sensemaking 0.22.2 → 0.23.1

package/dist/esm/chunk/extract.js

@@ -1,5 +1,6 @@
-// mdast has no wikilink, embed, or %%comment%% node -- literal text resolved by regex passes
-// over non-code text (code/inlineCode is placeholder-held and spliced back verbatim, below).
+import { parser } from './parser.js';
+// markdown-it, like mdast, has no wikilink, embed, or %%comment%% token -- literal text resolved
+// by regex passes over non-code text (code/inline code is placeholder-held and spliced back verbatim).
 const COMMENT_PAIR = /%%[\s\S]*?%%/g;
 // An embed always yields its target, whatever follows a pipe (a resize suffix, never an alias);
 // must run before the wikilink regex below, bang included, or that regex matches first.
@@ -11,7 +12,7 @@ const BLOCK_ID = / \^[A-Za-z0-9-]+$/gm;
 // A callout marker is only ever the first thing in a blockquote's text (Obsidian's grammar),
 // so this anchors to blockquote output alone -- never applied to prose in general.
 const CALLOUT_MARKER = /^\[!\w[\w-]*\][+-]?[ \t]?/;
-// An html node's value is raw HTML, block or inline: a full <!-- --> comment is dropped, any
+// An html token's content is raw HTML, block or inline: a full <!-- --> comment is dropped, any
 // remaining tags are stripped, and the text a browser would still render survives.
 const HTML_COMMENT = /<!--[\s\S]*?-->/g;
 const HTML_TAG = /<[^>]+>/g;
@@ -38,85 +39,297 @@ function embedText(inner) {
 function resolveFlavor(text) {
     return text.replace(HTML_COMMENT, '').replace(COMMENT_PAIR, '').replace(EMBED, (_, inner)=>embedText(inner)).replace(WIKILINK, (_, inner)=>wikilinkText(inner)).replace(BLOCK_ID, '');
 }
-// Sibling blocks (list items, blockquote content, footnote bodies) joined one per line;
-// empties dropped so a blank child never widens the gap between its neighbors.
-function joinBlocks(nodes, code) {
-    return nodes.map((n)=>extractNode(n, code)).filter((s)=>s.length > 0).join('\n');
-}
-// Inline content (headings, paragraphs, emphasis, links) concatenated with no separator: the
-// source text nodes already carry whatever spacing separates them.
-function joinInline(nodes, code) {
-    return nodes.map((n)=>extractNode(n, code)).join('');
-}
 // U+E000 (Private Use Area) never appears in real markdown text, so it is a collision-free
 // placeholder delimiter, spliced back to the literal code value after flavor resolution.
 function codePlaceholder(value, code) {
     const idx = code.push(value) - 1;
     return `\uE000${idx}\uE000`;
 }
-// A GFM autolink's display text is its own target, missing only the scheme
-// mdast-util-gfm-autolink-literal fills into node.url (http:// for www., mailto: for an email).
-const SCHEME_PREFIX = /^[a-z][a-z0-9+.-]*:\/{0,2}/i;
-function isAutolink(url, text) {
-    return url === text || url.replace(SCHEME_PREFIX, '') === text;
-}
-function extractNode(node, code) {
-    switch(node.type){
-        case 'root':
-        case 'list':
-        case 'listItem':
-        case 'footnoteDefinition':
-            return joinBlocks(node.children, code);
-        case 'blockquote':
-            return joinBlocks(node.children, code).replace(CALLOUT_MARKER, '');
-        case 'link':
+// GFM's autolink literal ends where its path machine stops: a trail of punctuation is a genuine
+// end (link stops before it) only when trailEndsAt accepts it, else the marks are part of the link.
+const PUNCT = new Set([
+    '!',
+    '"',
+    '&',
+    "'",
+    ')',
+    '*',
+    ',',
+    '.',
+    ';',
+    ':',
+    '<',
+    '?',
+    '_',
+    '~',
+    ']'
+]);
+const TRAIL_PUNCT = new Set([
+    '!',
+    '"',
+    "'",
+    ')',
+    '*',
+    ',',
+    '.',
+    ';',
+    ':',
+    '?',
+    '_',
+    '~'
+]);
+const isAlpha = (c)=>c !== undefined && (c >= 'a' && c <= 'z' || c >= 'A' && c <= 'Z');
+function trailEndsAt(s, j) {
+    for(;;){
+        const c = s[j];
+        if (c === undefined) return true;
+        if (TRAIL_PUNCT.has(c)) {
+            j += 1;
+            continue;
+        }
+        if (c === '&') {
+            j += 1;
+            if (!isAlpha(s[j])) return false;
+            j += 1;
+            while(isAlpha(s[j]))j += 1;
+            if (s[j] !== ';') return false;
+            j += 1;
+            continue;
+        }
+        if (c === ']') {
+            j += 1;
+            const d = s[j];
+            return d === undefined || d === '(' || d === '[' || /\s/u.test(d !== null && d !== void 0 ? d : ' ');
+        }
+        if (c === '<') return true;
+        if (c !== undefined && /^\s$/u.test(c)) return true;
+        return false;
+    }
+}
+// Where the GFM autolink ends in S (the linkifier's text plus the following text), old-style:
+// punctuation that fails the trail test, or a ) with closes <= opens, extends the link.
+function gfmAutolinkEnd(s) {
+    let open = 0;
+    let close = 0;
+    for(let i = 0; i < s.length; i++){
+        const c = s[i];
+        if (c === '(') {
+            open += 1;
+            continue;
+        }
+        if (c === ')' && close < open) {
+            close += 1;
+            continue;
+        }
+        if (PUNCT.has(c) && trailEndsAt(s, i)) return i;
+        if (c !== undefined && /^\s$/u.test(c)) return i;
+    }
+    return s.length;
+}
+// linkify-it links more than GFM's autolink literal (bare domains, ftp://, //host); only the
+// GFM forms (http(s)://, www., email) drop their text.
+function isGfmUrl(text) {
+    return /^https?:\/\//i.test(text) || /^www\./i.test(text);
+}
+// Only a bare address is GFM's email autolink: a url keeps its @, since a path can hold one.
+function isGfmEmail(text) {
+    return !isGfmUrl(text) && text.includes('@');
+}
+// The task-lists plugin injects its checkbox as the item's first inline child.
+function isTaskCheckbox(token) {
+    return (token === null || token === void 0 ? void 0 : token.type) === 'html_inline' && /^<input\b/i.test(token.content);
+}
+// A linkify span can swallow the & of a following entity (span ...c& + text amp; z); the
+// re-emitted lead is then a reference the entity rule never saw, so decode it old-style.
+const LEAD_ENTITY = /^&[a-z#][a-z0-9]{1,31};/i;
+// Split a flat token range into its top-level blocks: nesting 1 opens, nesting 0 is one block.
+function topLevelBlocks(tokens) {
+    const out = [];
+    let i = 0;
+    while(i < tokens.length){
+        const token = tokens[i];
+        if (token.nesting === 1) {
+            let depth = 1;
+            let j = i + 1;
+            while(j < tokens.length && depth > 0){
+                depth += tokens[j].nesting === 1 ? 1 : tokens[j].nesting === -1 ? -1 : 0;
+                j += 1;
+            }
+            out.push(tokens.slice(i, j));
+            i = j;
+        } else if (token.nesting === 0) {
+            out.push(tokens.slice(i, i + 1));
+            i += 1;
+        } else {
+            i += 1;
+        }
+    }
+    return out;
+}
+// A first line with no source content (a bare task checkbox, whitespace) contributes no line to
+// the value, so its softbreak is dropped; a first line whose content extraction removes (an
+// autolink, an image) still leaves its line ending behind.
+function paragraphText(children, code) {
+    var _children_k;
+    let k = 0;
+    const first = children[0];
+    if (isTaskCheckbox(first)) k = 1;
+    while(k < children.length && children[k].type === 'text' && children[k].content.trim() === '')k += 1;
+    if (k > 0 && ((_children_k = children[k]) === null || _children_k === void 0 ? void 0 : _children_k.type) === 'softbreak') return inlineText(children.slice(k + 1), code);
+    return inlineText(children, code);
+}
+function blockText(tokens, code) {
+    const first = tokens[0];
+    if (!first) return '';
+    switch(first.type){
+        case 'heading_open':
+        case 'paragraph_open':
             {
-                const text = joinInline(node.children, code);
-                return isAutolink(node.url, text) ? '' : text;
+                var _inline_children;
+                const inline = tokens.find((t)=>t.type === 'inline');
+                return inline ? paragraphText((_inline_children = inline.children) !== null && _inline_children !== void 0 ? _inline_children : [], code) : '';
             }
-        case 'heading':
-        case 'paragraph':
-        case 'linkReference':
-        case 'emphasis':
-        case 'strong':
-        case 'delete':
-            return joinInline(node.children, code);
-        case 'table':
-            return node.children.map((n)=>extractNode(n, code)).join('\n');
-        case 'tableRow':
-            return node.children.map((n)=>extractNode(n, code)).join(' ');
-        case 'tableCell':
-            return joinInline(node.children, code);
-        case 'image':
-        case 'imageReference':
-            var _node_alt;
-            return (_node_alt = node.alt) !== null && _node_alt !== void 0 ? _node_alt : '';
-        case 'text':
-            return node.value;
-        case 'inlineCode':
-        case 'code':
-            return codePlaceholder(node.value, code);
-        case 'html':
-            return stripHtml(node.value);
-        case 'break':
-            return ' ';
+        case 'fence':
+        case 'code_block':
+            // Fence and indented code tokens carry the closing newline their lines were joined with.
+            return codePlaceholder(first.content.replace(/\n$/, ''), code);
+        case 'table_open':
+            return tableText(tokens, code);
+        case 'ordered_list_open':
+        case 'bullet_list_open':
+            return listItemsText(tokens, code);
+        case 'blockquote_open':
+            return blocksIn(tokens.slice(1, -1), code).replace(CALLOUT_MARKER, '');
+        case 'html_block':
+            // A non-final-line html block carries its last line's terminator, which the mdast value lacked.
+            return stripHtml(first.content.replace(/\n$/, ''));
+        case 'footnote_reference_open':
+            return blocksIn(tokens.slice(1, -1), code);
         default:
             return '';
     }
 }
+// Sibling blocks (list items, blockquote children, footnote bodies) joined one per line;
+// empties dropped so a blank child never widens the gap between its neighbors.
+function blocksIn(tokens, code) {
+    return topLevelBlocks(tokens).map((b)=>blockText(b, code)).filter((s)=>s.length > 0).join('\n');
+}
+// The task-lists plugin slices 3 of the marker's 4 chars, leaving its trailing space on the text
+// after the checkbox token. Returns a copy with that space dropped, never a mutation: a block is
+// extracted more than once (parse pre-extracts headings, group.ts extracts again).
+function withoutTaskMarker(item) {
+    for(let i = 0; i < item.length; i++){
+        var _token_children;
+        const token = item[i];
+        // The checkbox, if any, sits in the item's first inline only.
+        if (token.type !== 'inline' || !((_token_children = token.children) === null || _token_children === void 0 ? void 0 : _token_children.length)) continue;
+        const second = token.children[1];
+        if (!isTaskCheckbox(token.children[0]) || (second === null || second === void 0 ? void 0 : second.type) !== 'text' || !second.content.startsWith(' ')) return item;
+        const children = token.children.slice();
+        children[1] = {
+            ...second,
+            content: second.content.slice(1)
+        };
+        const copy = item.slice();
+        copy[i] = {
+            ...token,
+            children
+        };
+        return copy;
+    }
+    return item;
+}
+function listItemsText(tokens, code) {
+    const items = [];
+    for (const item of topLevelBlocks(tokens.slice(1, -1))){
+        items.push(blocksIn(withoutTaskMarker(item).slice(1, -1), code));
+    }
+    return items.filter((s)=>s.length > 0).join('\n');
+}
+// Table rows joined by newline, cells by space (mdast's tableRow/tableCell joiners).
+function tableText(tokens, code) {
+    const rows = [];
+    let cells = [];
+    for (const token of tokens){
+        var _token_children;
+        if (token.type === 'tr_open') cells = [];
+        else if (token.type === 'inline') cells.push(inlineText((_token_children = token.children) !== null && _token_children !== void 0 ? _token_children : [], code));
+        else if (token.type === 'tr_close') rows.push(cells.join(' '));
+    }
+    return rows.join('\n');
+}
+// Inline content (headings, paragraphs, emphasis, links) concatenated with no separator: the
+// source text tokens already carry whatever spacing separates them.
+function inlineText(tokens, code, inAlt = false) {
+    let out = '';
+    for(let i = 0; i < tokens.length; i++){
+        const token = tokens[i];
+        if (token.type === 'link_open') {
+            let skip = 0;
+            let j = i + 1;
+            let depth = 1;
+            while(j < tokens.length && depth > 0){
+                if (tokens[j].type === 'link_open') depth += 1;
+                else if (tokens[j].type === 'link_close') depth -= 1;
+                j += 1;
+            }
+            const text = inlineText(tokens.slice(i + 1, j - 1), code, inAlt);
+            const markup = token.markup;
+            const next = tokens[j];
+            if (inAlt) {
+                // An image's alt is plain text: mdast kept every link's own text there, autolinks included.
+                out += text;
+            } else if (markup === 'linkify' && isGfmUrl(text)) {
+                // GFM drop: the next text token resumes at the old link's boundary, re-emitting any
+                // trimmed trail. <www.x> has no trail to absorb (the brackets bound it), keep the >.
+                // Emails keep their whole remainder: GFM's email machine has no trail, so the span
+                // already ends at the old boundary and the entity rule has decoded the rest.
+                const prev = tokens[i - 1];
+                const bracketed = (prev === null || prev === void 0 ? void 0 : prev.type) === 'text' && prev.content.endsWith('<') && (next === null || next === void 0 ? void 0 : next.type) === 'text' && next.content.startsWith('>');
+                const tail = (next === null || next === void 0 ? void 0 : next.type) === 'text' ? text + next.content : text;
+                const k = bracketed ? text.length : gfmAutolinkEnd(tail);
+                out += tail.slice(k).replace(LEAD_ENTITY, (m)=>parser().utils.unescapeAll(m));
+                // The trail came out of the next text token, so that token is consumed here, not re-read.
+                if ((next === null || next === void 0 ? void 0 : next.type) === 'text') skip = 1;
+            } else if (markup === 'linkify' && isGfmEmail(text)) {
+            // Email drop: nothing after the span to trim, so the span alone is dropped.
+            } else if (markup === 'autolink') {
+            // <...> leaf: dropped whole, its brackets live inside the token so nothing follows to trim.
+            } else {
+                out += text; // user link or non-GFM linkify target: keep the display text
+            }
+            i = j - 1 + skip;
+        } else if (token.type === 'text') {
+            out += token.content;
+        } else if (token.type === 'softbreak') {
+            out += '\n';
+        } else if (token.type === 'hardbreak') {
+            out += ' ';
+        } else if (token.type === 'code_inline') {
+            out += codePlaceholder(token.content, code);
+        } else if (token.type === 'html_inline') {
+            out += stripHtml(token.content);
+        } else if (token.type === 'image') {
+            var _token_children;
+            out += inlineText((_token_children = token.children) !== null && _token_children !== void 0 ? _token_children : [], code, true);
+        }
+    // Emphasis, strikethrough and footnote markers carry no text of their own.
+    }
+    return out;
+}
 const CODE_PLACEHOLDER = /\uE000(\d+)\uE000/g;
-// Plain text of one mdast node (or a whole tree): heading/list/table structure is kept as text,
-// markup (emphasis, link targets, task and callout markers) is dropped. Pure, synchronous.
-export function extractText(node) {
+// Plain text of one parsed block: heading/list/table structure is kept as text, markup (emphasis,
+// link targets, task and callout markers) is dropped. Pure, synchronous.
+export function extractText(block) {
     return extractTexts([
-        node
+        block
     ]);
 }
 // Blocks resolved together: a blank line inside %%...%% splits it across blocks, and a per-block
 // strip then never sees the closing %%. Code stays placeholder-held across all of them.
-export function extractTexts(nodes) {
+export function extractTexts(blocks) {
     const code = [];
-    const joined = nodes.map((n)=>extractNode(n, code)).filter((s)=>s.length > 0).join('\n');
+    const joined = blocks.map((b)=>blockText(b.node, code)).filter((s)=>s.length > 0).join('\n');
     const resolved = resolveFlavor(joined);
     return code.length === 0 ? resolved : resolved.replace(CODE_PLACEHOLDER, (_, i)=>code[Number(i)]);
 }

package/dist/esm/chunk/extract.js.map

@@ -1 +1 @@
-{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/chunk/extract.ts"],"sourcesContent":["import type { Nodes } from 'mdast';\n\n// mdast has no wikilink, embed, or %%comment%% node -- literal text resolved by regex passes\n// over non-code text (code/inlineCode is placeholder-held and spliced back verbatim, below).\nconst COMMENT_PAIR = /%%[\\s\\S]*?%%/g;\n// An embed always yields its target, whatever follows a pipe (a resize suffix, never an alias);\n// must run before the wikilink regex below, bang included, or that regex matches first.\nconst EMBED = /!\\[\\[([^\\]]+)\\]\\]/g;\nconst WIKILINK = /\\[\\[([^\\]]+)\\]\\]/g;\n// A trailing \" ^id\" is an Obsidian block anchor, valid only at a line's end; a caret elsewhere\n// (mid-line) is ordinary text and is left alone.\nconst BLOCK_ID = / \\^[A-Za-z0-9-]+$/gm;\n// A callout marker is only ever the first thing in a blockquote's text (Obsidian's grammar),\n// so this anchors to blockquote output alone -- never applied to prose in general.\nconst CALLOUT_MARKER = /^\\[!\\w[\\w-]*\\][+-]?[ \\t]?/;\n// An html node's value is raw HTML, block or inline: a full <!-- --> comment is dropped, any\n// remaining tags are stripped, and the text a browser would still render survives.\nconst HTML_COMMENT = /<!--[\\s\\S]*?-->/g;\nconst HTML_TAG = /<[^>]+>/g;\n\nfunction stripHtml(value: string): string {\n  return value.replace(HTML_COMMENT, '').replace(HTML_TAG, '');\n}\n\n// A `#anchor` keeps its text (hash dropped), unlike links.ts's parseWikilinkInner (the resolver\n// authority), which discards it entirely -- replicated here since that return value doesn't fit.\nfunction anchorText(base: string): string {\n  const hashIdx = base.indexOf('#');\n  if (hashIdx === -1) return base.trim();\n  const target = base.slice(0, hashIdx).trim();\n  const anchor = base.slice(hashIdx + 1).trim();\n  return target ? `${target} ${anchor}` : anchor;\n}\n\nfunction wikilinkText(inner: string): string {\n  const pipeIdx = inner.indexOf('|');\n  return pipeIdx === -1 ? anchorText(inner) : inner.slice(pipeIdx + 1).trim();\n}\n\nfunction embedText(inner: string): string {\n  const pipeIdx = inner.indexOf('|');\n  return anchorText(pipeIdx === -1 ? inner : inner.slice(0, pipeIdx));\n}\n\nfunction resolveFlavor(text: string): string {\n  return text\n    .replace(HTML_COMMENT, '')\n    .replace(COMMENT_PAIR, '')\n    .replace(EMBED, (_, inner: string) => embedText(inner))\n    .replace(WIKILINK, (_, inner: string) => wikilinkText(inner))\n    .replace(BLOCK_ID, '');\n}\n\n// Sibling blocks (list items, blockquote content, footnote bodies) joined one per line;\n// empties dropped so a blank child never widens the gap between its neighbors.\nfunction joinBlocks(nodes: Nodes[], code: string[]): string {\n  return nodes\n    .map((n) => extractNode(n, code))\n    .filter((s) => s.length > 0)\n    .join('\\n');\n}\n\n// Inline content (headings, paragraphs, emphasis, links) concatenated with no separator: the\n// source text nodes already carry whatever spacing separates them.\nfunction joinInline(nodes: Nodes[], code: string[]): string {\n  return nodes.map((n) => extractNode(n, code)).join('');\n}\n\n// U+E000 (Private Use Area) never appears in real markdown text, so it is a collision-free\n// placeholder delimiter, spliced back to the literal code value after flavor resolution.\nfunction codePlaceholder(value: string, code: string[]): string {\n  const idx = code.push(value) - 1;\n  return `\\uE000${idx}\\uE000`;\n}\n\n// A GFM autolink's display text is its own target, missing only the scheme\n// mdast-util-gfm-autolink-literal fills into node.url (http:// for www., mailto: for an email).\nconst SCHEME_PREFIX = /^[a-z][a-z0-9+.-]*:\\/{0,2}/i;\nfunction isAutolink(url: string, text: string): boolean {\n  return url === text || url.replace(SCHEME_PREFIX, '') === text;\n}\n\nfunction extractNode(node: Nodes, code: string[]): string {\n  switch (node.type) {\n    case 'root':\n    case 'list':\n    case 'listItem':\n    case 'footnoteDefinition':\n      return joinBlocks(node.children, code);\n    case 'blockquote':\n      return joinBlocks(node.children, code).replace(CALLOUT_MARKER, '');\n    case 'link': {\n      const text = joinInline(node.children, code);\n      return isAutolink(node.url, text) ? '' : text;\n    }\n    case 'heading':\n    case 'paragraph':\n    case 'linkReference':\n    case 'emphasis':\n    case 'strong':\n    case 'delete':\n      return joinInline(node.children, code);\n    case 'table':\n      return node.children.map((n) => extractNode(n, code)).join('\\n');\n    case 'tableRow':\n      return node.children.map((n) => extractNode(n, code)).join(' ');\n    case 'tableCell':\n      return joinInline(node.children, code);\n    case 'image':\n    case 'imageReference':\n      return node.alt ?? '';\n    case 'text':\n      return node.value;\n    case 'inlineCode':\n    case 'code':\n      return codePlaceholder(node.value, code);\n    case 'html':\n      return stripHtml(node.value);\n    case 'break':\n      return ' ';\n    default:\n      return '';\n  }\n}\n\nconst CODE_PLACEHOLDER = /\\uE000(\\d+)\\uE000/g;\n\n// Plain text of one mdast node (or a whole tree): heading/list/table structure is kept as text,\n// markup (emphasis, link targets, task and callout markers) is dropped. Pure, synchronous.\nexport function extractText(node: Nodes): string {\n  return extractTexts([node]);\n}\n\n// Blocks resolved together: a blank line inside %%...%% splits it across blocks, and a per-block\n// strip then never sees the closing %%. Code stays placeholder-held across all of them.\nexport function extractTexts(nodes: Nodes[]): string {\n  const code: string[] = [];\n  const joined = nodes\n    .map((n) => extractNode(n, code))\n    .filter((s) => s.length > 0)\n    .join('\\n');\n  const resolved = resolveFlavor(joined);\n  return code.length === 0 ? resolved : resolved.replace(CODE_PLACEHOLDER, (_, i: string) => code[Number(i)]);\n}\n"],"names":["COMMENT_PAIR","EMBED","WIKILINK","BLOCK_ID","CALLOUT_MARKER","HTML_COMMENT","HTML_TAG","stripHtml","value","replace","anchorText","base","hashIdx","indexOf","trim","target","slice","anchor","wikilinkText","inner","pipeIdx","embedText","resolveFlavor","text","_","joinBlocks","nodes","code","map","n","extractNode","filter","s","length","join","joinInline","codePlaceholder","idx","push","SCHEME_PREFIX","isAutolink","url","node","type","children","alt","CODE_PLACEHOLDER","extractText","extractTexts","joined","resolved","i","Number"],"mappings":"AAEA,6FAA6F;AAC7F,6FAA6F;AAC7F,MAAMA,eAAe;AACrB,gGAAgG;AAChG,wFAAwF;AACxF,MAAMC,QAAQ;AACd,MAAMC,WAAW;AACjB,+FAA+F;AAC/F,iDAAiD;AACjD,MAAMC,WAAW;AACjB,6FAA6F;AAC7F,mFAAmF;AACnF,MAAMC,iBAAiB;AACvB,6FAA6F;AAC7F,mFAAmF;AACnF,MAAMC,eAAe;AACrB,MAAMC,WAAW;AAEjB,SAASC,UAAUC,KAAa;IAC9B,OAAOA,MAAMC,OAAO,CAACJ,cAAc,IAAII,OAAO,CAACH,UAAU;AAC3D;AAEA,gGAAgG;AAChG,iGAAiG;AACjG,SAASI,WAAWC,IAAY;IAC9B,MAAMC,UAAUD,KAAKE,OAAO,CAAC;IAC7B,IAAID,YAAY,CAAC,GAAG,OAAOD,KAAKG,IAAI;IACpC,MAAMC,SAASJ,KAAKK,KAAK,CAAC,GAAGJ,SAASE,IAAI;IAC1C,MAAMG,SAASN,KAAKK,KAAK,CAACJ,UAAU,GAAGE,IAAI;IAC3C,OAAOC,SAAS,GAAGA,OAAO,CAAC,EAAEE,QAAQ,GAAGA;AAC1C;AAEA,SAASC,aAAaC,KAAa;IACjC,MAAMC,UAAUD,MAAMN,OAAO,CAAC;IAC9B,OAAOO,YAAY,CAAC,IAAIV,WAAWS,SAASA,MAAMH,KAAK,CAACI,UAAU,GAAGN,IAAI;AAC3E;AAEA,SAASO,UAAUF,KAAa;IAC9B,MAAMC,UAAUD,MAAMN,OAAO,CAAC;IAC9B,OAAOH,WAAWU,YAAY,CAAC,IAAID,QAAQA,MAAMH,KAAK,CAAC,GAAGI;AAC5D;AAEA,SAASE,cAAcC,IAAY;IACjC,OAAOA,KACJd,OAAO,CAACJ,cAAc,IACtBI,OAAO,CAACT,cAAc,IACtBS,OAAO,CAACR,OAAO,CAACuB,GAAGL,QAAkBE,UAAUF,QAC/CV,OAAO,CAACP,UAAU,CAACsB,GAAGL,QAAkBD,aAAaC,QACrDV,OAAO,CAACN,UAAU;AACvB;AAEA,wFAAwF;AACxF,+EAA+E;AAC/E,SAASsB,WAAWC,KAAc,EAAEC,IAAc;IAChD,OAAOD,MACJE,GAAG,CAAC,CAACC,IAAMC,YAAYD,GAAGF,OAC1BI,MAAM,CAAC,CAACC,IAAMA,EAAEC,MAAM,GAAG,GACzBC,IAAI,CAAC;AACV;AAEA,6FAA6F;AAC7F,mEAAmE;AACnE,SAASC,WAAWT,KAAc,EAAEC,IAAc;IAChD,OAAOD,MAAME,GAAG,CAAC,CAACC,IAAMC,YAAYD,GAAGF,OAAOO,IAAI,CAAC;AACrD;AAEA,2FAA2F;AAC3F,yFAAyF;AACzF,SAASE,gBAAgB5B,KAAa,EAAEmB,IAAc;IACpD,MAAMU,MAAMV,KAAKW,IAAI,CAAC9B,SAAS;IAC/B,OAAO,CAAC,MAAM,EAAE6B,IAAI,MAAM,CAAC;AAC7B;AAEA,2EAA2E;AAC3E,gGAAgG;AAChG,MAAME,gBAAgB;AACtB,SAASC,WAAWC,GAAW,EAAElB,IAAY;IAC3C,OAAOkB,QAAQlB,QAAQkB,IAAIhC,OAAO,CAAC8B,eAAe,QAAQhB;AAC5D;AAEA,SAASO,YAAYY,IAAW,EAAEf,IAAc;IAC9C,OAAQe,KAAKC,IAAI;QACf,KAAK;QACL,KAAK;QACL,KAAK;QACL,KAAK;YACH,OAAOlB,WAAWiB,KAAKE,QAAQ,EAAEjB;QACnC,KAAK;YACH,OAAOF,WAAWiB,KAAKE,QAAQ,EAAEjB,MAAMlB,OAAO,CAACL,gBAAgB;QACjE,KAAK;YAAQ;gBACX,MAAMmB,OAAOY,WAAWO,KAAKE,QAAQ,EAAEjB;gBACvC,OAAOa,WAAWE,KAAKD,GAAG,EAAElB,QAAQ,KAAKA;YAC3C;QACA,KAAK;QACL,KAAK;QACL,KAAK;QACL,KAAK;QACL,KAAK;QACL,KAAK;YACH,OAAOY,WAAWO,KAAKE,QAAQ,EAAEjB;QACnC,KAAK;YACH,OAAOe,KAAKE,QAAQ,CAAChB,GAAG,CAAC,CAACC,IAAMC,YAAYD,GAAGF,OAAOO,IAAI,CAAC;QAC7D,KAAK;YACH,OAAOQ,KAAKE,QAAQ,CAAChB,GAAG,CAAC,CAACC,IAAMC,YAAYD,GAAGF,OAAOO,IAAI,CAAC;QAC7D,KAAK;YACH,OAAOC,WAAWO,KAAKE,QAAQ,EAAEjB;QACnC,KAAK;QACL,KAAK;gBACIe;YAAP,QAAOA,YAAAA,KAAKG,GAAG,cAARH,uBAAAA,YAAY;QACrB,KAAK;YACH,OAAOA,KAAKlC,KAAK;QACnB,KAAK;QACL,KAAK;YACH,OAAO4B,gBAAgBM,KAAKlC,KAAK,EAAEmB;QACrC,KAAK;YACH,OAAOpB,UAAUmC,KAAKlC,KAAK;QAC7B,KAAK;YACH,OAAO;QACT;YACE,OAAO;IACX;AACF;AAEA,MAAMsC,mBAAmB;AAEzB,gGAAgG;AAChG,2FAA2F;AAC3F,OAAO,SAASC,YAAYL,IAAW;IACrC,OAAOM,aAAa;QAACN;KAAK;AAC5B;AAEA,iGAAiG;AACjG,wFAAwF;AACxF,OAAO,SAASM,aAAatB,KAAc;IACzC,MAAMC,OAAiB,EAAE;IACzB,MAAMsB,SAASvB,MACZE,GAAG,CAAC,CAACC,IAAMC,YAAYD,GAAGF,OAC1BI,MAAM,CAAC,CAACC,IAAMA,EAAEC,MAAM,GAAG,GACzBC,IAAI,CAAC;IACR,MAAMgB,WAAW5B,cAAc2B;IAC/B,OAAOtB,KAAKM,MAAM,KAAK,IAAIiB,WAAWA,SAASzC,OAAO,CAACqC,kBAAkB,CAACtB,GAAG2B,IAAcxB,IAAI,CAACyB,OAAOD,GAAG;AAC5G"}
+{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/chunk/extract.ts"],"sourcesContent":["import type { Token } from 'markdown-it';\nimport { parser } from './parser.ts';\nimport type { Block } from './types.ts';\n\n// markdown-it, like mdast, has no wikilink, embed, or %%comment%% token -- literal text resolved\n// by regex passes over non-code text (code/inline code is placeholder-held and spliced back verbatim).\nconst COMMENT_PAIR = /%%[\\s\\S]*?%%/g;\n// An embed always yields its target, whatever follows a pipe (a resize suffix, never an alias);\n// must run before the wikilink regex below, bang included, or that regex matches first.\nconst EMBED = /!\\[\\[([^\\]]+)\\]\\]/g;\nconst WIKILINK = /\\[\\[([^\\]]+)\\]\\]/g;\n// A trailing \" ^id\" is an Obsidian block anchor, valid only at a line's end; a caret elsewhere\n// (mid-line) is ordinary text and is left alone.\nconst BLOCK_ID = / \\^[A-Za-z0-9-]+$/gm;\n// A callout marker is only ever the first thing in a blockquote's text (Obsidian's grammar),\n// so this anchors to blockquote output alone -- never applied to prose in general.\nconst CALLOUT_MARKER = /^\\[!\\w[\\w-]*\\][+-]?[ \\t]?/;\n// An html token's content is raw HTML, block or inline: a full <!-- --> comment is dropped, any\n// remaining tags are stripped, and the text a browser would still render survives.\nconst HTML_COMMENT = /<!--[\\s\\S]*?-->/g;\nconst HTML_TAG = /<[^>]+>/g;\n\nfunction stripHtml(value: string): string {\n  return value.replace(HTML_COMMENT, '').replace(HTML_TAG, '');\n}\n\n// A `#anchor` keeps its text (hash dropped), unlike links.ts's parseWikilinkInner (the resolver\n// authority), which discards it entirely -- replicated here since that return value doesn't fit.\nfunction anchorText(base: string): string {\n  const hashIdx = base.indexOf('#');\n  if (hashIdx === -1) return base.trim();\n  const target = base.slice(0, hashIdx).trim();\n  const anchor = base.slice(hashIdx + 1).trim();\n  return target ? `${target} ${anchor}` : anchor;\n}\n\nfunction wikilinkText(inner: string): string {\n  const pipeIdx = inner.indexOf('|');\n  return pipeIdx === -1 ? anchorText(inner) : inner.slice(pipeIdx + 1).trim();\n}\n\nfunction embedText(inner: string): string {\n  const pipeIdx = inner.indexOf('|');\n  return anchorText(pipeIdx === -1 ? inner : inner.slice(0, pipeIdx));\n}\n\nfunction resolveFlavor(text: string): string {\n  return text\n    .replace(HTML_COMMENT, '')\n    .replace(COMMENT_PAIR, '')\n    .replace(EMBED, (_, inner: string) => embedText(inner))\n    .replace(WIKILINK, (_, inner: string) => wikilinkText(inner))\n    .replace(BLOCK_ID, '');\n}\n\n// U+E000 (Private Use Area) never appears in real markdown text, so it is a collision-free\n// placeholder delimiter, spliced back to the literal code value after flavor resolution.\nfunction codePlaceholder(value: string, code: string[]): string {\n  const idx = code.push(value) - 1;\n  return `\\uE000${idx}\\uE000`;\n}\n\n// GFM's autolink literal ends where its path machine stops: a trail of punctuation is a genuine\n// end (link stops before it) only when trailEndsAt accepts it, else the marks are part of the link.\nconst PUNCT = new Set(['!', '\"', '&', \"'\", ')', '*', ',', '.', ';', ':', '<', '?', '_', '~', ']']);\nconst TRAIL_PUNCT = new Set(['!', '\"', \"'\", ')', '*', ',', '.', ';', ':', '?', '_', '~']);\nconst isAlpha = (c: string | undefined): boolean => c !== undefined && ((c >= 'a' && c <= 'z') || (c >= 'A' && c <= 'Z'));\nfunction trailEndsAt(s: string, j: number): boolean {\n  for (;;) {\n    const c = s[j];\n    if (c === undefined) return true;\n    if (TRAIL_PUNCT.has(c)) {\n      j += 1;\n      continue;\n    }\n    if (c === '&') {\n      j += 1;\n      if (!isAlpha(s[j])) return false;\n      j += 1;\n      while (isAlpha(s[j])) j += 1;\n      if (s[j] !== ';') return false;\n      j += 1;\n      continue;\n    }\n    if (c === ']') {\n      j += 1;\n      const d = s[j];\n      return d === undefined || d === '(' || d === '[' || /\\s/u.test(d ?? ' ');\n    }\n    if (c === '<') return true;\n    if (c !== undefined && /^\\s$/u.test(c)) return true;\n    return false;\n  }\n}\n// Where the GFM autolink ends in S (the linkifier's text plus the following text), old-style:\n// punctuation that fails the trail test, or a ) with closes <= opens, extends the link.\nfunction gfmAutolinkEnd(s: string): number {\n  let open = 0;\n  let close = 0;\n  for (let i = 0; i < s.length; i++) {\n    const c = s[i];\n    if (c === '(') {\n      open += 1;\n      continue;\n    }\n    if (c === ')' && close < open) {\n      close += 1;\n      continue;\n    }\n    if (PUNCT.has(c) && trailEndsAt(s, i)) return i;\n    if (c !== undefined && /^\\s$/u.test(c)) return i;\n  }\n  return s.length;\n}\n// linkify-it links more than GFM's autolink literal (bare domains, ftp://, //host); only the\n// GFM forms (http(s)://, www., email) drop their text.\nfunction isGfmUrl(text: string): boolean {\n  return /^https?:\\/\\//i.test(text) || /^www\\./i.test(text);\n}\n// Only a bare address is GFM's email autolink: a url keeps its @, since a path can hold one.\nfunction isGfmEmail(text: string): boolean {\n  return !isGfmUrl(text) && text.includes('@');\n}\n// The task-lists plugin injects its checkbox as the item's first inline child.\nfunction isTaskCheckbox(token: Token | undefined): boolean {\n  return token?.type === 'html_inline' && /^<input\\b/i.test(token.content);\n}\n// A linkify span can swallow the & of a following entity (span ...c& + text amp; z); the\n// re-emitted lead is then a reference the entity rule never saw, so decode it old-style.\nconst LEAD_ENTITY = /^&[a-z#][a-z0-9]{1,31};/i;\n\n// Split a flat token range into its top-level blocks: nesting 1 opens, nesting 0 is one block.\nfunction topLevelBlocks(tokens: Token[]): Token[][] {\n  const out: Token[][] = [];\n  let i = 0;\n  while (i < tokens.length) {\n    const token = tokens[i];\n    if (token.nesting === 1) {\n      let depth = 1;\n      let j = i + 1;\n      while (j < tokens.length && depth > 0) {\n        depth += tokens[j].nesting === 1 ? 1 : tokens[j].nesting === -1 ? -1 : 0;\n        j += 1;\n      }\n      out.push(tokens.slice(i, j));\n      i = j;\n    } else if (token.nesting === 0) {\n      out.push(tokens.slice(i, i + 1));\n      i += 1;\n    } else {\n      i += 1;\n    }\n  }\n  return out;\n}\n\n// A first line with no source content (a bare task checkbox, whitespace) contributes no line to\n// the value, so its softbreak is dropped; a first line whose content extraction removes (an\n// autolink, an image) still leaves its line ending behind.\nfunction paragraphText(children: Token[], code: string[]): string {\n  let k = 0;\n  const first = children[0];\n  if (isTaskCheckbox(first)) k = 1;\n  while (k < children.length && children[k].type === 'text' && children[k].content.trim() === '') k += 1;\n  if (k > 0 && children[k]?.type === 'softbreak') return inlineText(children.slice(k + 1), code);\n  return inlineText(children, code);\n}\n\nfunction blockText(tokens: Token[], code: string[]): string {\n  const first = tokens[0];\n  if (!first) return '';\n  switch (first.type) {\n    case 'heading_open':\n    case 'paragraph_open': {\n      const inline = tokens.find((t) => t.type === 'inline');\n      return inline ? paragraphText(inline.children ?? [], code) : '';\n    }\n    case 'fence':\n    case 'code_block':\n      // Fence and indented code tokens carry the closing newline their lines were joined with.\n      return codePlaceholder(first.content.replace(/\\n$/, ''), code);\n    case 'table_open':\n      return tableText(tokens, code);\n    case 'ordered_list_open':\n    case 'bullet_list_open':\n      return listItemsText(tokens, code);\n    case 'blockquote_open':\n      return blocksIn(tokens.slice(1, -1), code).replace(CALLOUT_MARKER, '');\n    case 'html_block':\n      // A non-final-line html block carries its last line's terminator, which the mdast value lacked.\n      return stripHtml(first.content.replace(/\\n$/, ''));\n    case 'footnote_reference_open':\n      return blocksIn(tokens.slice(1, -1), code);\n    default:\n      return '';\n  }\n}\n\n// Sibling blocks (list items, blockquote children, footnote bodies) joined one per line;\n// empties dropped so a blank child never widens the gap between its neighbors.\nfunction blocksIn(tokens: Token[], code: string[]): string {\n  return topLevelBlocks(tokens)\n    .map((b) => blockText(b, code))\n    .filter((s) => s.length > 0)\n    .join('\\n');\n}\n\n// The task-lists plugin slices 3 of the marker's 4 chars, leaving its trailing space on the text\n// after the checkbox token. Returns a copy with that space dropped, never a mutation: a block is\n// extracted more than once (parse pre-extracts headings, group.ts extracts again).\nfunction withoutTaskMarker(item: Token[]): Token[] {\n  for (let i = 0; i < item.length; i++) {\n    const token = item[i];\n    // The checkbox, if any, sits in the item's first inline only.\n    if (token.type !== 'inline' || !token.children?.length) continue;\n    const second = token.children[1];\n    if (!isTaskCheckbox(token.children[0]) || second?.type !== 'text' || !second.content.startsWith(' ')) return item;\n    const children = token.children.slice();\n    children[1] = { ...second, content: second.content.slice(1) } as Token;\n    const copy = item.slice();\n    copy[i] = { ...token, children } as Token;\n    return copy;\n  }\n  return item;\n}\n\nfunction listItemsText(tokens: Token[], code: string[]): string {\n  const items: string[] = [];\n  for (const item of topLevelBlocks(tokens.slice(1, -1))) {\n    items.push(blocksIn(withoutTaskMarker(item).slice(1, -1), code));\n  }\n  return items.filter((s) => s.length > 0).join('\\n');\n}\n\n// Table rows joined by newline, cells by space (mdast's tableRow/tableCell joiners).\nfunction tableText(tokens: Token[], code: string[]): string {\n  const rows: string[] = [];\n  let cells: string[] = [];\n  for (const token of tokens) {\n    if (token.type === 'tr_open') cells = [];\n    else if (token.type === 'inline') cells.push(inlineText(token.children ?? [], code));\n    else if (token.type === 'tr_close') rows.push(cells.join(' '));\n  }\n  return rows.join('\\n');\n}\n\n// Inline content (headings, paragraphs, emphasis, links) concatenated with no separator: the\n// source text tokens already carry whatever spacing separates them.\nfunction inlineText(tokens: Token[], code: string[], inAlt = false): string {\n  let out = '';\n  for (let i = 0; i < tokens.length; i++) {\n    const token = tokens[i];\n    if (token.type === 'link_open') {\n      let skip = 0;\n      let j = i + 1;\n      let depth = 1;\n      while (j < tokens.length && depth > 0) {\n        if (tokens[j].type === 'link_open') depth += 1;\n        else if (tokens[j].type === 'link_close') depth -= 1;\n        j += 1;\n      }\n      const text = inlineText(tokens.slice(i + 1, j - 1), code, inAlt);\n      const markup = token.markup;\n      const next = tokens[j];\n      if (inAlt) {\n        // An image's alt is plain text: mdast kept every link's own text there, autolinks included.\n        out += text;\n      } else if (markup === 'linkify' && isGfmUrl(text)) {\n        // GFM drop: the next text token resumes at the old link's boundary, re-emitting any\n        // trimmed trail. <www.x> has no trail to absorb (the brackets bound it), keep the >.\n        // Emails keep their whole remainder: GFM's email machine has no trail, so the span\n        // already ends at the old boundary and the entity rule has decoded the rest.\n        const prev = tokens[i - 1];\n        const bracketed = prev?.type === 'text' && prev.content.endsWith('<') && next?.type === 'text' && next.content.startsWith('>');\n        const tail = next?.type === 'text' ? text + next.content : text;\n        const k = bracketed ? text.length : gfmAutolinkEnd(tail);\n        out += tail.slice(k).replace(LEAD_ENTITY, (m) => parser().utils.unescapeAll(m));\n        // The trail came out of the next text token, so that token is consumed here, not re-read.\n        if (next?.type === 'text') skip = 1;\n      } else if (markup === 'linkify' && isGfmEmail(text)) {\n        // Email drop: nothing after the span to trim, so the span alone is dropped.\n      } else if (markup === 'autolink') {\n        // <...> leaf: dropped whole, its brackets live inside the token so nothing follows to trim.\n      } else {\n        out += text; // user link or non-GFM linkify target: keep the display text\n      }\n      i = j - 1 + skip;\n    } else if (token.type === 'text') {\n      out += token.content;\n    } else if (token.type === 'softbreak') {\n      out += '\\n';\n    } else if (token.type === 'hardbreak') {\n      out += ' ';\n    } else if (token.type === 'code_inline') {\n      out += codePlaceholder(token.content, code);\n    } else if (token.type === 'html_inline') {\n      out += stripHtml(token.content);\n    } else if (token.type === 'image') {\n      out += inlineText(token.children ?? [], code, true);\n    }\n    // Emphasis, strikethrough and footnote markers carry no text of their own.\n  }\n  return out;\n}\n\nconst CODE_PLACEHOLDER = /\\uE000(\\d+)\\uE000/g;\n\n// Plain text of one parsed block: heading/list/table structure is kept as text, markup (emphasis,\n// link targets, task and callout markers) is dropped. Pure, synchronous.\nexport function extractText(block: Block): string {\n  return extractTexts([block]);\n}\n\n// Blocks resolved together: a blank line inside %%...%% splits it across blocks, and a per-block\n// strip then never sees the closing %%. Code stays placeholder-held across all of them.\nexport function extractTexts(blocks: Block[]): string {\n  const code: string[] = [];\n  const joined = blocks\n    .map((b) => blockText(b.node, code))\n    .filter((s) => s.length > 0)\n    .join('\\n');\n  const resolved = resolveFlavor(joined);\n  return code.length === 0 ? resolved : resolved.replace(CODE_PLACEHOLDER, (_, i: string) => code[Number(i)]);\n}\n"],"names":["parser","COMMENT_PAIR","EMBED","WIKILINK","BLOCK_ID","CALLOUT_MARKER","HTML_COMMENT","HTML_TAG","stripHtml","value","replace","anchorText","base","hashIdx","indexOf","trim","target","slice","anchor","wikilinkText","inner","pipeIdx","embedText","resolveFlavor","text","_","codePlaceholder","code","idx","push","PUNCT","Set","TRAIL_PUNCT","isAlpha","c","undefined","trailEndsAt","s","j","has","d","test","gfmAutolinkEnd","open","close","i","length","isGfmUrl","isGfmEmail","includes","isTaskCheckbox","token","type","content","LEAD_ENTITY","topLevelBlocks","tokens","out","nesting","depth","paragraphText","children","k","first","inlineText","blockText","inline","find","t","tableText","listItemsText","blocksIn","map","b","filter","join","withoutTaskMarker","item","second","startsWith","copy","items","rows","cells","inAlt","skip","markup","next","prev","bracketed","endsWith","tail","m","utils","unescapeAll","CODE_PLACEHOLDER","extractText","block","extractTexts","blocks","joined","node","resolved","Number"],"mappings":"AACA,SAASA,MAAM,QAAQ,cAAc;AAGrC,iGAAiG;AACjG,uGAAuG;AACvG,MAAMC,eAAe;AACrB,gGAAgG;AAChG,wFAAwF;AACxF,MAAMC,QAAQ;AACd,MAAMC,WAAW;AACjB,+FAA+F;AAC/F,iDAAiD;AACjD,MAAMC,WAAW;AACjB,6FAA6F;AAC7F,mFAAmF;AACnF,MAAMC,iBAAiB;AACvB,gGAAgG;AAChG,mFAAmF;AACnF,MAAMC,eAAe;AACrB,MAAMC,WAAW;AAEjB,SAASC,UAAUC,KAAa;IAC9B,OAAOA,MAAMC,OAAO,CAACJ,cAAc,IAAII,OAAO,CAACH,UAAU;AAC3D;AAEA,gGAAgG;AAChG,iGAAiG;AACjG,SAASI,WAAWC,IAAY;IAC9B,MAAMC,UAAUD,KAAKE,OAAO,CAAC;IAC7B,IAAID,YAAY,CAAC,GAAG,OAAOD,KAAKG,IAAI;IACpC,MAAMC,SAASJ,KAAKK,KAAK,CAAC,GAAGJ,SAASE,IAAI;IAC1C,MAAMG,SAASN,KAAKK,KAAK,CAACJ,UAAU,GAAGE,IAAI;IAC3C,OAAOC,SAAS,GAAGA,OAAO,CAAC,EAAEE,QAAQ,GAAGA;AAC1C;AAEA,SAASC,aAAaC,KAAa;IACjC,MAAMC,UAAUD,MAAMN,OAAO,CAAC;IAC9B,OAAOO,YAAY,CAAC,IAAIV,WAAWS,SAASA,MAAMH,KAAK,CAACI,UAAU,GAAGN,IAAI;AAC3E;AAEA,SAASO,UAAUF,KAAa;IAC9B,MAAMC,UAAUD,MAAMN,OAAO,CAAC;IAC9B,OAAOH,WAAWU,YAAY,CAAC,IAAID,QAAQA,MAAMH,KAAK,CAAC,GAAGI;AAC5D;AAEA,SAASE,cAAcC,IAAY;IACjC,OAAOA,KACJd,OAAO,CAACJ,cAAc,IACtBI,OAAO,CAACT,cAAc,IACtBS,OAAO,CAACR,OAAO,CAACuB,GAAGL,QAAkBE,UAAUF,QAC/CV,OAAO,CAACP,UAAU,CAACsB,GAAGL,QAAkBD,aAAaC,QACrDV,OAAO,CAACN,UAAU;AACvB;AAEA,2FAA2F;AAC3F,yFAAyF;AACzF,SAASsB,gBAAgBjB,KAAa,EAAEkB,IAAc;IACpD,MAAMC,MAAMD,KAAKE,IAAI,CAACpB,SAAS;IAC/B,OAAO,CAAC,MAAM,EAAEmB,IAAI,MAAM,CAAC;AAC7B;AAEA,gGAAgG;AAChG,oGAAoG;AACpG,MAAME,QAAQ,IAAIC,IAAI;IAAC;IAAK;IAAK;IAAK;IAAK;IAAK;IAAK;IAAK;IAAK;IAAK;IAAK;IAAK;IAAK;IAAK;IAAK;CAAI;AACjG,MAAMC,cAAc,IAAID,IAAI;IAAC;IAAK;IAAK;IAAK;IAAK;IAAK;IAAK;IAAK;IAAK;IAAK;IAAK;IAAK;CAAI;AACxF,MAAME,UAAU,CAACC,IAAmCA,MAAMC,aAAc,CAAA,AAACD,KAAK,OAAOA,KAAK,OAASA,KAAK,OAAOA,KAAK,GAAG;AACvH,SAASE,YAAYC,CAAS,EAAEC,CAAS;IACvC,OAAS;QACP,MAAMJ,IAAIG,CAAC,CAACC,EAAE;QACd,IAAIJ,MAAMC,WAAW,OAAO;QAC5B,IAAIH,YAAYO,GAAG,CAACL,IAAI;YACtBI,KAAK;YACL;QACF;QACA,IAAIJ,MAAM,KAAK;YACbI,KAAK;YACL,IAAI,CAACL,QAAQI,CAAC,CAACC,EAAE,GAAG,OAAO;YAC3BA,KAAK;YACL,MAAOL,QAAQI,CAAC,CAACC,EAAE,EAAGA,KAAK;YAC3B,IAAID,CAAC,CAACC,EAAE,KAAK,KAAK,OAAO;YACzBA,KAAK;YACL;QACF;QACA,IAAIJ,MAAM,KAAK;YACbI,KAAK;YACL,MAAME,IAAIH,CAAC,CAACC,EAAE;YACd,OAAOE,MAAML,aAAaK,MAAM,OAAOA,MAAM,OAAO,MAAMC,IAAI,CAACD,cAAAA,eAAAA,IAAK;QACtE;QACA,IAAIN,MAAM,KAAK,OAAO;QACtB,IAAIA,MAAMC,aAAa,QAAQM,IAAI,CAACP,IAAI,OAAO;QAC/C,OAAO;IACT;AACF;AACA,8FAA8F;AAC9F,wFAAwF;AACxF,SAASQ,eAAeL,CAAS;IAC/B,IAAIM,OAAO;IACX,IAAIC,QAAQ;IACZ,IAAK,IAAIC,IAAI,GAAGA,IAAIR,EAAES,MAAM,EAAED,IAAK;QACjC,MAAMX,IAAIG,CAAC,CAACQ,EAAE;QACd,IAAIX,MAAM,KAAK;YACbS,QAAQ;YACR;QACF;QACA,IAAIT,MAAM,OAAOU,QAAQD,MAAM;YAC7BC,SAAS;YACT;QACF;QACA,IAAId,MAAMS,GAAG,CAACL,MAAME,YAAYC,GAAGQ,IAAI,OAAOA;QAC9C,IAAIX,MAAMC,aAAa,QAAQM,IAAI,CAACP,IAAI,OAAOW;IACjD;IACA,OAAOR,EAAES,MAAM;AACjB;AACA,6FAA6F;AAC7F,uDAAuD;AACvD,SAASC,SAASvB,IAAY;IAC5B,OAAO,gBAAgBiB,IAAI,CAACjB,SAAS,UAAUiB,IAAI,CAACjB;AACtD;AACA,6FAA6F;AAC7F,SAASwB,WAAWxB,IAAY;IAC9B,OAAO,CAACuB,SAASvB,SAASA,KAAKyB,QAAQ,CAAC;AAC1C;AACA,+EAA+E;AAC/E,SAASC,eAAeC,KAAwB;IAC9C,OAAOA,CAAAA,kBAAAA,4BAAAA,MAAOC,IAAI,MAAK,iBAAiB,aAAaX,IAAI,CAACU,MAAME,OAAO;AACzE;AACA,yFAAyF;AACzF,yFAAyF;AACzF,MAAMC,cAAc;AAEpB,+FAA+F;AAC/F,SAASC,eAAeC,MAAe;IACrC,MAAMC,MAAiB,EAAE;IACzB,IAAIZ,IAAI;IACR,MAAOA,IAAIW,OAAOV,MAAM,CAAE;QACxB,MAAMK,QAAQK,MAAM,CAACX,EAAE;QACvB,IAAIM,MAAMO,OAAO,KAAK,GAAG;YACvB,IAAIC,QAAQ;YACZ,IAAIrB,IAAIO,IAAI;YACZ,MAAOP,IAAIkB,OAAOV,MAAM,IAAIa,QAAQ,EAAG;gBACrCA,SAASH,MAAM,CAAClB,EAAE,CAACoB,OAAO,KAAK,IAAI,IAAIF,MAAM,CAAClB,EAAE,CAACoB,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI;gBACvEpB,KAAK;YACP;YACAmB,IAAI5B,IAAI,CAAC2B,OAAOvC,KAAK,CAAC4B,GAAGP;YACzBO,IAAIP;QACN,OAAO,IAAIa,MAAMO,OAAO,KAAK,GAAG;YAC9BD,IAAI5B,IAAI,CAAC2B,OAAOvC,KAAK,CAAC4B,GAAGA,IAAI;YAC7BA,KAAK;QACP,OAAO;YACLA,KAAK;QACP;IACF;IACA,OAAOY;AACT;AAEA,gGAAgG;AAChG,4FAA4F;AAC5F,2DAA2D;AAC3D,SAASG,cAAcC,QAAiB,EAAElC,IAAc;QAKzCkC;IAJb,IAAIC,IAAI;IACR,MAAMC,QAAQF,QAAQ,CAAC,EAAE;IACzB,IAAIX,eAAea,QAAQD,IAAI;IAC/B,MAAOA,IAAID,SAASf,MAAM,IAAIe,QAAQ,CAACC,EAAE,CAACV,IAAI,KAAK,UAAUS,QAAQ,CAACC,EAAE,CAACT,OAAO,CAACtC,IAAI,OAAO,GAAI+C,KAAK;IACrG,IAAIA,IAAI,KAAKD,EAAAA,cAAAA,QAAQ,CAACC,EAAE,cAAXD,kCAAAA,YAAaT,IAAI,MAAK,aAAa,OAAOY,WAAWH,SAAS5C,KAAK,CAAC6C,IAAI,IAAInC;IACzF,OAAOqC,WAAWH,UAAUlC;AAC9B;AAEA,SAASsC,UAAUT,MAAe,EAAE7B,IAAc;IAChD,MAAMoC,QAAQP,MAAM,CAAC,EAAE;IACvB,IAAI,CAACO,OAAO,OAAO;IACnB,OAAQA,MAAMX,IAAI;QAChB,KAAK;QACL,KAAK;YAAkB;oBAESc;gBAD9B,MAAMA,SAASV,OAAOW,IAAI,CAAC,CAACC,IAAMA,EAAEhB,IAAI,KAAK;gBAC7C,OAAOc,SAASN,eAAcM,mBAAAA,OAAOL,QAAQ,cAAfK,8BAAAA,mBAAmB,EAAE,EAAEvC,QAAQ;YAC/D;QACA,KAAK;QACL,KAAK;YACH,yFAAyF;YACzF,OAAOD,gBAAgBqC,MAAMV,OAAO,CAAC3C,OAAO,CAAC,OAAO,KAAKiB;QAC3D,KAAK;YACH,OAAO0C,UAAUb,QAAQ7B;QAC3B,KAAK;QACL,KAAK;YACH,OAAO2C,cAAcd,QAAQ7B;QAC/B,KAAK;YACH,OAAO4C,SAASf,OAAOvC,KAAK,CAAC,GAAG,CAAC,IAAIU,MAAMjB,OAAO,CAACL,gBAAgB;QACrE,KAAK;YACH,gGAAgG;YAChG,OAAOG,UAAUuD,MAAMV,OAAO,CAAC3C,OAAO,CAAC,OAAO;QAChD,KAAK;YACH,OAAO6D,SAASf,OAAOvC,KAAK,CAAC,GAAG,CAAC,IAAIU;QACvC;YACE,OAAO;IACX;AACF;AAEA,yFAAyF;AACzF,+EAA+E;AAC/E,SAAS4C,SAASf,MAAe,EAAE7B,IAAc;IAC/C,OAAO4B,eAAeC,QACnBgB,GAAG,CAAC,CAACC,IAAMR,UAAUQ,GAAG9C,OACxB+C,MAAM,CAAC,CAACrC,IAAMA,EAAES,MAAM,GAAG,GACzB6B,IAAI,CAAC;AACV;AAEA,iGAAiG;AACjG,iGAAiG;AACjG,mFAAmF;AACnF,SAASC,kBAAkBC,IAAa;IACtC,IAAK,IAAIhC,IAAI,GAAGA,IAAIgC,KAAK/B,MAAM,EAAED,IAAK;YAGJM;QAFhC,MAAMA,QAAQ0B,IAAI,CAAChC,EAAE;QACrB,8DAA8D;QAC9D,IAAIM,MAAMC,IAAI,KAAK,YAAY,GAACD,kBAAAA,MAAMU,QAAQ,cAAdV,sCAAAA,gBAAgBL,MAAM,GAAE;QACxD,MAAMgC,SAAS3B,MAAMU,QAAQ,CAAC,EAAE;QAChC,IAAI,CAACX,eAAeC,MAAMU,QAAQ,CAAC,EAAE,KAAKiB,CAAAA,mBAAAA,6BAAAA,OAAQ1B,IAAI,MAAK,UAAU,CAAC0B,OAAOzB,OAAO,CAAC0B,UAAU,CAAC,MAAM,OAAOF;QAC7G,MAAMhB,WAAWV,MAAMU,QAAQ,CAAC5C,KAAK;QACrC4C,QAAQ,CAAC,EAAE,GAAG;YAAE,GAAGiB,MAAM;YAAEzB,SAASyB,OAAOzB,OAAO,CAACpC,KAAK,CAAC;QAAG;QAC5D,MAAM+D,OAAOH,KAAK5D,KAAK;QACvB+D,IAAI,CAACnC,EAAE,GAAG;YAAE,GAAGM,KAAK;YAAEU;QAAS;QAC/B,OAAOmB;IACT;IACA,OAAOH;AACT;AAEA,SAASP,cAAcd,MAAe,EAAE7B,IAAc;IACpD,MAAMsD,QAAkB,EAAE;IAC1B,KAAK,MAAMJ,QAAQtB,eAAeC,OAAOvC,KAAK,CAAC,GAAG,CAAC,IAAK;QACtDgE,MAAMpD,IAAI,CAAC0C,SAASK,kBAAkBC,MAAM5D,KAAK,CAAC,GAAG,CAAC,IAAIU;IAC5D;IACA,OAAOsD,MAAMP,MAAM,CAAC,CAACrC,IAAMA,EAAES,MAAM,GAAG,GAAG6B,IAAI,CAAC;AAChD;AAEA,qFAAqF;AACrF,SAASN,UAAUb,MAAe,EAAE7B,IAAc;IAChD,MAAMuD,OAAiB,EAAE;IACzB,IAAIC,QAAkB,EAAE;IACxB,KAAK,MAAMhC,SAASK,OAAQ;YAE8BL;QADxD,IAAIA,MAAMC,IAAI,KAAK,WAAW+B,QAAQ,EAAE;aACnC,IAAIhC,MAAMC,IAAI,KAAK,UAAU+B,MAAMtD,IAAI,CAACmC,YAAWb,kBAAAA,MAAMU,QAAQ,cAAdV,6BAAAA,kBAAkB,EAAE,EAAExB;aACzE,IAAIwB,MAAMC,IAAI,KAAK,YAAY8B,KAAKrD,IAAI,CAACsD,MAAMR,IAAI,CAAC;IAC3D;IACA,OAAOO,KAAKP,IAAI,CAAC;AACnB;AAEA,6FAA6F;AAC7F,oEAAoE;AACpE,SAASX,WAAWR,MAAe,EAAE7B,IAAc,EAAEyD,QAAQ,KAAK;IAChE,IAAI3B,MAAM;IACV,IAAK,IAAIZ,IAAI,GAAGA,IAAIW,OAAOV,MAAM,EAAED,IAAK;QACtC,MAAMM,QAAQK,MAAM,CAACX,EAAE;QACvB,IAAIM,MAAMC,IAAI,KAAK,aAAa;YAC9B,IAAIiC,OAAO;YACX,IAAI/C,IAAIO,IAAI;YACZ,IAAIc,QAAQ;YACZ,MAAOrB,IAAIkB,OAAOV,MAAM,IAAIa,QAAQ,EAAG;gBACrC,IAAIH,MAAM,CAAClB,EAAE,CAACc,IAAI,KAAK,aAAaO,SAAS;qBACxC,IAAIH,MAAM,CAAClB,EAAE,CAACc,IAAI,KAAK,cAAcO,SAAS;gBACnDrB,KAAK;YACP;YACA,MAAMd,OAAOwC,WAAWR,OAAOvC,KAAK,CAAC4B,IAAI,GAAGP,IAAI,IAAIX,MAAMyD;YAC1D,MAAME,SAASnC,MAAMmC,MAAM;YAC3B,MAAMC,OAAO/B,MAAM,CAAClB,EAAE;YACtB,IAAI8C,OAAO;gBACT,4FAA4F;gBAC5F3B,OAAOjC;YACT,OAAO,IAAI8D,WAAW,aAAavC,SAASvB,OAAO;gBACjD,oFAAoF;gBACpF,qFAAqF;gBACrF,mFAAmF;gBACnF,6EAA6E;gBAC7E,MAAMgE,OAAOhC,MAAM,CAACX,IAAI,EAAE;gBAC1B,MAAM4C,YAAYD,CAAAA,iBAAAA,2BAAAA,KAAMpC,IAAI,MAAK,UAAUoC,KAAKnC,OAAO,CAACqC,QAAQ,CAAC,QAAQH,CAAAA,iBAAAA,2BAAAA,KAAMnC,IAAI,MAAK,UAAUmC,KAAKlC,OAAO,CAAC0B,UAAU,CAAC;gBAC1H,MAAMY,OAAOJ,CAAAA,iBAAAA,2BAAAA,KAAMnC,IAAI,MAAK,SAAS5B,OAAO+D,KAAKlC,OAAO,GAAG7B;gBAC3D,MAAMsC,IAAI2B,YAAYjE,KAAKsB,MAAM,GAAGJ,eAAeiD;gBACnDlC,OAAOkC,KAAK1E,KAAK,CAAC6C,GAAGpD,OAAO,CAAC4C,aAAa,CAACsC,IAAM5F,SAAS6F,KAAK,CAACC,WAAW,CAACF;gBAC5E,0FAA0F;gBAC1F,IAAIL,CAAAA,iBAAAA,2BAAAA,KAAMnC,IAAI,MAAK,QAAQiC,OAAO;YACpC,OAAO,IAAIC,WAAW,aAAatC,WAAWxB,OAAO;YACnD,4EAA4E;YAC9E,OAAO,IAAI8D,WAAW,YAAY;YAChC,4FAA4F;YAC9F,OAAO;gBACL7B,OAAOjC,MAAM,6DAA6D;YAC5E;YACAqB,IAAIP,IAAI,IAAI+C;QACd,OAAO,IAAIlC,MAAMC,IAAI,KAAK,QAAQ;YAChCK,OAAON,MAAME,OAAO;QACtB,OAAO,IAAIF,MAAMC,IAAI,KAAK,aAAa;YACrCK,OAAO;QACT,OAAO,IAAIN,MAAMC,IAAI,KAAK,aAAa;YACrCK,OAAO;QACT,OAAO,IAAIN,MAAMC,IAAI,KAAK,eAAe;YACvCK,OAAO/B,gBAAgByB,MAAME,OAAO,EAAE1B;QACxC,OAAO,IAAIwB,MAAMC,IAAI,KAAK,eAAe;YACvCK,OAAOjD,UAAU2C,MAAME,OAAO;QAChC,OAAO,IAAIF,MAAMC,IAAI,KAAK,SAAS;gBACfD;YAAlBM,OAAOO,YAAWb,kBAAAA,MAAMU,QAAQ,cAAdV,6BAAAA,kBAAkB,EAAE,EAAExB,MAAM;QAChD;IACA,2EAA2E;IAC7E;IACA,OAAO8B;AACT;AAEA,MAAMsC,mBAAmB;AAEzB,kGAAkG;AAClG,yEAAyE;AACzE,OAAO,SAASC,YAAYC,KAAY;IACtC,OAAOC,aAAa;QAACD;KAAM;AAC7B;AAEA,iGAAiG;AACjG,wFAAwF;AACxF,OAAO,SAASC,aAAaC,MAAe;IAC1C,MAAMxE,OAAiB,EAAE;IACzB,MAAMyE,SAASD,OACZ3B,GAAG,CAAC,CAACC,IAAMR,UAAUQ,EAAE4B,IAAI,EAAE1E,OAC7B+C,MAAM,CAAC,CAACrC,IAAMA,EAAES,MAAM,GAAG,GACzB6B,IAAI,CAAC;IACR,MAAM2B,WAAW/E,cAAc6E;IAC/B,OAAOzE,KAAKmB,MAAM,KAAK,IAAIwD,WAAWA,SAAS5F,OAAO,CAACqF,kBAAkB,CAACtE,GAAGoB,IAAclB,IAAI,CAAC4E,OAAO1D,GAAG;AAC5G"}

package/dist/esm/chunk/group.js

@@ -105,7 +105,7 @@ const ATOMIC_TYPES = new Set([
 // Atomic (code/table/list) pieces are always a raw line slice: re-parsing a table's later pieces
 // without their header/delimiter rows would demote them to paragraph text.
 function piece(pieceLines, startLine, endLine, blockType, textMode) {
-    const text = ATOMIC_TYPES.has(blockType) || textMode === 'raw' ? pieceLines.join('\n') : parse(pieceLines.join('\n')).map((b)=>extractText(b.node)).join('\n');
+    const text = ATOMIC_TYPES.has(blockType) || textMode === 'raw' ? pieceLines.join('\n') : parse(pieceLines.join('\n')).map((b)=>extractText(b)).join('\n');
     return {
         startLine,
         endLine,
@@ -191,7 +191,7 @@ function groupScope(scopeBlocks, lines, resolved) {
             }
             continue;
         }
-        const extracted = extractText(block.node);
+        const extracted = extractText(block);
         const sizeText = resolved.text === 'raw' ? raw : extracted;
         if (block.type === 'heading') {
             addPart(extracted, sizeText, block.startLine, block.endLine);

package/dist/esm/chunk/group.js.map

@@ -1 +1 @@
-{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/chunk/group.ts"],"sourcesContent":["import { extractText } from './extract.ts';\nimport { parse } from './parse.ts';\nimport { DEFAULT_TARGET_TOKENS, estimateTokens } from './tokens.ts';\nimport type { Block, BlockType, Chunk, ChunkOptions } from './types.ts';\n\nconst PGC_GROUP_SIZE = 2;\nconst OVERSIZE_TRIGGER_MULTIPLE = 2;\n\ninterface ResolvedOptions {\n  targetTokens: number;\n  text: 'extracted' | 'raw';\n}\n\nfunction resolveOptions(opts?: ChunkOptions): ResolvedOptions {\n  return {\n    targetTokens: opts?.targetTokens ?? DEFAULT_TARGET_TOKENS,\n    text: opts?.text ?? 'raw',\n  };\n}\n\n// A heading of any depth ends the current scope and starts a new one (D1); the heading block\n// itself is carried into the new scope, where it joins that scope's first group (F7/F10).\nfunction splitScopes(blocks: Block[]): Block[][] {\n  const scopes: Block[][] = [];\n  let current: Block[] = [];\n  for (const block of blocks) {\n    if (block.type === 'heading' && current.length > 0) {\n      scopes.push(current);\n      current = [];\n    }\n    current.push(block);\n  }\n  if (current.length > 0) scopes.push(current);\n  return scopes;\n}\n\ninterface Part {\n  startLine: number;\n  endLine: number;\n  text: string;\n  // True for a sub-line split piece: its text is already the final slice, not the whole line --\n  // group()'s extent-based raw re-slice must not touch it (siblings share startLine === endLine).\n  final?: boolean;\n}\n\nfunction finalize(parts: Part[]): (Chunk & { final?: boolean }) | undefined {\n  if (parts.length === 0) return undefined;\n  const first = parts[0];\n  const last = parts[parts.length - 1];\n  return { startLine: first.startLine, endLine: last.endLine, text: parts.map((p) => p.text).join('\\n'), final: parts.some((p) => p.final) };\n}\n\nconst NEWLINE_TOKENS = estimateTokens('\\n');\n// Built on first use and kept: each construction is ~3.5 ms, and only an oversize block is ever\n// split, so no command pays for a segmenter it never reaches.\nconst SEGMENTERS = new Map<string, Intl.Segmenter>();\n\nfunction segmentsOf(text: string, granularity: 'sentence' | 'word'): string[] {\n  let segmenter = SEGMENTERS.get(granularity);\n  if (!segmenter) {\n    segmenter = new Intl.Segmenter(undefined, { granularity });\n    SEGMENTERS.set(granularity, segmenter);\n  }\n  return Array.from(segmenter.segment(text), (s) => s.segment);\n}\n\n// Greedily packs segments (already contiguous, tiling the source text with no gaps) into groups\n// of at most `working` estimated tokens; a lone segment over `working` still stands alone.\nfunction pack(segments: string[], working: number): string[] {\n  const groups: string[] = [];\n  let current = '';\n  let tokens = 0;\n  for (const segment of segments) {\n    const segmentTokens = estimateTokens(segment);\n    if (current.length > 0 && tokens + segmentTokens > working) {\n      groups.push(current);\n      current = '';\n      tokens = 0;\n    }\n    current += segment;\n    tokens += segmentTokens;\n  }\n  if (current.length > 0) groups.push(current);\n  return groups;\n}\n\n// Line-split alone can't shrink a lone dense line (the CJK case): falls back to sentence then\n// word boundaries (Intl.Segmenter, the same grapheme-safe engine as segment.ts), mode-agnostic on `text`.\nfunction splitLineText(text: string, working: number): string[] {\n  const sentences = segmentsOf(text, 'sentence');\n  const out: string[] = [];\n  let current = '';\n  let tokens = 0;\n  const flush = () => {\n    if (current.length > 0) {\n      out.push(current);\n      current = '';\n      tokens = 0;\n    }\n  };\n  for (const sentence of sentences) {\n    const sentenceTokens = estimateTokens(sentence);\n    if (sentenceTokens > working) {\n      flush();\n      out.push(...pack(segmentsOf(sentence, 'word'), working));\n      continue;\n    }\n    if (current.length > 0 && tokens + sentenceTokens > working) flush();\n    current += sentence;\n    tokens += sentenceTokens;\n  }\n  flush();\n  return out;\n}\n\nconst ATOMIC_TYPES: ReadonlySet<BlockType> = new Set(['code', 'table', 'list']);\n\n// Atomic (code/table/list) pieces are always a raw line slice: re-parsing a table's later pieces\n// without their header/delimiter rows would demote them to paragraph text.\nfunction piece(pieceLines: string[], startLine: number, endLine: number, blockType: BlockType, textMode: 'extracted' | 'raw'): Part {\n  const text =\n    ATOMIC_TYPES.has(blockType) || textMode === 'raw'\n      ? pieceLines.join('\\n')\n      : parse(pieceLines.join('\\n'))\n          .map((b) => extractText(b.node))\n          .join('\\n');\n  return { startLine, endLine, text };\n}\n\n// A one-line piece over working can't shrink via another line-boundary pass (rule 5's gap), so it\n// splits at sentence/word boundaries instead; `final` stops group() re-deriving its text by extent (F5).\nfunction finalizePiece(pieceLines: string[], startLine: number, endLine: number, working: number, blockType: BlockType, textMode: 'extracted' | 'raw'): Part[] {\n  const p = piece(pieceLines, startLine, endLine, blockType, textMode);\n  if (pieceLines.length === 1 && estimateTokens(p.text) > working) {\n    return splitLineText(p.text, working).map((text) => ({ startLine, endLine, text, final: true }));\n  }\n  return [p];\n}\n\n// A block over 2x working size splits at line boundaries into pieces each <= working size, never\n// mid-line. `seed`: pending tokens (e.g. a heading) the first piece must join, checked against the limit.\nfunction splitOversizeBlock(lines: string[], startLine: number, endLine: number, working: number, blockType: BlockType, textMode: 'extracted' | 'raw', seed = 0): Part[] {\n  const pieces: Part[] = [];\n  let pieceLines: string[] = [];\n  let pieceStart = startLine;\n  let tokens = seed;\n  for (let line = startLine; line <= endLine; line++) {\n    const lineText = lines[line - 1];\n    const sep = pieceLines.length > 0 || tokens > 0 ? NEWLINE_TOKENS : 0;\n    const lineTokens = estimateTokens(lineText);\n    if (pieceLines.length > 0 && tokens + sep + lineTokens > working) {\n      pieces.push(...finalizePiece(pieceLines, pieceStart, line - 1, working, blockType, textMode));\n      pieceLines = [];\n      tokens = 0;\n      pieceStart = line;\n      pieceLines.push(lineText);\n      tokens += lineTokens;\n      continue;\n    }\n    pieceLines.push(lineText);\n    tokens += sep + lineTokens;\n  }\n  if (pieceLines.length > 0) pieces.push(...finalizePiece(pieceLines, pieceStart, endLine, working, blockType, textMode));\n  return pieces;\n}\n\n// One heading scope's groups (D1): a heading opens the first group, and an oversize block\n// (rule 5, including an oversize heading) splits into pieces that each close their own group.\nfunction groupScope(scopeBlocks: Block[], lines: string[], resolved: ResolvedOptions): (Chunk & { final?: boolean })[] {\n  const working = resolved.targetTokens;\n  const trigger = working * OVERSIZE_TRIGGER_MULTIPLE;\n  const finished: (Chunk & { final?: boolean })[] = [];\n  let parts: Part[] = [];\n  let paragraphCount = 0;\n  let tokens = 0;\n\n  function close(): void {\n    const group = finalize(parts);\n    if (group) finished.push(group);\n    parts = [];\n    paragraphCount = 0;\n    tokens = 0;\n  }\n\n  // tokens tracks the active text mode's own estimate (a newline between parts costs\n  // NEWLINE_TOKENS too), so packing decisions size the text the chunk will actually ship as.\n  function addPart(text: string, sizeText: string, startLine: number, endLine: number): void {\n    tokens += (parts.length > 0 ? NEWLINE_TOKENS : 0) + estimateTokens(sizeText);\n    parts.push({ startLine, endLine, text });\n  }\n\n  for (const block of scopeBlocks) {\n    const raw = lines.slice(block.startLine - 1, block.endLine).join('\\n');\n    const blockTokens = estimateTokens(raw);\n\n    if (blockTokens > trigger) {\n      const seed = parts.length > 0 ? tokens : 0;\n      for (const p of splitOversizeBlock(lines, block.startLine, block.endLine, working, block.type, resolved.text, seed)) {\n        parts.push(p);\n        close();\n      }\n      continue;\n    }\n\n    const extracted = extractText(block.node);\n    const sizeText = resolved.text === 'raw' ? raw : extracted;\n\n    if (block.type === 'heading') {\n      addPart(extracted, sizeText, block.startLine, block.endLine);\n      continue;\n    }\n\n    // The 2x-working invariant holds even under pgc's paper-faithful 2-paragraph pairing --\n    // close first if the pair about to form would cross it.\n    const pairOversize = parts.length > 0 && tokens + NEWLINE_TOKENS + blockTokens > trigger;\n    if (pairOversize) close();\n\n    addPart(extracted, sizeText, block.startLine, block.endLine);\n    paragraphCount++;\n\n    if (paragraphCount >= PGC_GROUP_SIZE) close();\n  }\n  close();\n\n  return finished;\n}\n\n// Groups already-parsed blocks per opts (D1/D3), against the same body the blocks were parsed\n// from (line lookups for oversize splitting).\nexport function group(blocks: Block[], body: string, opts?: ChunkOptions): Chunk[] {\n  const resolved = resolveOptions(opts);\n  const lines = body.split('\\n');\n  const chunks: (Chunk & { final?: boolean })[] = [];\n  for (const scope of splitScopes(blocks)) chunks.push(...groupScope(scope, lines, resolved));\n  // 'raw': the chunk's own source lines verbatim, replacing the flavor-resolved join above (D9).\n  // A `final` chunk already carries its own slice's raw text; re-slicing by extent would return the whole shared line.\n  const texted =\n    resolved.text === 'raw'\n      ? chunks.map((c) =>\n          c.final\n            ? c\n            : {\n                ...c,\n                text: lines\n                  .slice(c.startLine - 1, c.endLine)\n                  .join('\\n')\n                  .trim(),\n              }\n        )\n      : chunks;\n  // A group can be all-blank (flavor-stripped to nothing, or a raw slice of pure syntax); it never produces a chunk.\n  return texted.filter((c) => c.text.trim().length > 0).map((c) => ({ startLine: c.startLine, endLine: c.endLine, text: c.text }));\n}\n"],"names":["extractText","parse","DEFAULT_TARGET_TOKENS","estimateTokens","PGC_GROUP_SIZE","OVERSIZE_TRIGGER_MULTIPLE","resolveOptions","opts","targetTokens","text","splitScopes","blocks","scopes","current","block","type","length","push","finalize","parts","undefined","first","last","startLine","endLine","map","p","join","final","some","NEWLINE_TOKENS","SEGMENTERS","Map","segmentsOf","granularity","segmenter","get","Intl","Segmenter","set","Array","from","segment","s","pack","segments","working","groups","tokens","segmentTokens","splitLineText","sentences","out","flush","sentence","sentenceTokens","ATOMIC_TYPES","Set","piece","pieceLines","blockType","textMode","has","b","node","finalizePiece","splitOversizeBlock","lines","seed","pieces","pieceStart","line","lineText","sep","lineTokens","groupScope","scopeBlocks","resolved","trigger","finished","paragraphCount","close","group","addPart","sizeText","raw","slice","blockTokens","extracted","pairOversize","body","split","chunks","scope","texted","c","trim","filter"],"mappings":"AAAA,SAASA,WAAW,QAAQ,eAAe;AAC3C,SAASC,KAAK,QAAQ,aAAa;AACnC,SAASC,qBAAqB,EAAEC,cAAc,QAAQ,cAAc;AAGpE,MAAMC,iBAAiB;AACvB,MAAMC,4BAA4B;AAOlC,SAASC,eAAeC,IAAmB;;IACzC,OAAO;QACLC,YAAY,UAAED,iBAAAA,2BAAAA,KAAMC,YAAY,uCAAIN;QACpCO,IAAI,WAAEF,iBAAAA,2BAAAA,KAAME,IAAI,yCAAI;IACtB;AACF;AAEA,6FAA6F;AAC7F,0FAA0F;AAC1F,SAASC,YAAYC,MAAe;IAClC,MAAMC,SAAoB,EAAE;IAC5B,IAAIC,UAAmB,EAAE;IACzB,KAAK,MAAMC,SAASH,OAAQ;QAC1B,IAAIG,MAAMC,IAAI,KAAK,aAAaF,QAAQG,MAAM,GAAG,GAAG;YAClDJ,OAAOK,IAAI,CAACJ;YACZA,UAAU,EAAE;QACd;QACAA,QAAQI,IAAI,CAACH;IACf;IACA,IAAID,QAAQG,MAAM,GAAG,GAAGJ,OAAOK,IAAI,CAACJ;IACpC,OAAOD;AACT;AAWA,SAASM,SAASC,KAAa;IAC7B,IAAIA,MAAMH,MAAM,KAAK,GAAG,OAAOI;IAC/B,MAAMC,QAAQF,KAAK,CAAC,EAAE;IACtB,MAAMG,OAAOH,KAAK,CAACA,MAAMH,MAAM,GAAG,EAAE;IACpC,OAAO;QAAEO,WAAWF,MAAME,SAAS;QAAEC,SAASF,KAAKE,OAAO;QAAEf,MAAMU,MAAMM,GAAG,CAAC,CAACC,IAAMA,EAAEjB,IAAI,EAAEkB,IAAI,CAAC;QAAOC,OAAOT,MAAMU,IAAI,CAAC,CAACH,IAAMA,EAAEE,KAAK;IAAE;AAC3I;AAEA,MAAME,iBAAiB3B,eAAe;AACtC,gGAAgG;AAChG,8DAA8D;AAC9D,MAAM4B,aAAa,IAAIC;AAEvB,SAASC,WAAWxB,IAAY,EAAEyB,WAAgC;IAChE,IAAIC,YAAYJ,WAAWK,GAAG,CAACF;IAC/B,IAAI,CAACC,WAAW;QACdA,YAAY,IAAIE,KAAKC,SAAS,CAAClB,WAAW;YAAEc;QAAY;QACxDH,WAAWQ,GAAG,CAACL,aAAaC;IAC9B;IACA,OAAOK,MAAMC,IAAI,CAACN,UAAUO,OAAO,CAACjC,OAAO,CAACkC,IAAMA,EAAED,OAAO;AAC7D;AAEA,gGAAgG;AAChG,2FAA2F;AAC3F,SAASE,KAAKC,QAAkB,EAAEC,OAAe;IAC/C,MAAMC,SAAmB,EAAE;IAC3B,IAAIlC,UAAU;IACd,IAAImC,SAAS;IACb,KAAK,MAAMN,WAAWG,SAAU;QAC9B,MAAMI,gBAAgB9C,eAAeuC;QACrC,IAAI7B,QAAQG,MAAM,GAAG,KAAKgC,SAASC,gBAAgBH,SAAS;YAC1DC,OAAO9B,IAAI,CAACJ;YACZA,UAAU;YACVmC,SAAS;QACX;QACAnC,WAAW6B;QACXM,UAAUC;IACZ;IACA,IAAIpC,QAAQG,MAAM,GAAG,GAAG+B,OAAO9B,IAAI,CAACJ;IACpC,OAAOkC;AACT;AAEA,8FAA8F;AAC9F,0GAA0G;AAC1G,SAASG,cAAczC,IAAY,EAAEqC,OAAe;IAClD,MAAMK,YAAYlB,WAAWxB,MAAM;IACnC,MAAM2C,MAAgB,EAAE;IACxB,IAAIvC,UAAU;IACd,IAAImC,SAAS;IACb,MAAMK,QAAQ;QACZ,IAAIxC,QAAQG,MAAM,GAAG,GAAG;YACtBoC,IAAInC,IAAI,CAACJ;YACTA,UAAU;YACVmC,SAAS;QACX;IACF;IACA,KAAK,MAAMM,YAAYH,UAAW;QAChC,MAAMI,iBAAiBpD,eAAemD;QACtC,IAAIC,iBAAiBT,SAAS;YAC5BO;YACAD,IAAInC,IAAI,IAAI2B,KAAKX,WAAWqB,UAAU,SAASR;YAC/C;QACF;QACA,IAAIjC,QAAQG,MAAM,GAAG,KAAKgC,SAASO,iBAAiBT,SAASO;QAC7DxC,WAAWyC;QACXN,UAAUO;IACZ;IACAF;IACA,OAAOD;AACT;AAEA,MAAMI,eAAuC,IAAIC,IAAI;IAAC;IAAQ;IAAS;CAAO;AAE9E,iGAAiG;AACjG,2EAA2E;AAC3E,SAASC,MAAMC,UAAoB,EAAEpC,SAAiB,EAAEC,OAAe,EAAEoC,SAAoB,EAAEC,QAA6B;IAC1H,MAAMpD,OACJ+C,aAAaM,GAAG,CAACF,cAAcC,aAAa,QACxCF,WAAWhC,IAAI,CAAC,QAChB1B,MAAM0D,WAAWhC,IAAI,CAAC,OACnBF,GAAG,CAAC,CAACsC,IAAM/D,YAAY+D,EAAEC,IAAI,GAC7BrC,IAAI,CAAC;IACd,OAAO;QAAEJ;QAAWC;QAASf;IAAK;AACpC;AAEA,kGAAkG;AAClG,yGAAyG;AACzG,SAASwD,cAAcN,UAAoB,EAAEpC,SAAiB,EAAEC,OAAe,EAAEsB,OAAe,EAAEc,SAAoB,EAAEC,QAA6B;IACnJ,MAAMnC,IAAIgC,MAAMC,YAAYpC,WAAWC,SAASoC,WAAWC;IAC3D,IAAIF,WAAW3C,MAAM,KAAK,KAAKb,eAAeuB,EAAEjB,IAAI,IAAIqC,SAAS;QAC/D,OAAOI,cAAcxB,EAAEjB,IAAI,EAAEqC,SAASrB,GAAG,CAAC,CAAChB,OAAU,CAAA;gBAAEc;gBAAWC;gBAASf;gBAAMmB,OAAO;YAAK,CAAA;IAC/F;IACA,OAAO;QAACF;KAAE;AACZ;AAEA,iGAAiG;AACjG,0GAA0G;AAC1G,SAASwC,mBAAmBC,KAAe,EAAE5C,SAAiB,EAAEC,OAAe,EAAEsB,OAAe,EAAEc,SAAoB,EAAEC,QAA6B,EAAEO,OAAO,CAAC;IAC7J,MAAMC,SAAiB,EAAE;IACzB,IAAIV,aAAuB,EAAE;IAC7B,IAAIW,aAAa/C;IACjB,IAAIyB,SAASoB;IACb,IAAK,IAAIG,OAAOhD,WAAWgD,QAAQ/C,SAAS+C,OAAQ;QAClD,MAAMC,WAAWL,KAAK,CAACI,OAAO,EAAE;QAChC,MAAME,MAAMd,WAAW3C,MAAM,GAAG,KAAKgC,SAAS,IAAIlB,iBAAiB;QACnE,MAAM4C,aAAavE,eAAeqE;QAClC,IAAIb,WAAW3C,MAAM,GAAG,KAAKgC,SAASyB,MAAMC,aAAa5B,SAAS;YAChEuB,OAAOpD,IAAI,IAAIgD,cAAcN,YAAYW,YAAYC,OAAO,GAAGzB,SAASc,WAAWC;YACnFF,aAAa,EAAE;YACfX,SAAS;YACTsB,aAAaC;YACbZ,WAAW1C,IAAI,CAACuD;YAChBxB,UAAU0B;YACV;QACF;QACAf,WAAW1C,IAAI,CAACuD;QAChBxB,UAAUyB,MAAMC;IAClB;IACA,IAAIf,WAAW3C,MAAM,GAAG,GAAGqD,OAAOpD,IAAI,IAAIgD,cAAcN,YAAYW,YAAY9C,SAASsB,SAASc,WAAWC;IAC7G,OAAOQ;AACT;AAEA,0FAA0F;AAC1F,8FAA8F;AAC9F,SAASM,WAAWC,WAAoB,EAAET,KAAe,EAAEU,QAAyB;IAClF,MAAM/B,UAAU+B,SAASrE,YAAY;IACrC,MAAMsE,UAAUhC,UAAUzC;IAC1B,MAAM0E,WAA4C,EAAE;IACpD,IAAI5D,QAAgB,EAAE;IACtB,IAAI6D,iBAAiB;IACrB,IAAIhC,SAAS;IAEb,SAASiC;QACP,MAAMC,QAAQhE,SAASC;QACvB,IAAI+D,OAAOH,SAAS9D,IAAI,CAACiE;QACzB/D,QAAQ,EAAE;QACV6D,iBAAiB;QACjBhC,SAAS;IACX;IAEA,mFAAmF;IACnF,2FAA2F;IAC3F,SAASmC,QAAQ1E,IAAY,EAAE2E,QAAgB,EAAE7D,SAAiB,EAAEC,OAAe;QACjFwB,UAAU,AAAC7B,CAAAA,MAAMH,MAAM,GAAG,IAAIc,iBAAiB,CAAA,IAAK3B,eAAeiF;QACnEjE,MAAMF,IAAI,CAAC;YAAEM;YAAWC;YAASf;QAAK;IACxC;IAEA,KAAK,MAAMK,SAAS8D,YAAa;QAC/B,MAAMS,MAAMlB,MAAMmB,KAAK,CAACxE,MAAMS,SAAS,GAAG,GAAGT,MAAMU,OAAO,EAAEG,IAAI,CAAC;QACjE,MAAM4D,cAAcpF,eAAekF;QAEnC,IAAIE,cAAcT,SAAS;YACzB,MAAMV,OAAOjD,MAAMH,MAAM,GAAG,IAAIgC,SAAS;YACzC,KAAK,MAAMtB,KAAKwC,mBAAmBC,OAAOrD,MAAMS,SAAS,EAAET,MAAMU,OAAO,EAAEsB,SAAShC,MAAMC,IAAI,EAAE8D,SAASpE,IAAI,EAAE2D,MAAO;gBACnHjD,MAAMF,IAAI,CAACS;gBACXuD;YACF;YACA;QACF;QAEA,MAAMO,YAAYxF,YAAYc,MAAMkD,IAAI;QACxC,MAAMoB,WAAWP,SAASpE,IAAI,KAAK,QAAQ4E,MAAMG;QAEjD,IAAI1E,MAAMC,IAAI,KAAK,WAAW;YAC5BoE,QAAQK,WAAWJ,UAAUtE,MAAMS,SAAS,EAAET,MAAMU,OAAO;YAC3D;QACF;QAEA,wFAAwF;QACxF,wDAAwD;QACxD,MAAMiE,eAAetE,MAAMH,MAAM,GAAG,KAAKgC,SAASlB,iBAAiByD,cAAcT;QACjF,IAAIW,cAAcR;QAElBE,QAAQK,WAAWJ,UAAUtE,MAAMS,SAAS,EAAET,MAAMU,OAAO;QAC3DwD;QAEA,IAAIA,kBAAkB5E,gBAAgB6E;IACxC;IACAA;IAEA,OAAOF;AACT;AAEA,8FAA8F;AAC9F,8CAA8C;AAC9C,OAAO,SAASG,MAAMvE,MAAe,EAAE+E,IAAY,EAAEnF,IAAmB;IACtE,MAAMsE,WAAWvE,eAAeC;IAChC,MAAM4D,QAAQuB,KAAKC,KAAK,CAAC;IACzB,MAAMC,SAA0C,EAAE;IAClD,KAAK,MAAMC,SAASnF,YAAYC,QAASiF,OAAO3E,IAAI,IAAI0D,WAAWkB,OAAO1B,OAAOU;IACjF,+FAA+F;IAC/F,qHAAqH;IACrH,MAAMiB,SACJjB,SAASpE,IAAI,KAAK,QACdmF,OAAOnE,GAAG,CAAC,CAACsE,IACVA,EAAEnE,KAAK,GACHmE,IACA;YACE,GAAGA,CAAC;YACJtF,MAAM0D,MACHmB,KAAK,CAACS,EAAExE,SAAS,GAAG,GAAGwE,EAAEvE,OAAO,EAChCG,IAAI,CAAC,MACLqE,IAAI;QACT,KAENJ;IACN,mHAAmH;IACnH,OAAOE,OAAOG,MAAM,CAAC,CAACF,IAAMA,EAAEtF,IAAI,CAACuF,IAAI,GAAGhF,MAAM,GAAG,GAAGS,GAAG,CAAC,CAACsE,IAAO,CAAA;YAAExE,WAAWwE,EAAExE,SAAS;YAAEC,SAASuE,EAAEvE,OAAO;YAAEf,MAAMsF,EAAEtF,IAAI;QAAC,CAAA;AAC/H"}
+{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/chunk/group.ts"],"sourcesContent":["import { extractText } from './extract.ts';\nimport { parse } from './parse.ts';\nimport { DEFAULT_TARGET_TOKENS, estimateTokens } from './tokens.ts';\nimport type { Block, BlockType, Chunk, ChunkOptions } from './types.ts';\n\nconst PGC_GROUP_SIZE = 2;\nconst OVERSIZE_TRIGGER_MULTIPLE = 2;\n\ninterface ResolvedOptions {\n  targetTokens: number;\n  text: 'extracted' | 'raw';\n}\n\nfunction resolveOptions(opts?: ChunkOptions): ResolvedOptions {\n  return {\n    targetTokens: opts?.targetTokens ?? DEFAULT_TARGET_TOKENS,\n    text: opts?.text ?? 'raw',\n  };\n}\n\n// A heading of any depth ends the current scope and starts a new one (D1); the heading block\n// itself is carried into the new scope, where it joins that scope's first group (F7/F10).\nfunction splitScopes(blocks: Block[]): Block[][] {\n  const scopes: Block[][] = [];\n  let current: Block[] = [];\n  for (const block of blocks) {\n    if (block.type === 'heading' && current.length > 0) {\n      scopes.push(current);\n      current = [];\n    }\n    current.push(block);\n  }\n  if (current.length > 0) scopes.push(current);\n  return scopes;\n}\n\ninterface Part {\n  startLine: number;\n  endLine: number;\n  text: string;\n  // True for a sub-line split piece: its text is already the final slice, not the whole line --\n  // group()'s extent-based raw re-slice must not touch it (siblings share startLine === endLine).\n  final?: boolean;\n}\n\nfunction finalize(parts: Part[]): (Chunk & { final?: boolean }) | undefined {\n  if (parts.length === 0) return undefined;\n  const first = parts[0];\n  const last = parts[parts.length - 1];\n  return { startLine: first.startLine, endLine: last.endLine, text: parts.map((p) => p.text).join('\\n'), final: parts.some((p) => p.final) };\n}\n\nconst NEWLINE_TOKENS = estimateTokens('\\n');\n// Built on first use and kept: each construction is ~3.5 ms, and only an oversize block is ever\n// split, so no command pays for a segmenter it never reaches.\nconst SEGMENTERS = new Map<string, Intl.Segmenter>();\n\nfunction segmentsOf(text: string, granularity: 'sentence' | 'word'): string[] {\n  let segmenter = SEGMENTERS.get(granularity);\n  if (!segmenter) {\n    segmenter = new Intl.Segmenter(undefined, { granularity });\n    SEGMENTERS.set(granularity, segmenter);\n  }\n  return Array.from(segmenter.segment(text), (s) => s.segment);\n}\n\n// Greedily packs segments (already contiguous, tiling the source text with no gaps) into groups\n// of at most `working` estimated tokens; a lone segment over `working` still stands alone.\nfunction pack(segments: string[], working: number): string[] {\n  const groups: string[] = [];\n  let current = '';\n  let tokens = 0;\n  for (const segment of segments) {\n    const segmentTokens = estimateTokens(segment);\n    if (current.length > 0 && tokens + segmentTokens > working) {\n      groups.push(current);\n      current = '';\n      tokens = 0;\n    }\n    current += segment;\n    tokens += segmentTokens;\n  }\n  if (current.length > 0) groups.push(current);\n  return groups;\n}\n\n// Line-split alone can't shrink a lone dense line (the CJK case): falls back to sentence then\n// word boundaries (Intl.Segmenter, the same grapheme-safe engine as segment.ts), mode-agnostic on `text`.\nfunction splitLineText(text: string, working: number): string[] {\n  const sentences = segmentsOf(text, 'sentence');\n  const out: string[] = [];\n  let current = '';\n  let tokens = 0;\n  const flush = () => {\n    if (current.length > 0) {\n      out.push(current);\n      current = '';\n      tokens = 0;\n    }\n  };\n  for (const sentence of sentences) {\n    const sentenceTokens = estimateTokens(sentence);\n    if (sentenceTokens > working) {\n      flush();\n      out.push(...pack(segmentsOf(sentence, 'word'), working));\n      continue;\n    }\n    if (current.length > 0 && tokens + sentenceTokens > working) flush();\n    current += sentence;\n    tokens += sentenceTokens;\n  }\n  flush();\n  return out;\n}\n\nconst ATOMIC_TYPES: ReadonlySet<BlockType> = new Set(['code', 'table', 'list']);\n\n// Atomic (code/table/list) pieces are always a raw line slice: re-parsing a table's later pieces\n// without their header/delimiter rows would demote them to paragraph text.\nfunction piece(pieceLines: string[], startLine: number, endLine: number, blockType: BlockType, textMode: 'extracted' | 'raw'): Part {\n  const text =\n    ATOMIC_TYPES.has(blockType) || textMode === 'raw'\n      ? pieceLines.join('\\n')\n      : parse(pieceLines.join('\\n'))\n          .map((b) => extractText(b))\n          .join('\\n');\n  return { startLine, endLine, text };\n}\n\n// A one-line piece over working can't shrink via another line-boundary pass (rule 5's gap), so it\n// splits at sentence/word boundaries instead; `final` stops group() re-deriving its text by extent (F5).\nfunction finalizePiece(pieceLines: string[], startLine: number, endLine: number, working: number, blockType: BlockType, textMode: 'extracted' | 'raw'): Part[] {\n  const p = piece(pieceLines, startLine, endLine, blockType, textMode);\n  if (pieceLines.length === 1 && estimateTokens(p.text) > working) {\n    return splitLineText(p.text, working).map((text) => ({ startLine, endLine, text, final: true }));\n  }\n  return [p];\n}\n\n// A block over 2x working size splits at line boundaries into pieces each <= working size, never\n// mid-line. `seed`: pending tokens (e.g. a heading) the first piece must join, checked against the limit.\nfunction splitOversizeBlock(lines: string[], startLine: number, endLine: number, working: number, blockType: BlockType, textMode: 'extracted' | 'raw', seed = 0): Part[] {\n  const pieces: Part[] = [];\n  let pieceLines: string[] = [];\n  let pieceStart = startLine;\n  let tokens = seed;\n  for (let line = startLine; line <= endLine; line++) {\n    const lineText = lines[line - 1];\n    const sep = pieceLines.length > 0 || tokens > 0 ? NEWLINE_TOKENS : 0;\n    const lineTokens = estimateTokens(lineText);\n    if (pieceLines.length > 0 && tokens + sep + lineTokens > working) {\n      pieces.push(...finalizePiece(pieceLines, pieceStart, line - 1, working, blockType, textMode));\n      pieceLines = [];\n      tokens = 0;\n      pieceStart = line;\n      pieceLines.push(lineText);\n      tokens += lineTokens;\n      continue;\n    }\n    pieceLines.push(lineText);\n    tokens += sep + lineTokens;\n  }\n  if (pieceLines.length > 0) pieces.push(...finalizePiece(pieceLines, pieceStart, endLine, working, blockType, textMode));\n  return pieces;\n}\n\n// One heading scope's groups (D1): a heading opens the first group, and an oversize block\n// (rule 5, including an oversize heading) splits into pieces that each close their own group.\nfunction groupScope(scopeBlocks: Block[], lines: string[], resolved: ResolvedOptions): (Chunk & { final?: boolean })[] {\n  const working = resolved.targetTokens;\n  const trigger = working * OVERSIZE_TRIGGER_MULTIPLE;\n  const finished: (Chunk & { final?: boolean })[] = [];\n  let parts: Part[] = [];\n  let paragraphCount = 0;\n  let tokens = 0;\n\n  function close(): void {\n    const group = finalize(parts);\n    if (group) finished.push(group);\n    parts = [];\n    paragraphCount = 0;\n    tokens = 0;\n  }\n\n  // tokens tracks the active text mode's own estimate (a newline between parts costs\n  // NEWLINE_TOKENS too), so packing decisions size the text the chunk will actually ship as.\n  function addPart(text: string, sizeText: string, startLine: number, endLine: number): void {\n    tokens += (parts.length > 0 ? NEWLINE_TOKENS : 0) + estimateTokens(sizeText);\n    parts.push({ startLine, endLine, text });\n  }\n\n  for (const block of scopeBlocks) {\n    const raw = lines.slice(block.startLine - 1, block.endLine).join('\\n');\n    const blockTokens = estimateTokens(raw);\n\n    if (blockTokens > trigger) {\n      const seed = parts.length > 0 ? tokens : 0;\n      for (const p of splitOversizeBlock(lines, block.startLine, block.endLine, working, block.type, resolved.text, seed)) {\n        parts.push(p);\n        close();\n      }\n      continue;\n    }\n\n    const extracted = extractText(block);\n    const sizeText = resolved.text === 'raw' ? raw : extracted;\n\n    if (block.type === 'heading') {\n      addPart(extracted, sizeText, block.startLine, block.endLine);\n      continue;\n    }\n\n    // The 2x-working invariant holds even under pgc's paper-faithful 2-paragraph pairing --\n    // close first if the pair about to form would cross it.\n    const pairOversize = parts.length > 0 && tokens + NEWLINE_TOKENS + blockTokens > trigger;\n    if (pairOversize) close();\n\n    addPart(extracted, sizeText, block.startLine, block.endLine);\n    paragraphCount++;\n\n    if (paragraphCount >= PGC_GROUP_SIZE) close();\n  }\n  close();\n\n  return finished;\n}\n\n// Groups already-parsed blocks per opts (D1/D3), against the same body the blocks were parsed\n// from (line lookups for oversize splitting).\nexport function group(blocks: Block[], body: string, opts?: ChunkOptions): Chunk[] {\n  const resolved = resolveOptions(opts);\n  const lines = body.split('\\n');\n  const chunks: (Chunk & { final?: boolean })[] = [];\n  for (const scope of splitScopes(blocks)) chunks.push(...groupScope(scope, lines, resolved));\n  // 'raw': the chunk's own source lines verbatim, replacing the flavor-resolved join above (D9).\n  // A `final` chunk already carries its own slice's raw text; re-slicing by extent would return the whole shared line.\n  const texted =\n    resolved.text === 'raw'\n      ? chunks.map((c) =>\n          c.final\n            ? c\n            : {\n                ...c,\n                text: lines\n                  .slice(c.startLine - 1, c.endLine)\n                  .join('\\n')\n                  .trim(),\n              }\n        )\n      : chunks;\n  // A group can be all-blank (flavor-stripped to nothing, or a raw slice of pure syntax); it never produces a chunk.\n  return texted.filter((c) => c.text.trim().length > 0).map((c) => ({ startLine: c.startLine, endLine: c.endLine, text: c.text }));\n}\n"],"names":["extractText","parse","DEFAULT_TARGET_TOKENS","estimateTokens","PGC_GROUP_SIZE","OVERSIZE_TRIGGER_MULTIPLE","resolveOptions","opts","targetTokens","text","splitScopes","blocks","scopes","current","block","type","length","push","finalize","parts","undefined","first","last","startLine","endLine","map","p","join","final","some","NEWLINE_TOKENS","SEGMENTERS","Map","segmentsOf","granularity","segmenter","get","Intl","Segmenter","set","Array","from","segment","s","pack","segments","working","groups","tokens","segmentTokens","splitLineText","sentences","out","flush","sentence","sentenceTokens","ATOMIC_TYPES","Set","piece","pieceLines","blockType","textMode","has","b","finalizePiece","splitOversizeBlock","lines","seed","pieces","pieceStart","line","lineText","sep","lineTokens","groupScope","scopeBlocks","resolved","trigger","finished","paragraphCount","close","group","addPart","sizeText","raw","slice","blockTokens","extracted","pairOversize","body","split","chunks","scope","texted","c","trim","filter"],"mappings":"AAAA,SAASA,WAAW,QAAQ,eAAe;AAC3C,SAASC,KAAK,QAAQ,aAAa;AACnC,SAASC,qBAAqB,EAAEC,cAAc,QAAQ,cAAc;AAGpE,MAAMC,iBAAiB;AACvB,MAAMC,4BAA4B;AAOlC,SAASC,eAAeC,IAAmB;;IACzC,OAAO;QACLC,YAAY,UAAED,iBAAAA,2BAAAA,KAAMC,YAAY,uCAAIN;QACpCO,IAAI,WAAEF,iBAAAA,2BAAAA,KAAME,IAAI,yCAAI;IACtB;AACF;AAEA,6FAA6F;AAC7F,0FAA0F;AAC1F,SAASC,YAAYC,MAAe;IAClC,MAAMC,SAAoB,EAAE;IAC5B,IAAIC,UAAmB,EAAE;IACzB,KAAK,MAAMC,SAASH,OAAQ;QAC1B,IAAIG,MAAMC,IAAI,KAAK,aAAaF,QAAQG,MAAM,GAAG,GAAG;YAClDJ,OAAOK,IAAI,CAACJ;YACZA,UAAU,EAAE;QACd;QACAA,QAAQI,IAAI,CAACH;IACf;IACA,IAAID,QAAQG,MAAM,GAAG,GAAGJ,OAAOK,IAAI,CAACJ;IACpC,OAAOD;AACT;AAWA,SAASM,SAASC,KAAa;IAC7B,IAAIA,MAAMH,MAAM,KAAK,GAAG,OAAOI;IAC/B,MAAMC,QAAQF,KAAK,CAAC,EAAE;IACtB,MAAMG,OAAOH,KAAK,CAACA,MAAMH,MAAM,GAAG,EAAE;IACpC,OAAO;QAAEO,WAAWF,MAAME,SAAS;QAAEC,SAASF,KAAKE,OAAO;QAAEf,MAAMU,MAAMM,GAAG,CAAC,CAACC,IAAMA,EAAEjB,IAAI,EAAEkB,IAAI,CAAC;QAAOC,OAAOT,MAAMU,IAAI,CAAC,CAACH,IAAMA,EAAEE,KAAK;IAAE;AAC3I;AAEA,MAAME,iBAAiB3B,eAAe;AACtC,gGAAgG;AAChG,8DAA8D;AAC9D,MAAM4B,aAAa,IAAIC;AAEvB,SAASC,WAAWxB,IAAY,EAAEyB,WAAgC;IAChE,IAAIC,YAAYJ,WAAWK,GAAG,CAACF;IAC/B,IAAI,CAACC,WAAW;QACdA,YAAY,IAAIE,KAAKC,SAAS,CAAClB,WAAW;YAAEc;QAAY;QACxDH,WAAWQ,GAAG,CAACL,aAAaC;IAC9B;IACA,OAAOK,MAAMC,IAAI,CAACN,UAAUO,OAAO,CAACjC,OAAO,CAACkC,IAAMA,EAAED,OAAO;AAC7D;AAEA,gGAAgG;AAChG,2FAA2F;AAC3F,SAASE,KAAKC,QAAkB,EAAEC,OAAe;IAC/C,MAAMC,SAAmB,EAAE;IAC3B,IAAIlC,UAAU;IACd,IAAImC,SAAS;IACb,KAAK,MAAMN,WAAWG,SAAU;QAC9B,MAAMI,gBAAgB9C,eAAeuC;QACrC,IAAI7B,QAAQG,MAAM,GAAG,KAAKgC,SAASC,gBAAgBH,SAAS;YAC1DC,OAAO9B,IAAI,CAACJ;YACZA,UAAU;YACVmC,SAAS;QACX;QACAnC,WAAW6B;QACXM,UAAUC;IACZ;IACA,IAAIpC,QAAQG,MAAM,GAAG,GAAG+B,OAAO9B,IAAI,CAACJ;IACpC,OAAOkC;AACT;AAEA,8FAA8F;AAC9F,0GAA0G;AAC1G,SAASG,cAAczC,IAAY,EAAEqC,OAAe;IAClD,MAAMK,YAAYlB,WAAWxB,MAAM;IACnC,MAAM2C,MAAgB,EAAE;IACxB,IAAIvC,UAAU;IACd,IAAImC,SAAS;IACb,MAAMK,QAAQ;QACZ,IAAIxC,QAAQG,MAAM,GAAG,GAAG;YACtBoC,IAAInC,IAAI,CAACJ;YACTA,UAAU;YACVmC,SAAS;QACX;IACF;IACA,KAAK,MAAMM,YAAYH,UAAW;QAChC,MAAMI,iBAAiBpD,eAAemD;QACtC,IAAIC,iBAAiBT,SAAS;YAC5BO;YACAD,IAAInC,IAAI,IAAI2B,KAAKX,WAAWqB,UAAU,SAASR;YAC/C;QACF;QACA,IAAIjC,QAAQG,MAAM,GAAG,KAAKgC,SAASO,iBAAiBT,SAASO;QAC7DxC,WAAWyC;QACXN,UAAUO;IACZ;IACAF;IACA,OAAOD;AACT;AAEA,MAAMI,eAAuC,IAAIC,IAAI;IAAC;IAAQ;IAAS;CAAO;AAE9E,iGAAiG;AACjG,2EAA2E;AAC3E,SAASC,MAAMC,UAAoB,EAAEpC,SAAiB,EAAEC,OAAe,EAAEoC,SAAoB,EAAEC,QAA6B;IAC1H,MAAMpD,OACJ+C,aAAaM,GAAG,CAACF,cAAcC,aAAa,QACxCF,WAAWhC,IAAI,CAAC,QAChB1B,MAAM0D,WAAWhC,IAAI,CAAC,OACnBF,GAAG,CAAC,CAACsC,IAAM/D,YAAY+D,IACvBpC,IAAI,CAAC;IACd,OAAO;QAAEJ;QAAWC;QAASf;IAAK;AACpC;AAEA,kGAAkG;AAClG,yGAAyG;AACzG,SAASuD,cAAcL,UAAoB,EAAEpC,SAAiB,EAAEC,OAAe,EAAEsB,OAAe,EAAEc,SAAoB,EAAEC,QAA6B;IACnJ,MAAMnC,IAAIgC,MAAMC,YAAYpC,WAAWC,SAASoC,WAAWC;IAC3D,IAAIF,WAAW3C,MAAM,KAAK,KAAKb,eAAeuB,EAAEjB,IAAI,IAAIqC,SAAS;QAC/D,OAAOI,cAAcxB,EAAEjB,IAAI,EAAEqC,SAASrB,GAAG,CAAC,CAAChB,OAAU,CAAA;gBAAEc;gBAAWC;gBAASf;gBAAMmB,OAAO;YAAK,CAAA;IAC/F;IACA,OAAO;QAACF;KAAE;AACZ;AAEA,iGAAiG;AACjG,0GAA0G;AAC1G,SAASuC,mBAAmBC,KAAe,EAAE3C,SAAiB,EAAEC,OAAe,EAAEsB,OAAe,EAAEc,SAAoB,EAAEC,QAA6B,EAAEM,OAAO,CAAC;IAC7J,MAAMC,SAAiB,EAAE;IACzB,IAAIT,aAAuB,EAAE;IAC7B,IAAIU,aAAa9C;IACjB,IAAIyB,SAASmB;IACb,IAAK,IAAIG,OAAO/C,WAAW+C,QAAQ9C,SAAS8C,OAAQ;QAClD,MAAMC,WAAWL,KAAK,CAACI,OAAO,EAAE;QAChC,MAAME,MAAMb,WAAW3C,MAAM,GAAG,KAAKgC,SAAS,IAAIlB,iBAAiB;QACnE,MAAM2C,aAAatE,eAAeoE;QAClC,IAAIZ,WAAW3C,MAAM,GAAG,KAAKgC,SAASwB,MAAMC,aAAa3B,SAAS;YAChEsB,OAAOnD,IAAI,IAAI+C,cAAcL,YAAYU,YAAYC,OAAO,GAAGxB,SAASc,WAAWC;YACnFF,aAAa,EAAE;YACfX,SAAS;YACTqB,aAAaC;YACbX,WAAW1C,IAAI,CAACsD;YAChBvB,UAAUyB;YACV;QACF;QACAd,WAAW1C,IAAI,CAACsD;QAChBvB,UAAUwB,MAAMC;IAClB;IACA,IAAId,WAAW3C,MAAM,GAAG,GAAGoD,OAAOnD,IAAI,IAAI+C,cAAcL,YAAYU,YAAY7C,SAASsB,SAASc,WAAWC;IAC7G,OAAOO;AACT;AAEA,0FAA0F;AAC1F,8FAA8F;AAC9F,SAASM,WAAWC,WAAoB,EAAET,KAAe,EAAEU,QAAyB;IAClF,MAAM9B,UAAU8B,SAASpE,YAAY;IACrC,MAAMqE,UAAU/B,UAAUzC;IAC1B,MAAMyE,WAA4C,EAAE;IACpD,IAAI3D,QAAgB,EAAE;IACtB,IAAI4D,iBAAiB;IACrB,IAAI/B,SAAS;IAEb,SAASgC;QACP,MAAMC,QAAQ/D,SAASC;QACvB,IAAI8D,OAAOH,SAAS7D,IAAI,CAACgE;QACzB9D,QAAQ,EAAE;QACV4D,iBAAiB;QACjB/B,SAAS;IACX;IAEA,mFAAmF;IACnF,2FAA2F;IAC3F,SAASkC,QAAQzE,IAAY,EAAE0E,QAAgB,EAAE5D,SAAiB,EAAEC,OAAe;QACjFwB,UAAU,AAAC7B,CAAAA,MAAMH,MAAM,GAAG,IAAIc,iBAAiB,CAAA,IAAK3B,eAAegF;QACnEhE,MAAMF,IAAI,CAAC;YAAEM;YAAWC;YAASf;QAAK;IACxC;IAEA,KAAK,MAAMK,SAAS6D,YAAa;QAC/B,MAAMS,MAAMlB,MAAMmB,KAAK,CAACvE,MAAMS,SAAS,GAAG,GAAGT,MAAMU,OAAO,EAAEG,IAAI,CAAC;QACjE,MAAM2D,cAAcnF,eAAeiF;QAEnC,IAAIE,cAAcT,SAAS;YACzB,MAAMV,OAAOhD,MAAMH,MAAM,GAAG,IAAIgC,SAAS;YACzC,KAAK,MAAMtB,KAAKuC,mBAAmBC,OAAOpD,MAAMS,SAAS,EAAET,MAAMU,OAAO,EAAEsB,SAAShC,MAAMC,IAAI,EAAE6D,SAASnE,IAAI,EAAE0D,MAAO;gBACnHhD,MAAMF,IAAI,CAACS;gBACXsD;YACF;YACA;QACF;QAEA,MAAMO,YAAYvF,YAAYc;QAC9B,MAAMqE,WAAWP,SAASnE,IAAI,KAAK,QAAQ2E,MAAMG;QAEjD,IAAIzE,MAAMC,IAAI,KAAK,WAAW;YAC5BmE,QAAQK,WAAWJ,UAAUrE,MAAMS,SAAS,EAAET,MAAMU,OAAO;YAC3D;QACF;QAEA,wFAAwF;QACxF,wDAAwD;QACxD,MAAMgE,eAAerE,MAAMH,MAAM,GAAG,KAAKgC,SAASlB,iBAAiBwD,cAAcT;QACjF,IAAIW,cAAcR;QAElBE,QAAQK,WAAWJ,UAAUrE,MAAMS,SAAS,EAAET,MAAMU,OAAO;QAC3DuD;QAEA,IAAIA,kBAAkB3E,gBAAgB4E;IACxC;IACAA;IAEA,OAAOF;AACT;AAEA,8FAA8F;AAC9F,8CAA8C;AAC9C,OAAO,SAASG,MAAMtE,MAAe,EAAE8E,IAAY,EAAElF,IAAmB;IACtE,MAAMqE,WAAWtE,eAAeC;IAChC,MAAM2D,QAAQuB,KAAKC,KAAK,CAAC;IACzB,MAAMC,SAA0C,EAAE;IAClD,KAAK,MAAMC,SAASlF,YAAYC,QAASgF,OAAO1E,IAAI,IAAIyD,WAAWkB,OAAO1B,OAAOU;IACjF,+FAA+F;IAC/F,qHAAqH;IACrH,MAAMiB,SACJjB,SAASnE,IAAI,KAAK,QACdkF,OAAOlE,GAAG,CAAC,CAACqE,IACVA,EAAElE,KAAK,GACHkE,IACA;YACE,GAAGA,CAAC;YACJrF,MAAMyD,MACHmB,KAAK,CAACS,EAAEvE,SAAS,GAAG,GAAGuE,EAAEtE,OAAO,EAChCG,IAAI,CAAC,MACLoE,IAAI;QACT,KAENJ;IACN,mHAAmH;IACnH,OAAOE,OAAOG,MAAM,CAAC,CAACF,IAAMA,EAAErF,IAAI,CAACsF,IAAI,GAAG/E,MAAM,GAAG,GAAGS,GAAG,CAAC,CAACqE,IAAO,CAAA;YAAEvE,WAAWuE,EAAEvE,SAAS;YAAEC,SAASsE,EAAEtE,OAAO;YAAEf,MAAMqF,EAAErF,IAAI;QAAC,CAAA;AAC/H"}