";import{c as n,d as r,f as i,n as a,s as o}from"./runner-BnFvID1z.mjs";import{a as s,i as c,r as l}from"./publish-DCXIPwTE.mjs";import{runAsCli as u}from"@murumets-ee/core";import{z as d}from"zod";import{and as f,desc as p,eq as m}from"drizzle-orm";import{defineJob as h}from"@murumets-ee/queue/client";const g=d.object({importRunId:d.string().uuid()}),_=h({name:`imports:run`,description:`Stream a feed file, transform rows, bulk-write to Elasticsearch.`,schema:g,defaultRetries:0});function v(l){let{importRuns:d,transforms:h,esClient:g,esIndex:_,resolveFilePath:v,ensureIndex:b,pairSink:x,logger:S}=l;return async l=>u(async()=>{let{importRunId:u}=l.payload,C=S?.child({jobId:l.id,importRunId:u,type:`imports:run`}),w=await d.findById(u);if(!w){C?.warn(`import_run row not found — skipping`);return}w.status===`running`&&C?.warn({previousQueueJobId:w.queueJobId,previousStartedAt:w.startedAt},`imports: re-claiming a row left in running by a prior worker — likely a crash recovered by queue lock-recovery`);let T=h.get(w.transformName);if(!T)throw await y(d,u,`Unknown transform "${w.transformName}"`),Error(`No transform registered for name "${w.transformName}"`);let E=w.params&&typeof w.params==`object`&&!Array.isArray(w.params)?w.params:{},D=t(E);if(!D.success){let e=D.error.issues[0]?.message??`invalid feed reader config`;throw await y(d,u,`Invalid params.feed: ${e}`),Error(`Invalid params.feed for import_run ${u}: ${e}`)}let O=D.data;await d.update(u,{status:`running`,queueJobId:l.id,startedAt:new Date}),c(u);let k;try{k=await g()}catch(e){throw await y(d,u,`Could not resolve ES client: ${e instanceof Error?e.message:String(e)}`),e}if(b)try{await b()}catch(e){throw await y(d,u,`Could not ensure ES index/alias: ${e instanceof Error?e.message:String(e)}`),e}let A,j;if(v)try{let e=await v(w.filePath);A=e.localPath,j=e.cleanup}catch(e){throw await y(d,u,`Could not resolve filePath: ${e instanceof Error?e.message:String(e)}`),e}else A=w.filePath;let M;if(O.format!==`xlsx`)try{M=await e(A,O.hasHeader===void 0?{}:{hasHeader:O.hasHeader}),C?.info({importRunId:u,totalRows:M},`imports: pre-pass row count`)}catch(e){C?.warn({err:e},`imports: pre-pass row count failed — continuing without total`)}let N;try{let e={filePath:A,...O.format!==void 0&&{format:O.format},...O.delimiter!==void 0&&{delimiter:O.delimiter},...O.hasHeader!==void 0&&{hasHeader:O.hasHeader},...O.headerRowIndex!==void 0&&{headerRowIndex:O.headerRowIndex},...O.sheet!==void 0&&{sheet:O.sheet},...O.quote!==void 0&&{quote:O.quote}};N=await a({importRunId:u,runLabel:w.label,params:E,transform:T,feed:e,esClient:k,esIndex:_,onProgress:e=>l.updateProgress(M===void 0?e:{...e,totalRows:M}),...C!==void 0&&{logger:C}})}catch(e){throw await y(d,u,e instanceof Error?e.message:String(e)),j&&await j().catch(e=>{C?.warn({err:e},`filePath cleanup after failed run threw — ignoring`)}),e}j&&await j().catch(e=>{C?.warn({err:e},`filePath cleanup after successful run threw — ignoring`)});let P=N.warnings,F=typeof E.brandId==`string`?E.brandId:void 0,I=typeof E.supplierId==`string`?E.supplierId:void 0;if(F&&I&&N.priceStats.median!==null)try{let e=d.getTable(),t=await d.findMany({where:f(m(e.transformName,w.transformName),m(e.status,`succeeded`)),orderBy:p(e.createdAt),limit:50}),a=t.find(e=>{if(e.id===u)return!1;let t=e.params;if(!t||typeof t!=`object`||Array.isArray(t))return!1;let n=t;return n.brandId===F&&n.supplierId===I});!a&&t.length>=50&&C?.warn({brandId:F,supplierId:I,transformName:w.transformName},`imports: price-drift check found no prior run for this (brand, supplier) within the 50 most recent succeeded runs for this transform — a real prior run may exist beyond the bound, not necessarily absent`);let s=a?n(a.priceStats):null,c=s?o(s,N.priceStats):null;if(c){let e=c.ratio>=1,t=new r;t.addError(-1,`price_anomaly`,`Median purchase price is ${(e?c.ratio:1/c.ratio).toFixed(1)}x ${e?`higher`:`lower`} than the previous successful run for this supplier/brand (previous run ${a?.id??`unknown`}: median ${c.prevMedian}; this run: median ${c.currMedian}, minor units). Check for a decimal or thousands-separator format change in the source feed.`,`priceStats.median`,{previousImportRunId:a?.id??null,previousMedian:c.prevMedian,currentMedian:c.currMedian,ratio:c.ratio}),P=i(N.warnings,t.snapshot()),C?.warn({previousImportRunId:a?.id,ratio:c.ratio},`imports: price_anomaly — median purchase price shifted by a magnitude vs the prior run`)}}catch(e){C?.warn({err:e},`imports: price-drift comparison failed — continuing without it`)}if(await d.update(u,{status:`succeeded`,finishedAt:new Date,totals:{rowsRead:N.rowsRead,rowsSucceeded:N.rowsSucceeded,rowsFailed:N.rowsFailed,rowsSkipped:N.rowsSkipped,rowsWarned:N.rowsWarned,batchesCompleted:N.batchesCompleted,patchesApplied:N.patchesApplied,patchesOrphaned:N.patchesOrphaned},errorSummary:JSON.parse(JSON.stringify(N.errors)),warningSummary:JSON.parse(JSON.stringify(P)),priceStats:JSON.parse(JSON.stringify(N.priceStats))}),s(u,{rowsRead:N.rowsRead,rowsSucceeded:N.rowsSucceeded,rowsFailed:N.rowsFailed}),x&&N.pairs.length>0)try{await x(N.pairs,{importRunId:u,params:E,...C!==void 0&&{logger:C}}),C?.info({importRunId:u,pairs:N.pairs.length},`imports: pairSink invoked`)}catch(e){C?.warn({importRunId:u,pairs:N.pairs.length,err:e},`imports: pairSink threw — replacement ingestion incomplete, catalogue import unaffected`)}C?.info({rowsRead:N.rowsRead,rowsSucceeded:N.rowsSucceeded,rowsFailed:N.rowsFailed,rowsSkipped:N.rowsSkipped,batches:N.batchesCompleted},`import_run completed`)})}async function y(e,t,n){try{await e.update(t,{status:`failed`,finishedAt:new Date,errorSummary:{fatal:n}}),l(t,n)}catch{}}function b(e){let{importRuns:t,logger:n}=e;return{onJobDead:async e=>{e.jobType===`imports:run`&&await u(async()=>{let r=n?.child({event:`imports.run.dead`,jobId:e.jobId,type:`imports:run`}),i=t.getTable(),a=(await t.findMany({where:m(i.queueJobId,e.jobId),limit:1}))[0];if(!a){r?.warn(`queue.job.dead with no matching import_run — skipping`);return}if(a.status!==`running`&&a.status!==`pending`){r?.info({importRunId:a.id,status:a.status},`queue.job.dead arrived for a row already in a terminal state — leaving as-is`);return}a.status===`pending`&&r?.warn({importRunId:a.id,attempts:e.attempts},`queue.job.dead arrived for an import_run still in pending — worker crashed BEFORE claim/lookup`),await y(t,a.id,e.error),r?.info({importRunId:a.id,attempts:e.attempts,priorStatus:a.status},`queue.job.dead → import_run marked failed`)})}}}export{b as createImportsRunDeadListener,v as createRunImportHandler,_ as importsRunJob,g as importsRunJobPayloadSchema};
 //# sourceMappingURL=worker.mjs.map

package/dist/xlsx-reader-D9Ntxm29.mjs

@@ -0,0 +1,2 @@
+import{i as e}from"./read-feed-DsLgRDNR.mjs";import{open as t}from"node:fs/promises";import{Duplex as n}from"node:stream";var r=class extends Error{headerRowNumber;constructor(e){super(`xlsx header row ${e} has no readable columns — the file may be corrupt, or its shared strings failed to load. Refusing rather than importing empty rows.`),this.name=`XlsxEmptyHeaderError`,this.headerRowNumber=e}},i=class extends Error{constructor(e){super(`xlsx path "${e}" is not a regular file`),this.name=`XlsxNotAFileError`}};function a(e){if(e==null)return``;if(typeof e==`string`)return e;if(typeof e==`number`||typeof e==`boolean`)return String(e);if(e instanceof Date)return e.toISOString();if(`richText`in e&&Array.isArray(e.richText))return e.richText.map(e=>e.text).join(``);if(`text`in e&&typeof e.text==`string`)return e.text;if(`result`in e&&e.result!==void 0&&e.result!==null){let t=e.result;return t instanceof Date?t.toISOString():typeof t==`object`?``:String(t)}return``}function o(e){let t=[];if(Array.isArray(e))for(let n=1;n<e.length;n+=1)t[n-1]=a(e[n]);else for(let[n,r]of Object.entries(e)){let e=Number.parseInt(n,10);!Number.isNaN(e)&&e>=1&&(t[e-1]=a(r))}for(let e=0;e<t.length;e+=1)t[e]===void 0&&(t[e]=``);return t}function s(e){let t=e.pipe.bind(e);e.pipe=(e,r)=>{if(e instanceof n){let t=e,n=t.emit.bind(t);t.emit=(e,...r)=>e===`end`&&t.readableLength>0?!1:n(e,...r)}return t(e,r)}}async function*c(n){let{filePath:a,hasHeader:c=!0,headerRowIndex:l=0,sheet:u}=n,d=await import(`exceljs`),f;if(u!==void 0&&(f=Number.parseInt(u,10),Number.isNaN(f)||f<1))throw Error(`Invalid xlsx sheet "${u}" — expected a 1-based worksheet index`);let p=await t(a,`r`),m;try{if(!(await p.stat()).isFile())throw new i(a);m=p.createReadStream({autoClose:!1}),s(m);let t=new d.stream.xlsx.WorkbookReader(m,{worksheets:`emit`,sharedStrings:`cache`,styles:`ignore`,hyperlinks:`ignore`,entries:`ignore`});t.model={sheets:[]};let n=0,u=0,h=!1,g=t[Symbol.asyncIterator](),_=!1;try{for(;;){let t=await g.next();if(t.done){_=!0;break}let i=t.value;if(u+=1,!(f===void 0?u===1:u===f)){for await(let e of i);continue}h=!0;let a=0,s=null;for await(let t of i){a+=1;let i=o(t.values);if(c&&s===null){if(a<l+1)continue;if(s=i.map(e=>e.trim()),s.every(e=>e===``))throw new r(a);e(s);continue}n+=1;let u={};if(s)for(let e=0;e<s.length;e+=1){let t=s[e];t&&(u[t]=i[e]??``)}else for(let e=0;e<i.length;e+=1)u[String(e)]=i[e]??``;yield{rowNumber:n,row:u}}break}}finally{try{for(;!_;){let e=await g.next();if(e.done)break;for await(let t of e.value);}}catch{}}if(f!==void 0&&!h)throw Error(`xlsx worksheet ${f} not found`)}finally{m?.destroy(),await p.close().catch(()=>{})}}export{r as XlsxEmptyHeaderError,i as XlsxNotAFileError,c as streamXlsx};
+//# sourceMappingURL=xlsx-reader-D9Ntxm29.mjs.map

package/dist/xlsx-reader-D9Ntxm29.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"xlsx-reader-D9Ntxm29.mjs","names":[],"sources":["../src/xlsx-reader.ts"],"sourcesContent":["/**\n * Streaming XLSX reader (commerce-demo phase 02, D005) — the exceljs branch of\n * the feed reader. Yields the SAME `AsyncIterable<{ rowNumber, row }>` contract as\n * the csv-parse `streamFeed`, so the runner / transform layer downstream is\n * format-agnostic.\n *\n * **exceljs over SheetJS** (R02): we parse operator-uploaded spreadsheets — the\n * exact threat model of the unfixed npm `xlsx` prototype-pollution CVE\n * (CVE-2023-30533). exceljs is MIT, ships its own types, has a real streaming\n * `WorkbookReader`, and carries no equivalent advisory.\n *\n * **exceljs is `await import()`-ed**, never statically imported, so it stays out\n * of this package's static module graph — heavy + Node-only, it must not be\n * inlinable into any `*-ui` server-pages bundle (CLAUDE.md bundle budgets). The\n * `read-feed` dispatcher additionally only loads THIS module for `format: 'xlsx'`.\n */\n\nimport { type FileHandle, open } from 'node:fs/promises'\nimport { Duplex } from 'node:stream'\nimport type { CellValue } from 'exceljs'\nimport { assertUniqueHeaders, type StreamFeedRow } from './streaming.js'\n\n/**\n * Thrown by {@link streamXlsx} when the configured header row parses to zero\n * readable columns — the signature of the silent-corruption defect documented\n * on the guard below.\n *\n * A distinct CLASS rather than a bare `Error` because the retroactive audit\n * (`auditEmptyHeaderRuns`, #842) has to tell \"this file's header is\n * unreadable\" apart from every other way reading a spreadsheet can fail\n * (missing object, unreadable zip, wrong sheet index). Matching on the message\n * would make that classification a regex over prose that anyone is free to\n * reword — which is exactly the kind of detector that silently stops detecting.\n * The message text is unchanged from the original throw.\n */\nexport class XlsxEmptyHeaderError extends Error {\n  /** 1-based worksheet row the empty header was read from. */\n  readonly headerRowNumber: number\n\n  constructor(headerRowNumber: number) {\n    super(\n      `xlsx header row ${headerRowNumber} has no readable columns — the file may be corrupt, ` +\n        'or its shared strings failed to load. Refusing rather than importing empty rows.',\n    )\n    this.name = 'XlsxEmptyHeaderError'\n    this.headerRowNumber = headerRowNumber\n  }\n}\n\n/**\n * Thrown by {@link streamXlsx} when the path resolves to something that is not a\n * regular file — a directory, most often. `open()` SUCCEEDS on a directory on\n * Linux (POSIX allows `O_RDONLY` on one) and only the read fails, so this is a\n * check the reader has to make itself.\n *\n * Typed for the same reason {@link XlsxEmptyHeaderError} is: the #842 audit has\n * to tell \"the evidence is unreadable\" from \"the evidence is readable and bad\",\n * and matching prose is how a classifier quietly stops classifying.\n */\nexport class XlsxNotAFileError extends Error {\n  constructor(filePath: string) {\n    super(`xlsx path \"${filePath}\" is not a regular file`)\n    this.name = 'XlsxNotAFileError'\n  }\n}\n\nexport interface XlsxReadOptions {\n  /** Path to the .xlsx file on local disk. */\n  filePath: string\n  /** First row is the header (column names). `false` → positional `{ \"0\": … }` keys. Default `true`. */\n  hasHeader?: boolean\n  /** 0-based index of the header row (rows before it are skipped). Default `0`. */\n  headerRowIndex?: number\n  /** Worksheet to read, as a 1-based index string. Omit → the first worksheet. */\n  sheet?: string\n}\n\n/** Convert one exceljs cell value to a plain string (feeds are text/number cells). */\nfunction cellToString(v: CellValue): string {\n  if (v === null || v === undefined) return ''\n  if (typeof v === 'string') return v\n  if (typeof v === 'number' || typeof v === 'boolean') return String(v)\n  if (v instanceof Date) return v.toISOString()\n  // Object cell values: rich text, hyperlink, formula result, error.\n  if ('richText' in v && Array.isArray(v.richText)) {\n    return v.richText.map((r) => r.text).join('')\n  }\n  if ('text' in v && typeof v.text === 'string') return v.text\n  if ('result' in v && v.result !== undefined && v.result !== null) {\n    const r = v.result\n    if (r instanceof Date) return r.toISOString()\n    // A nested error object (formula error) surfaces as empty; primitives stringify.\n    return typeof r === 'object' ? '' : String(r)\n  }\n  return ''\n}\n\n/** Normalise an exceljs `row.values` (1-indexed array OR object form) to a 0-indexed string[]. */\nfunction rowCells(values: CellValue[] | { [key: string]: CellValue }): string[] {\n  const out: string[] = []\n  if (Array.isArray(values)) {\n    // exceljs row arrays are 1-indexed (`values[0]` is empty); shift to 0-based.\n    for (let i = 1; i < values.length; i += 1) out[i - 1] = cellToString(values[i])\n  } else {\n    for (const [k, v] of Object.entries(values)) {\n      const idx = Number.parseInt(k, 10)\n      if (!Number.isNaN(idx) && idx >= 1) out[idx - 1] = cellToString(v)\n    }\n  }\n  // Fill any holes (sparse rows) with empty strings — the contract is \"missing\n  // cell = ''\", matching streamFeed.\n  for (let i = 0; i < out.length; i += 1) if (out[i] === undefined) out[i] = ''\n  return out\n}\n\n/**\n * Stop unzipper's PREMATURE `'end'` from ending exceljs's entry loop while zip\n * entries are still buffered — the cause of the \"clean workbook reads as\n * corrupt\" flake (plan frontend-delivery F067 / F008), and a production hazard\n * for the same reason the `reader.model` seed below is one.\n *\n * The mechanism, read from source and reproduced deterministically:\n *\n *  1. unzipper 0.10's `Parse` (what exceljs pipes our source into,\n *     `workbook-reader.js:81`) re-emits its WRITABLE side's `'finish'` as\n *     `'end'` (`unzipper/lib/parse.js`, constructor) — i.e. \"all input bytes\n *     consumed\", not \"every entry delivered\". Small entries fit their\n *     PassThrough buffers, so `Parse` races through the whole zip, pushing\n *     entry objects into its readable buffer, regardless of the consumer.\n *  2. exceljs's `iterateStream` PAUSES the zip while it handles an entry, and\n *     for a worksheet that precedes `sharedStrings.xml` (exceljs's own write\n *     order, and Excel's) \"handling\" is a `tmp.file()` + write — filesystem\n *     round-trips. It treats the first `'end'` it hears as the end of the zip\n *     and exits with the rest of the entries still sitting in the buffer.\n *  3. So when the temp-file write is slower than inflating the remaining\n *     entries, `xl/sharedStrings.xml` is never parsed. Every string cell then\n *     comes back as `{ sharedString: n }` (exceljs's no-table shape),\n *     `cellToString` renders it `''`, and the header guard in `streamXlsx`\n *     refuses a perfectly good file as `XlsxEmptyHeaderError`.\n *\n * Measured by delaying `tmp.file` inside exceljs: 0-1ms reads correctly,\n * 5ms+ loses the shared strings every time. A loaded CI runner or a busy\n * queue worker supplies those milliseconds; the regression test supplies them\n * on purpose (`xlsx-reader.premature-end.test.ts`).\n *\n * The fix drops an `'end'` only while the zip's readable buffer still holds\n * undelivered entries. By the time `Parse` finishes its writable side it has\n * reached the end-of-central-directory record and called `push(null)`, so\n * Node emits the REAL `'end'` itself once those entries have been read — this\n * delays the event, it never loses one. An `'end'` with an empty buffer passes\n * straight through, so a read that was already correct is untouched.\n *\n * Not a `pnpm patch` of exceljs/unzipper: a patch lives in THIS workspace's\n * install only, and `@murumets-ee/imports` is published — every consumer\n * would keep the defect. If a future exceljs stops piping into a `Duplex`,\n * the hook simply does not apply and the regression test goes red.\n */\nfunction holdPrematureZipEnd(source: ReturnType<FileHandle['createReadStream']>): void {\n  const pipe = source.pipe.bind(source)\n  source.pipe = <T extends NodeJS.WritableStream>(\n    destination: T,\n    options?: { end?: boolean },\n  ): T => {\n    if (destination instanceof Duplex) {\n      const zip = destination\n      const emit = zip.emit.bind(zip)\n      zip.emit = (event: string | symbol, ...args: unknown[]): boolean => {\n        if (event === 'end' && zip.readableLength > 0) return false\n        return emit(event, ...args)\n      }\n    }\n    return pipe(destination, options)\n  }\n}\n\n/**\n * Async-iterable over the data rows of an .xlsx file. `rowNumber` is the 1-based\n * DATA row index (header excluded). Cell values are coerced to strings so the\n * mapping engine's per-column transforms (which expect raw string cells) apply\n * uniformly across csv/tsv/xlsx.\n */\nexport async function* streamXlsx(options: XlsxReadOptions): AsyncIterable<StreamFeedRow> {\n  const { filePath, hasHeader = true, headerRowIndex = 0, sheet } = options\n  const ExcelJS = await import('exceljs')\n\n  // `WorksheetReader` doesn't expose its sheet id, so count emitted worksheets\n  // (1-based) to match the configured `sheet`. Default: the first worksheet.\n  // A provided-but-invalid sheet FAILS FAST rather than silently reading sheet 1\n  // — and before we touch the filesystem, so a config typo costs no I/O.\n  let targetSheetIndex: number | undefined\n  if (sheet !== undefined) {\n    targetSheetIndex = Number.parseInt(sheet, 10)\n    if (Number.isNaN(targetSheetIndex) || targetSheetIndex < 1) {\n      throw new Error(`Invalid xlsx sheet \"${sheet}\" — expected a 1-based worksheet index`)\n    }\n  }\n\n  // OPEN THE FILE OURSELVES rather than handing exceljs a path, because a path\n  // makes every filesystem failure an UNCAUGHT EXCEPTION that kills the process\n  // instead of a rejection the caller can handle.\n  //\n  // `_getStream` does `fs.createReadStream(input)` for a string input and then\n  // `stream.pipe(zip)` (`workbook-reader.js:38,81`). `pipe()` does not forward\n  // SOURCE errors, and exceljs registers no `error` listener on that stream — so\n  // an ENOENT / EACCES / EISDIR surfaces as `Unhandled 'error' event`. Verified\n  // by execution, not inference: all three crash the process outright. In the\n  // queue worker that kills the job mid-flight instead of routing through\n  // `markFailed`; in the #842 audit one unreadable file discards the whole sweep.\n  //\n  // An fd fixes it at the root rather than patching the symptom:\n  //   - `open()` rejects normally on ENOENT and EACCES.\n  //   - a DIRECTORY still opens on Linux (checked — O_RDONLY on a dir succeeds\n  //     and only the read gives EISDIR), so the `isFile()` check below is\n  //     load-bearing. It runs against the fd, so it cannot be raced.\n  //   - once the fd is held, the inode survives `unlink` (checked: the read\n  //     still returns the bytes). That closes the window against the\n  //     `@every 1h` retention sweep, which deletes exactly these files and which\n  //     the audit is documented as racing.\n  const handle = await open(filePath, 'r')\n  // Held so the outer `finally` can destroy it BEFORE the fd goes away. With\n  // `autoClose: false` nothing tears this stream down except reaching the end,\n  // and the drain below deliberately swallows errors — so a malformed trailing\n  // worksheet can leave the source mid-read while `handle.close()` invalidates\n  // the descriptor underneath it. exceljs registers no `error` listener on the\n  // stream it pipes (the whole reason this function opens the file itself), so\n  // that EBADF would arrive as an uncaught event: the exact failure mode the fd\n  // was introduced to eliminate, re-entering through the back door.\n  let source: ReturnType<typeof handle.createReadStream> | undefined\n  try {\n    const stats = await handle.stat()\n    if (!stats.isFile()) {\n      throw new XlsxNotAFileError(filePath)\n    }\n\n    source = handle.createReadStream({ autoClose: false })\n    holdPrematureZipEnd(source)\n    const reader = new ExcelJS.stream.xlsx.WorkbookReader(source, {\n      worksheets: 'emit',\n      sharedStrings: 'cache',\n      styles: 'ignore',\n      hyperlinks: 'ignore',\n      entries: 'ignore',\n    })\n\n    // Seed `model` so exceljs cannot dereference it before it has parsed\n    // `xl/workbook.xml` — an UPSTREAM defect, and the cause of the xlsx-reader\n    // \"flake\" that failed CI intermittently for months (#619, #835).\n    //\n    // Its streaming reader parses a worksheet INLINE, rather than deferring it to\n    // a temp file, when `this.sharedStrings && this.workbookRels` are both set as\n    // the entry arrives (`workbook-reader.js:109`). Neither implies\n    // `xl/workbook.xml` has been read. The inline path then does\n    // `matchingRel && (this.model.sheets || []).find(…)` (`:303`) — guarding\n    // `this.workbookRels` on the line above and `this.model.sheets` not at all —\n    // so any workbook whose zip orders `sharedStrings.xml` before the worksheet,\n    // with `workbook.xml` after it, throws `Cannot read properties of undefined\n    // (reading 'sheets')`.\n    //\n    // That is a PRODUCTION hazard, not a test artifact: this function reads\n    // operator-uploaded spreadsheets and the writer chooses the entry order. It\n    // presented as flakiness only because the fixtures our own tests write happen\n    // to sit on the safe side of the condition.\n    //\n    // An empty `sheets` makes the lookup miss, so the worksheet keeps its default\n    // id/name/state until `_parseWorkbook` overwrites `model` with the real one a\n    // few entries later — which is exactly what the deferred path already does.\n    // Preferred over a `pnpm patch` (exceljs 4.4.0 is the latest release, so there\n    // is no upstream fix to take, but a patch is a standing maintenance cost and\n    // this is one assignment that stays correct even if upstream fixes it).\n    //\n    // `sheets: []` rather than `{}` because `(this.model.sheets || [])` is what is\n    // read.\n    //\n    // The double assertion is NOT laziness and a single `as typeof reader.model`\n    // does not compile — checked, not assumed:\n    //\n    //   TS2352: Conversion of type '{ sheets: never[]; }' to type 'WorkbookModel'\n    //   may be a mistake because neither type sufficiently overlaps with the\n    //   other. … missing the following properties: creator, lastModifiedBy,\n    //   lastPrinted, created, and 17 more.\n    //\n    // A deliberately PARTIAL seed is the whole point — exceljs overwrites it with\n    // the real model a few entries later, and filling 21 fields with invented\n    // values to satisfy the compiler would be worse than the assertion: it would\n    // put fabricated workbook metadata one `_parseWorkbook` failure away from\n    // being read as real.\n    // toolkit-smell-allow: partial seed for a 21-field type; single assertion rejected by TS2352 (see above)\n    reader.model = { sheets: [] } as unknown as typeof reader.model\n\n    let dataRowNumber = 0\n    let sheetIndex = 0\n    let matchedTarget = false\n\n    // Drive the worksheet iterator BY HAND rather than with `for await`, so the\n    // `finally` below can resume it. Abandoning exceljs's `parse()` generator —\n    // which `return`/`break` out of a `for await` does — LEAKS a temp file and an\n    // fd per deferred worksheet, because `tempFileCleanupCallback()` sits after\n    // the `yield*` with no `finally` of its own (`workbook-reader.js:140-149`).\n    //\n    // Measured, with a negative control: 25 early-returning reads of a two-sheet\n    // deferred-order workbook leaked 50 fds and 50 temp files; the draining shape\n    // below leaks 0 of each. `tmp`'s graceful-cleanup hook reclaims the files at\n    // NORMAL process exit only, which is no help to a long-lived queue worker and\n    // no help at all against the fd ceiling: at the #842 audit's 5000-run cap\n    // this is an EMFILE, and an EMFILE arrives as the uncaught stream error the\n    // `open()` above exists to prevent.\n    //\n    // THE COST IS PROPORTIONAL TO THE WORKSHEETS AFTER THE TARGET, which is zero\n    // for the single-sheet feeds this reader mostly sees and 4.7× on a workbook\n    // whose target is sheet 1 and whose sheet 2 holds 40k rows. Advancing the\n    // parent iterator WITHOUT consuming each trailing worksheet is far cheaper\n    // and does release the temp files — but it strands exceljs's\n    // `createReadStream` on each one, i.e. it trades the disk leak for the fd\n    // leak, and the fd leak is the one that ends in EMFILE. So: full drain.\n    const worksheets = reader[Symbol.asyncIterator]()\n    let readerExhausted = false\n    try {\n      while (true) {\n        const next = await worksheets.next()\n        if (next.done) {\n          readerExhausted = true\n          break\n        }\n        const worksheet = next.value\n        sheetIndex += 1\n        const isTarget =\n          targetSheetIndex === undefined ? sheetIndex === 1 : sheetIndex === targetSheetIndex\n\n        if (!isTarget) {\n          // Drain a non-target worksheet so the underlying zip stream advances to the\n          // next; exceljs's streaming reader stalls if a worksheet is left unconsumed.\n          for await (const _skip of worksheet) {\n            void _skip\n          }\n          continue\n        }\n        matchedTarget = true\n\n        let excelRowNum = 0\n        let header: string[] | null = null\n        for await (const row of worksheet) {\n          excelRowNum += 1\n          const cells = rowCells(row.values)\n\n          if (hasHeader && header === null) {\n            // Skip rows before the configured header row, then capture it (trimmed).\n            if (excelRowNum < headerRowIndex + 1) continue\n            header = cells.map((c) => c.trim())\n            // FAIL LOUDLY on an empty header, because the alternative is SILENT DATA\n            // CORRUPTION and we have watched it happen.\n            //\n            // exceljs assigns `this.sharedStrings = []` at the TOP of\n            // `_parseSharedStrings`, before parsing a single entry — and an empty\n            // array is truthy, which is what the inline-worksheet guard tests. So a\n            // shared-strings entry that yields nothing (a truncated or starved read;\n            // observed in CI under load) leaves every string cell resolving to\n            // `sharedStrings[index]` → `undefined`. The header parses to zero\n            // columns, every data row maps to `{}`, and the import reports SUCCESS\n            // with every value dropped. Reproduced deterministically by rebuilding a\n            // workbook with an empty `sst`: two rows read, no values, no error.\n            //\n            // A header row with no usable column is never a legitimate import, so\n            // there is no case to preserve. Refusing costs an operator a clear error\n            // and saves them a silently empty dataset they would have to notice\n            // themselves — the same posture the rest of this package takes toward\n            // config mistakes.\n            if (header.every((c) => c === '')) {\n              throw new XlsxEmptyHeaderError(excelRowNum)\n            }\n            assertUniqueHeaders(header)\n            continue\n          }\n\n          dataRowNumber += 1\n          const rec: Record<string, string> = {}\n          if (header) {\n            for (let i = 0; i < header.length; i += 1) {\n              const key = header[i]\n              if (key) rec[key] = cells[i] ?? ''\n            }\n          } else {\n            for (let i = 0; i < cells.length; i += 1) rec[String(i)] = cells[i] ?? ''\n          }\n          yield { rowNumber: dataRowNumber, row: rec }\n        }\n\n        // Target worksheet fully read. Stop yielding — the `finally` walks the\n        // remaining worksheets so exceljs reaches its cleanup.\n        break\n      }\n    } finally {\n      // Runs on every exit: target read, an empty-header throw, or the CONSUMER\n      // abandoning us mid-stream (a `break` in the caller's `for await` calls\n      // `.return()`, which lands here). Whatever the reason, exceljs still owns\n      // temp files it only releases as the consumer pulls past each worksheet.\n      //\n      // The whole body is CONTAINED, for the same reason `handle.close()` below\n      // is: a throw from a `finally` REPLACES whatever the try block was\n      // propagating. Trailing worksheets are parsed here, so a malformed one\n      // would overwrite an `XlsxEmptyHeaderError` with an XML error — turning\n      // the #842 audit's `corrupt` verdict into a meaningless `error` and\n      // failing an import whose target sheet read perfectly. Measured: two of\n      // four malformed-trailing-sheet variants masked the refusal before this\n      // catch existed. Cleanup is best-effort by definition; the error the\n      // caller came for is not.\n      try {\n        while (!readerExhausted) {\n          const rest = await worksheets.next()\n          if (rest.done) break\n          for await (const _skip of rest.value) {\n            void _skip\n          }\n        }\n      } catch {\n        // Deliberately swallowed — see above. A trailing worksheet we only read\n        // to satisfy exceljs's cleanup contract cannot be allowed to speak for\n        // the read the caller actually asked for. Worst case we leak the temp\n        // files this drain existed to release, which is what `main` did on\n        // EVERY read.\n      }\n    }\n\n    // Reached only when a specific sheet was requested but never matched — the\n    // first-sheet default always matches on a non-empty workbook.\n    if (targetSheetIndex !== undefined && !matchedTarget) {\n      throw new Error(`xlsx worksheet ${targetSheetIndex} not found`)\n    }\n  } finally {\n    // Order matters: destroy the reader of the fd before closing the fd.\n    source?.destroy()\n    await handle.close().catch(() => {\n      // Swallowed deliberately, and biome's `noUnsafeFinally` agrees: a throw\n      // here would REPLACE whatever the try block was propagating, and what it\n      // propagates is the point of this reader — an empty-header refusal has to\n      // reach the caller intact. Closing a read-only fd we are finished with can\n      // fail only in ways that are not actionable, and the fd is reclaimed at\n      // process exit regardless.\n    })\n  }\n}\n"],"mappings":"0HAmCA,IAAa,EAAb,cAA0C,KAAM,CAE9C,gBAEA,YAAY,EAAyB,CACnC,MACE,mBAAmB,EAAgB,qIAErC,EACA,KAAK,KAAO,uBACZ,KAAK,gBAAkB,CACzB,CACF,EAYa,EAAb,cAAuC,KAAM,CAC3C,YAAY,EAAkB,CAC5B,MAAM,cAAc,EAAS,wBAAwB,EACrD,KAAK,KAAO,mBACd,CACF,EAcA,SAAS,EAAa,EAAsB,CAC1C,GAAI,GAAM,KAAyB,MAAO,GAC1C,GAAI,OAAO,GAAM,SAAU,OAAO,EAClC,GAAI,OAAO,GAAM,UAAY,OAAO,GAAM,UAAW,OAAO,OAAO,CAAC,EACpE,GAAI,aAAa,KAAM,OAAO,EAAE,YAAY,EAE5C,GAAI,aAAc,GAAK,MAAM,QAAQ,EAAE,QAAQ,EAC7C,OAAO,EAAE,SAAS,IAAK,GAAM,EAAE,IAAI,CAAC,CAAC,KAAK,EAAE,EAE9C,GAAI,SAAU,GAAK,OAAO,EAAE,MAAS,SAAU,OAAO,EAAE,KACxD,GAAI,WAAY,GAAK,EAAE,SAAW,IAAA,IAAa,EAAE,SAAW,KAAM,CAChE,IAAM,EAAI,EAAE,OAGZ,OAFI,aAAa,KAAa,EAAE,YAAY,EAErC,OAAO,GAAM,SAAW,GAAK,OAAO,CAAC,CAC9C,CACA,MAAO,EACT,CAGA,SAAS,EAAS,EAA8D,CAC9E,IAAM,EAAgB,CAAC,EACvB,GAAI,MAAM,QAAQ,CAAM,EAEtB,IAAK,IAAI,EAAI,EAAG,EAAI,EAAO,OAAQ,GAAK,EAAG,EAAI,EAAI,GAAK,EAAa,EAAO,EAAE,OAE9E,IAAK,GAAM,CAAC,EAAG,KAAM,OAAO,QAAQ,CAAM,EAAG,CAC3C,IAAM,EAAM,OAAO,SAAS,EAAG,EAAE,EAC7B,CAAC,OAAO,MAAM,CAAG,GAAK,GAAO,IAAG,EAAI,EAAM,GAAK,EAAa,CAAC,EACnE,CAIF,IAAK,IAAI,EAAI,EAAG,EAAI,EAAI,OAAQ,GAAK,EAAO,EAAI,KAAO,IAAA,KAAW,EAAI,GAAK,IAC3E,OAAO,CACT,CA4CA,SAAS,EAAoB,EAA0D,CACrF,IAAM,EAAO,EAAO,KAAK,KAAK,CAAM,EACpC,EAAO,MACL,EACA,IACM,CACN,GAAI,aAAuB,EAAQ,CACjC,IAAM,EAAM,EACN,EAAO,EAAI,KAAK,KAAK,CAAG,EAC9B,EAAI,MAAQ,EAAwB,GAAG,IACjC,IAAU,OAAS,EAAI,eAAiB,EAAU,GAC/C,EAAK,EAAO,GAAG,CAAI,CAE9B,CACA,OAAO,EAAK,EAAa,CAAO,CAClC,CACF,CAQA,eAAuB,EAAW,EAAwD,CACxF,GAAM,CAAE,WAAU,YAAY,GAAM,iBAAiB,EAAG,SAAU,EAC5D,EAAU,MAAM,OAAO,WAMzB,EACJ,GAAI,IAAU,IAAA,KACZ,EAAmB,OAAO,SAAS,EAAO,EAAE,EACxC,OAAO,MAAM,CAAgB,GAAK,EAAmB,GACvD,MAAU,MAAM,uBAAuB,EAAM,uCAAuC,EAyBxF,IAAM,EAAS,MAAM,EAAK,EAAU,GAAG,EASnC,EACJ,GAAI,CAEF,GAAI,EAAC,MADe,EAAO,KAAK,EAAA,CACrB,OAAO,EAChB,MAAM,IAAI,EAAkB,CAAQ,EAGtC,EAAS,EAAO,iBAAiB,CAAE,UAAW,EAAM,CAAC,EACrD,EAAoB,CAAM,EAC1B,IAAM,EAAS,IAAI,EAAQ,OAAO,KAAK,eAAe,EAAQ,CAC5D,WAAY,OACZ,cAAe,QACf,OAAQ,SACR,WAAY,SACZ,QAAS,QACX,CAAC,EA6CD,EAAO,MAAQ,CAAE,OAAQ,CAAC,CAAE,EAE5B,IAAI,EAAgB,EAChB,EAAa,EACb,EAAgB,GAuBd,EAAa,EAAO,OAAO,cAAc,CAAC,EAC5C,EAAkB,GACtB,GAAI,CACF,OAAa,CACX,IAAM,EAAO,MAAM,EAAW,KAAK,EACnC,GAAI,EAAK,KAAM,CACb,EAAkB,GAClB,KACF,CACA,IAAM,EAAY,EAAK,MAKvB,GAJA,GAAc,EAIV,EAFF,IAAqB,IAAA,GAAY,IAAe,EAAI,IAAe,GAEtD,CAGb,UAAW,IAAM,KAAS,GAG1B,QACF,CACA,EAAgB,GAEhB,IAAI,EAAc,EACd,EAA0B,KAC9B,UAAW,IAAM,KAAO,EAAW,CACjC,GAAe,EACf,IAAM,EAAQ,EAAS,EAAI,MAAM,EAEjC,GAAI,GAAa,IAAW,KAAM,CAEhC,GAAI,EAAc,EAAiB,EAAG,SAoBtC,GAnBA,EAAS,EAAM,IAAK,GAAM,EAAE,KAAK,CAAC,EAmB9B,EAAO,MAAO,GAAM,IAAM,EAAE,EAC9B,MAAM,IAAI,EAAqB,CAAW,EAE5C,EAAoB,CAAM,EAC1B,QACF,CAEA,GAAiB,EACjB,IAAM,EAA8B,CAAC,EACrC,GAAI,EACF,IAAK,IAAI,EAAI,EAAG,EAAI,EAAO,OAAQ,GAAK,EAAG,CACzC,IAAM,EAAM,EAAO,GACf,IAAK,EAAI,GAAO,EAAM,IAAM,GAClC,MAEA,IAAK,IAAI,EAAI,EAAG,EAAI,EAAM,OAAQ,GAAK,EAAG,EAAI,OAAO,CAAC,GAAK,EAAM,IAAM,GAEzE,KAAM,CAAE,UAAW,EAAe,IAAK,CAAI,CAC7C,CAIA,KACF,CACF,QAAU,CAeR,GAAI,CACF,KAAO,CAAC,GAAiB,CACvB,IAAM,EAAO,MAAM,EAAW,KAAK,EACnC,GAAI,EAAK,KAAM,MACf,UAAW,IAAM,KAAS,EAAK,OAGjC,CACF,MAAQ,CAMR,CACF,CAIA,GAAI,IAAqB,IAAA,IAAa,CAAC,EACrC,MAAU,MAAM,kBAAkB,EAAiB,WAAW,CAElE,QAAU,CAER,GAAQ,QAAQ,EAChB,MAAM,EAAO,MAAM,CAAC,CAAC,UAAY,CAOjC,CAAC,CACH,CACF"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@murumets-ee/imports",
-  "version": "0.83.0",
+  "version": "0.84.0",
   "license": "Elastic-2.0",
   "type": "module",
   "exports": {
@@ -53,13 +53,13 @@
     "drizzle-orm": "^0.45.2",
     "exceljs": "^4.4.0",
     "zod": "^3.24.1",
-    "@murumets-ee/core": "0.83.0",
-    "@murumets-ee/db": "0.83.0",
-    "@murumets-ee/entity": "0.83.0",
-    "@murumets-ee/logging": "0.83.0",
-    "@murumets-ee/queue": "0.83.0",
-    "@murumets-ee/search-elasticsearch": "0.83.0",
-    "@murumets-ee/storage": "0.83.0"
+    "@murumets-ee/core": "0.84.0",
+    "@murumets-ee/logging": "0.84.0",
+    "@murumets-ee/db": "0.84.0",
+    "@murumets-ee/entity": "0.84.0",
+    "@murumets-ee/queue": "0.84.0",
+    "@murumets-ee/search-elasticsearch": "0.84.0",
+    "@murumets-ee/storage": "0.84.0"
   },
   "devDependencies": {
     "@types/node": "^20.19.42",

package/dist/xlsx-reader-_ioPAqm5.mjs

@@ -1,2 +0,0 @@
-import{i as e}from"./read-feed-DGjqejw3.mjs";import{open as t}from"node:fs/promises";var n=class extends Error{headerRowNumber;constructor(e){super(`xlsx header row ${e} has no readable columns — the file may be corrupt, or its shared strings failed to load. Refusing rather than importing empty rows.`),this.name=`XlsxEmptyHeaderError`,this.headerRowNumber=e}},r=class extends Error{constructor(e){super(`xlsx path "${e}" is not a regular file`),this.name=`XlsxNotAFileError`}};function i(e){if(e==null)return``;if(typeof e==`string`)return e;if(typeof e==`number`||typeof e==`boolean`)return String(e);if(e instanceof Date)return e.toISOString();if(`richText`in e&&Array.isArray(e.richText))return e.richText.map(e=>e.text).join(``);if(`text`in e&&typeof e.text==`string`)return e.text;if(`result`in e&&e.result!==void 0&&e.result!==null){let t=e.result;return t instanceof Date?t.toISOString():typeof t==`object`?``:String(t)}return``}function a(e){let t=[];if(Array.isArray(e))for(let n=1;n<e.length;n+=1)t[n-1]=i(e[n]);else for(let[n,r]of Object.entries(e)){let e=Number.parseInt(n,10);!Number.isNaN(e)&&e>=1&&(t[e-1]=i(r))}for(let e=0;e<t.length;e+=1)t[e]===void 0&&(t[e]=``);return t}async function*o(i){let{filePath:o,hasHeader:s=!0,headerRowIndex:c=0,sheet:l}=i,u=await import(`exceljs`),d;if(l!==void 0&&(d=Number.parseInt(l,10),Number.isNaN(d)||d<1))throw Error(`Invalid xlsx sheet "${l}" — expected a 1-based worksheet index`);let f=await t(o,`r`),p;try{if(!(await f.stat()).isFile())throw new r(o);p=f.createReadStream({autoClose:!1});let t=new u.stream.xlsx.WorkbookReader(p,{worksheets:`emit`,sharedStrings:`cache`,styles:`ignore`,hyperlinks:`ignore`,entries:`ignore`});t.model={sheets:[]};let i=0,l=0,m=!1,h=t[Symbol.asyncIterator](),g=!1;try{for(;;){let t=await h.next();if(t.done){g=!0;break}let r=t.value;if(l+=1,!(d===void 0?l===1:l===d)){for await(let e of r);continue}m=!0;let o=0,u=null;for await(let t of r){o+=1;let r=a(t.values);if(s&&u===null){if(o<c+1)continue;if(u=r.map(e=>e.trim()),u.every(e=>e===``))throw new n(o);e(u);continue}i+=1;let l={};if(u)for(let e=0;e<u.length;e+=1){let t=u[e];t&&(l[t]=r[e]??``)}else for(let e=0;e<r.length;e+=1)l[String(e)]=r[e]??``;yield{rowNumber:i,row:l}}break}}finally{try{for(;!g;){let e=await h.next();if(e.done)break;for await(let t of e.value);}}catch{}}if(d!==void 0&&!m)throw Error(`xlsx worksheet ${d} not found`)}finally{p?.destroy(),await f.close().catch(()=>{})}}export{n as XlsxEmptyHeaderError,r as XlsxNotAFileError,o as streamXlsx};
-//# sourceMappingURL=xlsx-reader-_ioPAqm5.mjs.map

package/dist/xlsx-reader-_ioPAqm5.mjs.map

@@ -1 +0,0 @@
-{"version":3,"file":"xlsx-reader-_ioPAqm5.mjs","names":[],"sources":["../src/xlsx-reader.ts"],"sourcesContent":["/**\n * Streaming XLSX reader (commerce-demo phase 02, D005) — the exceljs branch of\n * the feed reader. Yields the SAME `AsyncIterable<{ rowNumber, row }>` contract as\n * the csv-parse `streamFeed`, so the runner / transform layer downstream is\n * format-agnostic.\n *\n * **exceljs over SheetJS** (R02): we parse operator-uploaded spreadsheets — the\n * exact threat model of the unfixed npm `xlsx` prototype-pollution CVE\n * (CVE-2023-30533). exceljs is MIT, ships its own types, has a real streaming\n * `WorkbookReader`, and carries no equivalent advisory.\n *\n * **exceljs is `await import()`-ed**, never statically imported, so it stays out\n * of this package's static module graph — heavy + Node-only, it must not be\n * inlinable into any `*-ui` server-pages bundle (CLAUDE.md bundle budgets). The\n * `read-feed` dispatcher additionally only loads THIS module for `format: 'xlsx'`.\n */\n\nimport { open } from 'node:fs/promises'\nimport type { CellValue } from 'exceljs'\nimport { assertUniqueHeaders, type StreamFeedRow } from './streaming.js'\n\n/**\n * Thrown by {@link streamXlsx} when the configured header row parses to zero\n * readable columns — the signature of the silent-corruption defect documented\n * on the guard below.\n *\n * A distinct CLASS rather than a bare `Error` because the retroactive audit\n * (`auditEmptyHeaderRuns`, #842) has to tell \"this file's header is\n * unreadable\" apart from every other way reading a spreadsheet can fail\n * (missing object, unreadable zip, wrong sheet index). Matching on the message\n * would make that classification a regex over prose that anyone is free to\n * reword — which is exactly the kind of detector that silently stops detecting.\n * The message text is unchanged from the original throw.\n */\nexport class XlsxEmptyHeaderError extends Error {\n  /** 1-based worksheet row the empty header was read from. */\n  readonly headerRowNumber: number\n\n  constructor(headerRowNumber: number) {\n    super(\n      `xlsx header row ${headerRowNumber} has no readable columns — the file may be corrupt, ` +\n        'or its shared strings failed to load. Refusing rather than importing empty rows.',\n    )\n    this.name = 'XlsxEmptyHeaderError'\n    this.headerRowNumber = headerRowNumber\n  }\n}\n\n/**\n * Thrown by {@link streamXlsx} when the path resolves to something that is not a\n * regular file — a directory, most often. `open()` SUCCEEDS on a directory on\n * Linux (POSIX allows `O_RDONLY` on one) and only the read fails, so this is a\n * check the reader has to make itself.\n *\n * Typed for the same reason {@link XlsxEmptyHeaderError} is: the #842 audit has\n * to tell \"the evidence is unreadable\" from \"the evidence is readable and bad\",\n * and matching prose is how a classifier quietly stops classifying.\n */\nexport class XlsxNotAFileError extends Error {\n  constructor(filePath: string) {\n    super(`xlsx path \"${filePath}\" is not a regular file`)\n    this.name = 'XlsxNotAFileError'\n  }\n}\n\nexport interface XlsxReadOptions {\n  /** Path to the .xlsx file on local disk. */\n  filePath: string\n  /** First row is the header (column names). `false` → positional `{ \"0\": … }` keys. Default `true`. */\n  hasHeader?: boolean\n  /** 0-based index of the header row (rows before it are skipped). Default `0`. */\n  headerRowIndex?: number\n  /** Worksheet to read, as a 1-based index string. Omit → the first worksheet. */\n  sheet?: string\n}\n\n/** Convert one exceljs cell value to a plain string (feeds are text/number cells). */\nfunction cellToString(v: CellValue): string {\n  if (v === null || v === undefined) return ''\n  if (typeof v === 'string') return v\n  if (typeof v === 'number' || typeof v === 'boolean') return String(v)\n  if (v instanceof Date) return v.toISOString()\n  // Object cell values: rich text, hyperlink, formula result, error.\n  if ('richText' in v && Array.isArray(v.richText)) {\n    return v.richText.map((r) => r.text).join('')\n  }\n  if ('text' in v && typeof v.text === 'string') return v.text\n  if ('result' in v && v.result !== undefined && v.result !== null) {\n    const r = v.result\n    if (r instanceof Date) return r.toISOString()\n    // A nested error object (formula error) surfaces as empty; primitives stringify.\n    return typeof r === 'object' ? '' : String(r)\n  }\n  return ''\n}\n\n/** Normalise an exceljs `row.values` (1-indexed array OR object form) to a 0-indexed string[]. */\nfunction rowCells(values: CellValue[] | { [key: string]: CellValue }): string[] {\n  const out: string[] = []\n  if (Array.isArray(values)) {\n    // exceljs row arrays are 1-indexed (`values[0]` is empty); shift to 0-based.\n    for (let i = 1; i < values.length; i += 1) out[i - 1] = cellToString(values[i])\n  } else {\n    for (const [k, v] of Object.entries(values)) {\n      const idx = Number.parseInt(k, 10)\n      if (!Number.isNaN(idx) && idx >= 1) out[idx - 1] = cellToString(v)\n    }\n  }\n  // Fill any holes (sparse rows) with empty strings — the contract is \"missing\n  // cell = ''\", matching streamFeed.\n  for (let i = 0; i < out.length; i += 1) if (out[i] === undefined) out[i] = ''\n  return out\n}\n\n/**\n * Async-iterable over the data rows of an .xlsx file. `rowNumber` is the 1-based\n * DATA row index (header excluded). Cell values are coerced to strings so the\n * mapping engine's per-column transforms (which expect raw string cells) apply\n * uniformly across csv/tsv/xlsx.\n */\nexport async function* streamXlsx(options: XlsxReadOptions): AsyncIterable<StreamFeedRow> {\n  const { filePath, hasHeader = true, headerRowIndex = 0, sheet } = options\n  const ExcelJS = await import('exceljs')\n\n  // `WorksheetReader` doesn't expose its sheet id, so count emitted worksheets\n  // (1-based) to match the configured `sheet`. Default: the first worksheet.\n  // A provided-but-invalid sheet FAILS FAST rather than silently reading sheet 1\n  // — and before we touch the filesystem, so a config typo costs no I/O.\n  let targetSheetIndex: number | undefined\n  if (sheet !== undefined) {\n    targetSheetIndex = Number.parseInt(sheet, 10)\n    if (Number.isNaN(targetSheetIndex) || targetSheetIndex < 1) {\n      throw new Error(`Invalid xlsx sheet \"${sheet}\" — expected a 1-based worksheet index`)\n    }\n  }\n\n  // OPEN THE FILE OURSELVES rather than handing exceljs a path, because a path\n  // makes every filesystem failure an UNCAUGHT EXCEPTION that kills the process\n  // instead of a rejection the caller can handle.\n  //\n  // `_getStream` does `fs.createReadStream(input)` for a string input and then\n  // `stream.pipe(zip)` (`workbook-reader.js:38,81`). `pipe()` does not forward\n  // SOURCE errors, and exceljs registers no `error` listener on that stream — so\n  // an ENOENT / EACCES / EISDIR surfaces as `Unhandled 'error' event`. Verified\n  // by execution, not inference: all three crash the process outright. In the\n  // queue worker that kills the job mid-flight instead of routing through\n  // `markFailed`; in the #842 audit one unreadable file discards the whole sweep.\n  //\n  // An fd fixes it at the root rather than patching the symptom:\n  //   - `open()` rejects normally on ENOENT and EACCES.\n  //   - a DIRECTORY still opens on Linux (checked — O_RDONLY on a dir succeeds\n  //     and only the read gives EISDIR), so the `isFile()` check below is\n  //     load-bearing. It runs against the fd, so it cannot be raced.\n  //   - once the fd is held, the inode survives `unlink` (checked: the read\n  //     still returns the bytes). That closes the window against the\n  //     `@every 1h` retention sweep, which deletes exactly these files and which\n  //     the audit is documented as racing.\n  const handle = await open(filePath, 'r')\n  // Held so the outer `finally` can destroy it BEFORE the fd goes away. With\n  // `autoClose: false` nothing tears this stream down except reaching the end,\n  // and the drain below deliberately swallows errors — so a malformed trailing\n  // worksheet can leave the source mid-read while `handle.close()` invalidates\n  // the descriptor underneath it. exceljs registers no `error` listener on the\n  // stream it pipes (the whole reason this function opens the file itself), so\n  // that EBADF would arrive as an uncaught event: the exact failure mode the fd\n  // was introduced to eliminate, re-entering through the back door.\n  let source: ReturnType<typeof handle.createReadStream> | undefined\n  try {\n    const stats = await handle.stat()\n    if (!stats.isFile()) {\n      throw new XlsxNotAFileError(filePath)\n    }\n\n    source = handle.createReadStream({ autoClose: false })\n    const reader = new ExcelJS.stream.xlsx.WorkbookReader(source, {\n      worksheets: 'emit',\n      sharedStrings: 'cache',\n      styles: 'ignore',\n      hyperlinks: 'ignore',\n      entries: 'ignore',\n    })\n\n    // Seed `model` so exceljs cannot dereference it before it has parsed\n    // `xl/workbook.xml` — an UPSTREAM defect, and the cause of the xlsx-reader\n    // \"flake\" that failed CI intermittently for months (#619, #835).\n    //\n    // Its streaming reader parses a worksheet INLINE, rather than deferring it to\n    // a temp file, when `this.sharedStrings && this.workbookRels` are both set as\n    // the entry arrives (`workbook-reader.js:109`). Neither implies\n    // `xl/workbook.xml` has been read. The inline path then does\n    // `matchingRel && (this.model.sheets || []).find(…)` (`:303`) — guarding\n    // `this.workbookRels` on the line above and `this.model.sheets` not at all —\n    // so any workbook whose zip orders `sharedStrings.xml` before the worksheet,\n    // with `workbook.xml` after it, throws `Cannot read properties of undefined\n    // (reading 'sheets')`.\n    //\n    // That is a PRODUCTION hazard, not a test artifact: this function reads\n    // operator-uploaded spreadsheets and the writer chooses the entry order. It\n    // presented as flakiness only because the fixtures our own tests write happen\n    // to sit on the safe side of the condition.\n    //\n    // An empty `sheets` makes the lookup miss, so the worksheet keeps its default\n    // id/name/state until `_parseWorkbook` overwrites `model` with the real one a\n    // few entries later — which is exactly what the deferred path already does.\n    // Preferred over a `pnpm patch` (exceljs 4.4.0 is the latest release, so there\n    // is no upstream fix to take, but a patch is a standing maintenance cost and\n    // this is one assignment that stays correct even if upstream fixes it).\n    //\n    // `sheets: []` rather than `{}` because `(this.model.sheets || [])` is what is\n    // read.\n    //\n    // The double assertion is NOT laziness and a single `as typeof reader.model`\n    // does not compile — checked, not assumed:\n    //\n    //   TS2352: Conversion of type '{ sheets: never[]; }' to type 'WorkbookModel'\n    //   may be a mistake because neither type sufficiently overlaps with the\n    //   other. … missing the following properties: creator, lastModifiedBy,\n    //   lastPrinted, created, and 17 more.\n    //\n    // A deliberately PARTIAL seed is the whole point — exceljs overwrites it with\n    // the real model a few entries later, and filling 21 fields with invented\n    // values to satisfy the compiler would be worse than the assertion: it would\n    // put fabricated workbook metadata one `_parseWorkbook` failure away from\n    // being read as real.\n    // toolkit-smell-allow: partial seed for a 21-field type; single assertion rejected by TS2352 (see above)\n    reader.model = { sheets: [] } as unknown as typeof reader.model\n\n    let dataRowNumber = 0\n    let sheetIndex = 0\n    let matchedTarget = false\n\n    // Drive the worksheet iterator BY HAND rather than with `for await`, so the\n    // `finally` below can resume it. Abandoning exceljs's `parse()` generator —\n    // which `return`/`break` out of a `for await` does — LEAKS a temp file and an\n    // fd per deferred worksheet, because `tempFileCleanupCallback()` sits after\n    // the `yield*` with no `finally` of its own (`workbook-reader.js:140-149`).\n    //\n    // Measured, with a negative control: 25 early-returning reads of a two-sheet\n    // deferred-order workbook leaked 50 fds and 50 temp files; the draining shape\n    // below leaks 0 of each. `tmp`'s graceful-cleanup hook reclaims the files at\n    // NORMAL process exit only, which is no help to a long-lived queue worker and\n    // no help at all against the fd ceiling: at the #842 audit's 5000-run cap\n    // this is an EMFILE, and an EMFILE arrives as the uncaught stream error the\n    // `open()` above exists to prevent.\n    //\n    // THE COST IS PROPORTIONAL TO THE WORKSHEETS AFTER THE TARGET, which is zero\n    // for the single-sheet feeds this reader mostly sees and 4.7× on a workbook\n    // whose target is sheet 1 and whose sheet 2 holds 40k rows. Advancing the\n    // parent iterator WITHOUT consuming each trailing worksheet is far cheaper\n    // and does release the temp files — but it strands exceljs's\n    // `createReadStream` on each one, i.e. it trades the disk leak for the fd\n    // leak, and the fd leak is the one that ends in EMFILE. So: full drain.\n    const worksheets = reader[Symbol.asyncIterator]()\n    let readerExhausted = false\n    try {\n      while (true) {\n        const next = await worksheets.next()\n        if (next.done) {\n          readerExhausted = true\n          break\n        }\n        const worksheet = next.value\n        sheetIndex += 1\n        const isTarget =\n          targetSheetIndex === undefined ? sheetIndex === 1 : sheetIndex === targetSheetIndex\n\n        if (!isTarget) {\n          // Drain a non-target worksheet so the underlying zip stream advances to the\n          // next; exceljs's streaming reader stalls if a worksheet is left unconsumed.\n          for await (const _skip of worksheet) {\n            void _skip\n          }\n          continue\n        }\n        matchedTarget = true\n\n        let excelRowNum = 0\n        let header: string[] | null = null\n        for await (const row of worksheet) {\n          excelRowNum += 1\n          const cells = rowCells(row.values)\n\n          if (hasHeader && header === null) {\n            // Skip rows before the configured header row, then capture it (trimmed).\n            if (excelRowNum < headerRowIndex + 1) continue\n            header = cells.map((c) => c.trim())\n            // FAIL LOUDLY on an empty header, because the alternative is SILENT DATA\n            // CORRUPTION and we have watched it happen.\n            //\n            // exceljs assigns `this.sharedStrings = []` at the TOP of\n            // `_parseSharedStrings`, before parsing a single entry — and an empty\n            // array is truthy, which is what the inline-worksheet guard tests. So a\n            // shared-strings entry that yields nothing (a truncated or starved read;\n            // observed in CI under load) leaves every string cell resolving to\n            // `sharedStrings[index]` → `undefined`. The header parses to zero\n            // columns, every data row maps to `{}`, and the import reports SUCCESS\n            // with every value dropped. Reproduced deterministically by rebuilding a\n            // workbook with an empty `sst`: two rows read, no values, no error.\n            //\n            // A header row with no usable column is never a legitimate import, so\n            // there is no case to preserve. Refusing costs an operator a clear error\n            // and saves them a silently empty dataset they would have to notice\n            // themselves — the same posture the rest of this package takes toward\n            // config mistakes.\n            if (header.every((c) => c === '')) {\n              throw new XlsxEmptyHeaderError(excelRowNum)\n            }\n            assertUniqueHeaders(header)\n            continue\n          }\n\n          dataRowNumber += 1\n          const rec: Record<string, string> = {}\n          if (header) {\n            for (let i = 0; i < header.length; i += 1) {\n              const key = header[i]\n              if (key) rec[key] = cells[i] ?? ''\n            }\n          } else {\n            for (let i = 0; i < cells.length; i += 1) rec[String(i)] = cells[i] ?? ''\n          }\n          yield { rowNumber: dataRowNumber, row: rec }\n        }\n\n        // Target worksheet fully read. Stop yielding — the `finally` walks the\n        // remaining worksheets so exceljs reaches its cleanup.\n        break\n      }\n    } finally {\n      // Runs on every exit: target read, an empty-header throw, or the CONSUMER\n      // abandoning us mid-stream (a `break` in the caller's `for await` calls\n      // `.return()`, which lands here). Whatever the reason, exceljs still owns\n      // temp files it only releases as the consumer pulls past each worksheet.\n      //\n      // The whole body is CONTAINED, for the same reason `handle.close()` below\n      // is: a throw from a `finally` REPLACES whatever the try block was\n      // propagating. Trailing worksheets are parsed here, so a malformed one\n      // would overwrite an `XlsxEmptyHeaderError` with an XML error — turning\n      // the #842 audit's `corrupt` verdict into a meaningless `error` and\n      // failing an import whose target sheet read perfectly. Measured: two of\n      // four malformed-trailing-sheet variants masked the refusal before this\n      // catch existed. Cleanup is best-effort by definition; the error the\n      // caller came for is not.\n      try {\n        while (!readerExhausted) {\n          const rest = await worksheets.next()\n          if (rest.done) break\n          for await (const _skip of rest.value) {\n            void _skip\n          }\n        }\n      } catch {\n        // Deliberately swallowed — see above. A trailing worksheet we only read\n        // to satisfy exceljs's cleanup contract cannot be allowed to speak for\n        // the read the caller actually asked for. Worst case we leak the temp\n        // files this drain existed to release, which is what `main` did on\n        // EVERY read.\n      }\n    }\n\n    // Reached only when a specific sheet was requested but never matched — the\n    // first-sheet default always matches on a non-empty workbook.\n    if (targetSheetIndex !== undefined && !matchedTarget) {\n      throw new Error(`xlsx worksheet ${targetSheetIndex} not found`)\n    }\n  } finally {\n    // Order matters: destroy the reader of the fd before closing the fd.\n    source?.destroy()\n    await handle.close().catch(() => {\n      // Swallowed deliberately, and biome's `noUnsafeFinally` agrees: a throw\n      // here would REPLACE whatever the try block was propagating, and what it\n      // propagates is the point of this reader — an empty-header refusal has to\n      // reach the caller intact. Closing a read-only fd we are finished with can\n      // fail only in ways that are not actionable, and the fd is reclaimed at\n      // process exit regardless.\n    })\n  }\n}\n"],"mappings":"qFAkCA,IAAa,EAAb,cAA0C,KAAM,CAE9C,gBAEA,YAAY,EAAyB,CACnC,MACE,mBAAmB,EAAgB,qIAErC,EACA,KAAK,KAAO,uBACZ,KAAK,gBAAkB,CACzB,CACF,EAYa,EAAb,cAAuC,KAAM,CAC3C,YAAY,EAAkB,CAC5B,MAAM,cAAc,EAAS,wBAAwB,EACrD,KAAK,KAAO,mBACd,CACF,EAcA,SAAS,EAAa,EAAsB,CAC1C,GAAI,GAAM,KAAyB,MAAO,GAC1C,GAAI,OAAO,GAAM,SAAU,OAAO,EAClC,GAAI,OAAO,GAAM,UAAY,OAAO,GAAM,UAAW,OAAO,OAAO,CAAC,EACpE,GAAI,aAAa,KAAM,OAAO,EAAE,YAAY,EAE5C,GAAI,aAAc,GAAK,MAAM,QAAQ,EAAE,QAAQ,EAC7C,OAAO,EAAE,SAAS,IAAK,GAAM,EAAE,IAAI,CAAC,CAAC,KAAK,EAAE,EAE9C,GAAI,SAAU,GAAK,OAAO,EAAE,MAAS,SAAU,OAAO,EAAE,KACxD,GAAI,WAAY,GAAK,EAAE,SAAW,IAAA,IAAa,EAAE,SAAW,KAAM,CAChE,IAAM,EAAI,EAAE,OAGZ,OAFI,aAAa,KAAa,EAAE,YAAY,EAErC,OAAO,GAAM,SAAW,GAAK,OAAO,CAAC,CAC9C,CACA,MAAO,EACT,CAGA,SAAS,EAAS,EAA8D,CAC9E,IAAM,EAAgB,CAAC,EACvB,GAAI,MAAM,QAAQ,CAAM,EAEtB,IAAK,IAAI,EAAI,EAAG,EAAI,EAAO,OAAQ,GAAK,EAAG,EAAI,EAAI,GAAK,EAAa,EAAO,EAAE,OAE9E,IAAK,GAAM,CAAC,EAAG,KAAM,OAAO,QAAQ,CAAM,EAAG,CAC3C,IAAM,EAAM,OAAO,SAAS,EAAG,EAAE,EAC7B,CAAC,OAAO,MAAM,CAAG,GAAK,GAAO,IAAG,EAAI,EAAM,GAAK,EAAa,CAAC,EACnE,CAIF,IAAK,IAAI,EAAI,EAAG,EAAI,EAAI,OAAQ,GAAK,EAAO,EAAI,KAAO,IAAA,KAAW,EAAI,GAAK,IAC3E,OAAO,CACT,CAQA,eAAuB,EAAW,EAAwD,CACxF,GAAM,CAAE,WAAU,YAAY,GAAM,iBAAiB,EAAG,SAAU,EAC5D,EAAU,MAAM,OAAO,WAMzB,EACJ,GAAI,IAAU,IAAA,KACZ,EAAmB,OAAO,SAAS,EAAO,EAAE,EACxC,OAAO,MAAM,CAAgB,GAAK,EAAmB,GACvD,MAAU,MAAM,uBAAuB,EAAM,uCAAuC,EAyBxF,IAAM,EAAS,MAAM,EAAK,EAAU,GAAG,EASnC,EACJ,GAAI,CAEF,GAAI,EAAC,MADe,EAAO,KAAK,EAAA,CACrB,OAAO,EAChB,MAAM,IAAI,EAAkB,CAAQ,EAGtC,EAAS,EAAO,iBAAiB,CAAE,UAAW,EAAM,CAAC,EACrD,IAAM,EAAS,IAAI,EAAQ,OAAO,KAAK,eAAe,EAAQ,CAC5D,WAAY,OACZ,cAAe,QACf,OAAQ,SACR,WAAY,SACZ,QAAS,QACX,CAAC,EA6CD,EAAO,MAAQ,CAAE,OAAQ,CAAC,CAAE,EAE5B,IAAI,EAAgB,EAChB,EAAa,EACb,EAAgB,GAuBd,EAAa,EAAO,OAAO,cAAc,CAAC,EAC5C,EAAkB,GACtB,GAAI,CACF,OAAa,CACX,IAAM,EAAO,MAAM,EAAW,KAAK,EACnC,GAAI,EAAK,KAAM,CACb,EAAkB,GAClB,KACF,CACA,IAAM,EAAY,EAAK,MAKvB,GAJA,GAAc,EAIV,EAFF,IAAqB,IAAA,GAAY,IAAe,EAAI,IAAe,GAEtD,CAGb,UAAW,IAAM,KAAS,GAG1B,QACF,CACA,EAAgB,GAEhB,IAAI,EAAc,EACd,EAA0B,KAC9B,UAAW,IAAM,KAAO,EAAW,CACjC,GAAe,EACf,IAAM,EAAQ,EAAS,EAAI,MAAM,EAEjC,GAAI,GAAa,IAAW,KAAM,CAEhC,GAAI,EAAc,EAAiB,EAAG,SAoBtC,GAnBA,EAAS,EAAM,IAAK,GAAM,EAAE,KAAK,CAAC,EAmB9B,EAAO,MAAO,GAAM,IAAM,EAAE,EAC9B,MAAM,IAAI,EAAqB,CAAW,EAE5C,EAAoB,CAAM,EAC1B,QACF,CAEA,GAAiB,EACjB,IAAM,EAA8B,CAAC,EACrC,GAAI,EACF,IAAK,IAAI,EAAI,EAAG,EAAI,EAAO,OAAQ,GAAK,EAAG,CACzC,IAAM,EAAM,EAAO,GACf,IAAK,EAAI,GAAO,EAAM,IAAM,GAClC,MAEA,IAAK,IAAI,EAAI,EAAG,EAAI,EAAM,OAAQ,GAAK,EAAG,EAAI,OAAO,CAAC,GAAK,EAAM,IAAM,GAEzE,KAAM,CAAE,UAAW,EAAe,IAAK,CAAI,CAC7C,CAIA,KACF,CACF,QAAU,CAeR,GAAI,CACF,KAAO,CAAC,GAAiB,CACvB,IAAM,EAAO,MAAM,EAAW,KAAK,EACnC,GAAI,EAAK,KAAM,MACf,UAAW,IAAM,KAAS,EAAK,OAGjC,CACF,MAAQ,CAMR,CACF,CAIA,GAAI,IAAqB,IAAA,IAAa,CAAC,EACrC,MAAU,MAAM,kBAAkB,EAAiB,WAAW,CAElE,QAAU,CAER,GAAQ,QAAQ,EAChB,MAAM,EAAO,MAAM,CAAC,CAAC,UAAY,CAOjC,CAAC,CACH,CACF"}