@pgpmjs/transform 0.22.0 → 0.23.1

package/byte-slice.d.ts

@@ -0,0 +1,23 @@
+/**
+ * Byte-correct statement slicing.
+ *
+ * libpg-query reports statement offsets (`stmt_location` / `stmt_len`, and the
+ * `span` that `@pgsql/semantics` derives from them) as UTF-8 **byte** offsets,
+ * but JavaScript string indexing is UTF-16. The two agree only while a script
+ * is pure ASCII; a single non-ASCII byte (an em dash in a comment, an accented
+ * word in a seed row, an emoji) shifts every later byte offset past its JS
+ * index, so slicing the source string directly with `String.prototype.slice`
+ * mis-cuts every statement that follows it — e.g. `COMMENT ON ...` reappears as
+ * `MMENT ON ...` and fails to reparse.
+ *
+ * Slice on the UTF-8 bytes instead and decode back to text. Statement offsets
+ * always fall on character boundaries, so decoding a byte range never splits a
+ * multibyte character.
+ */
+export declare const sqlSourceBytes: (source: string) => Buffer;
+/**
+ * Slice a statement's verbatim text out of the source bytes by its byte-offset
+ * span. `len === undefined` slices to the end of the source (libpg-query omits
+ * `stmt_len` for the final statement).
+ */
+export declare const sliceStatementBytes: (sourceBytes: Buffer, start: number, len?: number) => string;

package/byte-slice.js

@@ -0,0 +1,31 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.sliceStatementBytes = exports.sqlSourceBytes = void 0;
+/**
+ * Byte-correct statement slicing.
+ *
+ * libpg-query reports statement offsets (`stmt_location` / `stmt_len`, and the
+ * `span` that `@pgsql/semantics` derives from them) as UTF-8 **byte** offsets,
+ * but JavaScript string indexing is UTF-16. The two agree only while a script
+ * is pure ASCII; a single non-ASCII byte (an em dash in a comment, an accented
+ * word in a seed row, an emoji) shifts every later byte offset past its JS
+ * index, so slicing the source string directly with `String.prototype.slice`
+ * mis-cuts every statement that follows it — e.g. `COMMENT ON ...` reappears as
+ * `MMENT ON ...` and fails to reparse.
+ *
+ * Slice on the UTF-8 bytes instead and decode back to text. Statement offsets
+ * always fall on character boundaries, so decoding a byte range never splits a
+ * multibyte character.
+ */
+const sqlSourceBytes = (source) => Buffer.from(source, 'utf8');
+exports.sqlSourceBytes = sqlSourceBytes;
+/**
+ * Slice a statement's verbatim text out of the source bytes by its byte-offset
+ * span. `len === undefined` slices to the end of the source (libpg-query omits
+ * `stmt_len` for the final statement).
+ */
+const sliceStatementBytes = (sourceBytes, start, len) => {
+    const end = len === undefined ? sourceBytes.length : start + len;
+    return sourceBytes.subarray(start, end).toString('utf8');
+};
+exports.sliceStatementBytes = sliceStatementBytes;

package/esm/byte-slice.js

@@ -0,0 +1,26 @@
+/**
+ * Byte-correct statement slicing.
+ *
+ * libpg-query reports statement offsets (`stmt_location` / `stmt_len`, and the
+ * `span` that `@pgsql/semantics` derives from them) as UTF-8 **byte** offsets,
+ * but JavaScript string indexing is UTF-16. The two agree only while a script
+ * is pure ASCII; a single non-ASCII byte (an em dash in a comment, an accented
+ * word in a seed row, an emoji) shifts every later byte offset past its JS
+ * index, so slicing the source string directly with `String.prototype.slice`
+ * mis-cuts every statement that follows it — e.g. `COMMENT ON ...` reappears as
+ * `MMENT ON ...` and fails to reparse.
+ *
+ * Slice on the UTF-8 bytes instead and decode back to text. Statement offsets
+ * always fall on character boundaries, so decoding a byte range never splits a
+ * multibyte character.
+ */
+export const sqlSourceBytes = (source) => Buffer.from(source, 'utf8');
+/**
+ * Slice a statement's verbatim text out of the source bytes by its byte-offset
+ * span. `len === undefined` slices to the end of the source (libpg-query omits
+ * `stmt_len` for the final statement).
+ */
+export const sliceStatementBytes = (sourceBytes, start, len) => {
+    const end = len === undefined ? sourceBytes.length : start + len;
+    return sourceBytes.subarray(start, end).toString('utf8');
+};

package/esm/granularity-driver.js

@@ -2,6 +2,7 @@ import { pathFor } from '@pgpmjs/naming-spec';
 import { revertFor, verifyFor } from '@pgsql/scripts';
 import { classifyStatements } from '@pgsql/semantics';
 import { buildStatementGraph, identityOf, restructureSql } from '@pgsql/transform';
+import { sliceStatementBytes, sqlSourceBytes } from './byte-slice';
 import { nameUnnamedConstraints, subObjectIdentityOf } from './sub-object';
 export const CHANGE_GRANULARITIES = ['alteration', 'object', 'single'];
 export const isChangeGranularity = (value) => CHANGE_GRANULARITIES.includes(value);
@@ -53,12 +54,13 @@ export function restructureChanges(changes, options) {
     // target (creates[0]), so a table's CREATE and its remaining ALTERs land
     // in the same change regardless of statement kind.
     const facts = classifyStatements(sql);
+    const sqlBytes = sqlSourceBytes(sql);
     if (options.changeGranularity === 'single') {
         const name = options.singleChangeName ?? 'module/init';
         const inputNames = new Set(changes.map(c => c.name));
         const dependencies = [...new Set(changes.flatMap(c => c.dependencies).filter(d => !inputNames.has(d)))].sort();
         const statements = facts
-            .map(f => sql.slice(f.span.start, f.span.start + f.span.len).trim())
+            .map(f => sliceStatementBytes(sqlBytes, f.span.start, f.span.len).trim())
             .filter(Boolean)
             .map(text => (text.endsWith(';') ? text : `${text};`));
         const revert = revertFor(facts);
@@ -137,7 +139,7 @@ export function restructureChanges(changes, options) {
     const groupSql = groupNames.map(() => []);
     const groupStatements = groupNames.map(() => []);
     facts.forEach((f, i) => {
-        const text = sql.slice(f.span.start, f.span.start + f.span.len).trim();
+        const text = sliceStatementBytes(sqlBytes, f.span.start, f.span.len).trim();
         const g = groupOf[i];
         if (g !== -1 && text) {
             groupSql[g].push(text.endsWith(';') ? text : `${text};`);

package/esm/module-source.js

@@ -11,29 +11,45 @@ export const stripTransactionWrapper = (sql) => sql
     .filter(line => !TX_LINE.test(line))
     .join('\n')
     .trim();
+const CROSS_PACKAGE_TAG = /^([^:]+):@(.+)$/;
 /**
- * Resolve a plan dependency token to a concrete change name.
+ * Resolve a plan dependency token to a concrete change identity.
  *
  * Non-tag tokens (a bare change name, or a `pkg:change` cross-package name)
  * pass through unchanged — they are already the identity the ledger and
- * generated plans record. A tag token (`@tag`) is resolved to its change via
- * the plan that defines it, so a dependency written against a tag is not
- * re-emitted verbatim into a generated plan or a ledger row. Cross-package
- * tags (`pkg:@tag`) cannot be resolved from a single module's plan and are
- * left as-is with a warning; the resolver returns them unchanged. Anything
- * that fails to resolve is kept verbatim (never dropped) with a warning.
+ * generated plans record. A tag token is resolved so a dependency written
+ * against a tag is never re-emitted verbatim into a generated plan or a
+ * ledger row:
+ * - a local tag (`@tag`) resolves to its change name via this plan;
+ * - a cross-package tag (`pkg:@tag`) resolves to the canonical `pkg:change`
+ *   qualified name via `crossPackagePlans[pkg]` when available.
+ *
+ * Anything that cannot be resolved (unknown tag, or a cross-package tag with
+ * no plan in context) is kept verbatim — never dropped — with a warning.
  */
-const resolveDependencyToken = (token, plan, warnings, changeName) => {
+const resolveDependencyToken = (token, plan, warnings, changeName, crossPackagePlans) => {
     if (!token.includes('@'))
         return token;
+    const cross = token.match(CROSS_PACKAGE_TAG);
+    if (cross && cross[1] !== plan.package) {
+        const pkg = cross[1];
+        const otherPlan = crossPackagePlans?.get(pkg);
+        if (!otherPlan) {
+            warnings.push(`${changeName}: dependency tag "${token}" is cross-package; left unresolved (no plan for "${pkg}" in context)`);
+            return token;
+        }
+        const resolved = resolveReference(token, otherPlan, pkg);
+        if (resolved.error || !resolved.change) {
+            warnings.push(`${changeName}: could not resolve cross-package dependency tag "${token}" (${resolved.error ?? 'no matching change'}); left as-is`);
+            return token;
+        }
+        return `${pkg}:${resolved.change}`;
+    }
     const resolved = resolveReference(token, plan, plan.package);
     if (resolved.error || !resolved.change) {
         warnings.push(`${changeName}: could not resolve dependency tag "${token}" (${resolved.error ?? 'no matching change'}); left as-is`);
         return token;
     }
-    if (resolved.change === token) {
-        warnings.push(`${changeName}: dependency tag "${token}" is cross-package; left unresolved (resolve it in workspace context)`);
-    }
     return resolved.change;
 };
 /**
@@ -42,7 +58,7 @@ const resolveDependencyToken = (token, plan, warnings, changeName) => {
  * result can be flattened straight into the dials pipeline
  * (`restructureChanges` / `partitionUnits`).
  */
-export const loadModuleSource = (moduleDir) => {
+export const loadModuleSource = (moduleDir, options = {}) => {
     const modulePath = path.resolve(moduleDir);
     const planResult = parsePlanFile(path.join(modulePath, 'pgpm.plan'));
     if (!planResult.data) {
@@ -61,7 +77,7 @@ export const loadModuleSource = (moduleDir) => {
         const { body } = parsePgpmHeader(raw);
         changes.push({
             name: change.name,
-            dependencies: (change.dependencies ?? []).map(dep => resolveDependencyToken(dep, plan, warnings, change.name)),
+            dependencies: (change.dependencies ?? []).map(dep => resolveDependencyToken(dep, plan, warnings, change.name, options.crossPackagePlans)),
             deploy: stripTransactionWrapper(body)
         });
     }

package/esm/partition-driver.js

@@ -23,6 +23,7 @@
 import { pathFor } from '@pgpmjs/naming-spec';
 import { classifyStatements } from '@pgsql/semantics';
 import { buildStatementGraph, identityOf } from '@pgsql/transform';
+import { sliceStatementBytes, sqlSourceBytes } from './byte-slice';
 /** A partition whose packages require each other in a cycle — unshippable. */
 export class PartitionCycleError extends Error {
     packages;
@@ -81,6 +82,7 @@ export const RESIDUAL_UNIT_PATH = 'misc/statements';
  */
 function buildUnitGraph(sql, style, splitRiders) {
     const facts = classifyStatements(sql);
+    const sqlBytes = sqlSourceBytes(sql);
     const graph = buildStatementGraph(facts);
     const unitOf = new Array(facts.length).fill(-1);
     const units = [];
@@ -163,7 +165,7 @@ function buildUnitGraph(sql, style, splitRiders) {
         unitOf[i] = unitFor(null, f.kind);
     });
     facts.forEach((f, i) => {
-        const text = sql.slice(f.span.start, f.span.start + f.span.len).trim();
+        const text = sliceStatementBytes(sqlBytes, f.span.start, f.span.len).trim();
         if (!text)
             return;
         const unit = units[unitOf[i]];

package/esm/program.js

@@ -1,5 +1,6 @@
 import { classifyStatements } from '@pgsql/semantics';
 import { Deparser, parseSql } from 'plpgsql-parser';
+import { sliceStatementBytes, sqlSourceBytes } from './byte-slice';
 /**
  * Parse a SQL script into a statement program. Requires `loadModule()` from
  * `plpgsql-parser` to have completed.
@@ -8,6 +9,7 @@ export function parseSqlProgram(source) {
     const parseResult = parseSql(source);
     const stmts = parseResult?.stmts ?? [];
     const facts = classifyStatements(source);
+    const sourceBytes = sqlSourceBytes(source);
     const statements = [];
     for (let i = 0; i < stmts.length; i++) {
         const rawStmt = stmts[i];
@@ -18,7 +20,7 @@ export function parseSqlProgram(source) {
             stmt: rawStmt.stmt,
             facts: facts[i],
             span: { start, len },
-            raw: source.slice(start, start + len),
+            raw: sliceStatementBytes(sourceBytes, start, len),
             dirty: false
         });
     }
@@ -34,16 +36,22 @@ export function emitSqlProgram(program) {
     const { source, statements } = program;
     if (!statements.some(s => s.dirty))
         return source;
+    // Spans are UTF-8 byte offsets; slice statements and the verbatim gaps
+    // between them on the bytes so a multibyte character never mis-cuts the
+    // reassembly. Statement boundaries fall on character boundaries.
+    const sourceBytes = sqlSourceBytes(source);
     const pieces = [];
     let cursor = 0;
     for (const statement of statements) {
         const { start, len } = statement.span;
         if (start > cursor)
-            pieces.push(source.slice(cursor, start));
-        pieces.push(statement.dirty ? Deparser.deparse(statement.stmt) : source.slice(start, start + len));
+            pieces.push(sliceStatementBytes(sourceBytes, cursor, start - cursor));
+        pieces.push(statement.dirty
+            ? Deparser.deparse(statement.stmt)
+            : sliceStatementBytes(sourceBytes, start, len));
         cursor = start + len;
     }
-    if (cursor < source.length)
-        pieces.push(source.slice(cursor));
+    if (cursor < sourceBytes.length)
+        pieces.push(sliceStatementBytes(sourceBytes, cursor));
     return pieces.join('');
 }

package/esm/semantic-diff-driver.js

@@ -31,6 +31,7 @@ import { existenceCheck, invertStatement, revertFor, verifyFor } from '@pgsql/sc
 import { classifyStatements } from '@pgsql/semantics';
 import { buildStatementGraph, cleanTree, identityOf } from '@pgsql/transform';
 import { Deparser, parseSync } from 'plpgsql-parser';
+import { sliceStatementBytes, sqlSourceBytes } from './byte-slice';
 import { defaultConstraintName } from './constraint-names';
 import { restructureChanges } from './granularity-driver';
 const identityKey = (id) => `${id.kind}\u0000${id.schema ?? ''}\u0000${id.table ?? ''}\u0000${id.name}`;
@@ -47,8 +48,9 @@ function readSide(sql, warnings, label) {
     const order = [];
     const unitOf = new Array(facts.length).fill(null);
     const byObject = new Map();
+    const sqlBytes = sqlSourceBytes(sql);
     facts.forEach((f, i) => {
-        const text = sql.slice(f.span.start, f.span.start + f.span.len).trim();
+        const text = sliceStatementBytes(sqlBytes, f.span.start, f.span.len).trim();
         if (!text)
             return;
         let identity = groupingIdentity(f);

package/esm/sub-object.js

@@ -1,4 +1,5 @@
 import { Deparser, parseSync } from 'plpgsql-parser';
+import { sliceStatementBytes, sqlSourceBytes } from './byte-slice';
 import { defaultConstraintName } from './constraint-names';
 /** The single AlterTableCmd of a one-command ALTER TABLE, or null. */
 function singleCmd(stmt) {
@@ -46,6 +47,7 @@ export function subObjectIdentityOf(facts) {
  */
 export function nameUnnamedConstraints(sql) {
     const parsed = parseSync(sql);
+    const sqlBytes = sqlSourceBytes(sql);
     const pieces = [];
     for (const raw of parsed.sql.stmts) {
         const node = raw.stmt;
@@ -71,7 +73,7 @@ export function nameUnnamedConstraints(sql) {
         else {
             const start = raw.stmt_location ?? 0;
             const len = raw.stmt_len;
-            const text = len === undefined ? sql.slice(start) : sql.slice(start, start + len);
+            const text = sliceStatementBytes(sqlBytes, start, len);
             const trimmed = text.trim();
             pieces.push(trimmed.endsWith(';') ? trimmed : `${trimmed};`);
         }

package/granularity-driver.js

@@ -7,6 +7,7 @@ const naming_spec_1 = require("@pgpmjs/naming-spec");
 const scripts_1 = require("@pgsql/scripts");
 const semantics_1 = require("@pgsql/semantics");
 const transform_1 = require("@pgsql/transform");
+const byte_slice_1 = require("./byte-slice");
 const sub_object_1 = require("./sub-object");
 exports.CHANGE_GRANULARITIES = ['alteration', 'object', 'single'];
 const isChangeGranularity = (value) => exports.CHANGE_GRANULARITIES.includes(value);
@@ -59,12 +60,13 @@ function restructureChanges(changes, options) {
     // target (creates[0]), so a table's CREATE and its remaining ALTERs land
     // in the same change regardless of statement kind.
     const facts = (0, semantics_1.classifyStatements)(sql);
+    const sqlBytes = (0, byte_slice_1.sqlSourceBytes)(sql);
     if (options.changeGranularity === 'single') {
         const name = options.singleChangeName ?? 'module/init';
         const inputNames = new Set(changes.map(c => c.name));
         const dependencies = [...new Set(changes.flatMap(c => c.dependencies).filter(d => !inputNames.has(d)))].sort();
         const statements = facts
-            .map(f => sql.slice(f.span.start, f.span.start + f.span.len).trim())
+            .map(f => (0, byte_slice_1.sliceStatementBytes)(sqlBytes, f.span.start, f.span.len).trim())
             .filter(Boolean)
             .map(text => (text.endsWith(';') ? text : `${text};`));
         const revert = (0, scripts_1.revertFor)(facts);
@@ -143,7 +145,7 @@ function restructureChanges(changes, options) {
     const groupSql = groupNames.map(() => []);
     const groupStatements = groupNames.map(() => []);
     facts.forEach((f, i) => {
-        const text = sql.slice(f.span.start, f.span.start + f.span.len).trim();
+        const text = (0, byte_slice_1.sliceStatementBytes)(sqlBytes, f.span.start, f.span.len).trim();
         const g = groupOf[i];
         if (g !== -1 && text) {
             groupSql[g].push(text.endsWith(';') ? text : `${text};`);

package/module-source.d.ts

@@ -1,3 +1,10 @@
+/**
+ * Module ingestion for the dials pipeline: load an on-disk pgpm module into
+ * the flattened, classified change list the transform/partition drivers
+ * consume. `pgpm transform` uses this today; a future `pgpm import` (dumps)
+ * and `pgpm diff` are expected to reuse the same seam.
+ */
+import type { ExtendedPlanFile } from '@pgpmjs/ast';
 /** One change of a loaded module: headerless, unwrapped deploy SQL. */
 export interface ModuleSourceChange {
     /** Change name (plan token). */
@@ -24,10 +31,21 @@ export interface ModuleSource {
  * transaction control.
  */
 export declare const stripTransactionWrapper: (sql: string) => string;
+/** Options controlling how {@link loadModuleSource} ingests a module. */
+export interface LoadModuleSourceOptions {
+    /**
+     * Parsed plans of the other packages in the same workspace, keyed by
+     * package name (`%project`). Lets a cross-package tag dependency
+     * (`pkg:@tag`) resolve against the plan that defines the tag. Without it,
+     * cross-package tags cannot be resolved from a single module's plan and are
+     * left verbatim with a warning.
+     */
+    crossPackagePlans?: Map<string, ExtendedPlanFile>;
+}
 /**
  * Load a pgpm module from disk: parse pgpm.plan and read each change's
  * deploy script, stripping pgpm headers and transaction wrappers so the
  * result can be flattened straight into the dials pipeline
  * (`restructureChanges` / `partitionUnits`).
  */
-export declare const loadModuleSource: (moduleDir: string) => ModuleSource;
+export declare const loadModuleSource: (moduleDir: string, options?: LoadModuleSourceOptions) => ModuleSource;

package/module-source.js

@@ -48,29 +48,45 @@ const stripTransactionWrapper = (sql) => sql
     .join('\n')
     .trim();
 exports.stripTransactionWrapper = stripTransactionWrapper;
+const CROSS_PACKAGE_TAG = /^([^:]+):@(.+)$/;
 /**
- * Resolve a plan dependency token to a concrete change name.
+ * Resolve a plan dependency token to a concrete change identity.
  *
  * Non-tag tokens (a bare change name, or a `pkg:change` cross-package name)
  * pass through unchanged — they are already the identity the ledger and
- * generated plans record. A tag token (`@tag`) is resolved to its change via
- * the plan that defines it, so a dependency written against a tag is not
- * re-emitted verbatim into a generated plan or a ledger row. Cross-package
- * tags (`pkg:@tag`) cannot be resolved from a single module's plan and are
- * left as-is with a warning; the resolver returns them unchanged. Anything
- * that fails to resolve is kept verbatim (never dropped) with a warning.
+ * generated plans record. A tag token is resolved so a dependency written
+ * against a tag is never re-emitted verbatim into a generated plan or a
+ * ledger row:
+ * - a local tag (`@tag`) resolves to its change name via this plan;
+ * - a cross-package tag (`pkg:@tag`) resolves to the canonical `pkg:change`
+ *   qualified name via `crossPackagePlans[pkg]` when available.
+ *
+ * Anything that cannot be resolved (unknown tag, or a cross-package tag with
+ * no plan in context) is kept verbatim — never dropped — with a warning.
  */
-const resolveDependencyToken = (token, plan, warnings, changeName) => {
+const resolveDependencyToken = (token, plan, warnings, changeName, crossPackagePlans) => {
     if (!token.includes('@'))
         return token;
+    const cross = token.match(CROSS_PACKAGE_TAG);
+    if (cross && cross[1] !== plan.package) {
+        const pkg = cross[1];
+        const otherPlan = crossPackagePlans?.get(pkg);
+        if (!otherPlan) {
+            warnings.push(`${changeName}: dependency tag "${token}" is cross-package; left unresolved (no plan for "${pkg}" in context)`);
+            return token;
+        }
+        const resolved = (0, ast_1.resolveReference)(token, otherPlan, pkg);
+        if (resolved.error || !resolved.change) {
+            warnings.push(`${changeName}: could not resolve cross-package dependency tag "${token}" (${resolved.error ?? 'no matching change'}); left as-is`);
+            return token;
+        }
+        return `${pkg}:${resolved.change}`;
+    }
     const resolved = (0, ast_1.resolveReference)(token, plan, plan.package);
     if (resolved.error || !resolved.change) {
         warnings.push(`${changeName}: could not resolve dependency tag "${token}" (${resolved.error ?? 'no matching change'}); left as-is`);
         return token;
     }
-    if (resolved.change === token) {
-        warnings.push(`${changeName}: dependency tag "${token}" is cross-package; left unresolved (resolve it in workspace context)`);
-    }
     return resolved.change;
 };
 /**
@@ -79,7 +95,7 @@ const resolveDependencyToken = (token, plan, warnings, changeName) => {
  * result can be flattened straight into the dials pipeline
  * (`restructureChanges` / `partitionUnits`).
  */
-const loadModuleSource = (moduleDir) => {
+const loadModuleSource = (moduleDir, options = {}) => {
     const modulePath = path.resolve(moduleDir);
     const planResult = (0, ast_1.parsePlanFile)(path.join(modulePath, 'pgpm.plan'));
     if (!planResult.data) {
@@ -98,7 +114,7 @@ const loadModuleSource = (moduleDir) => {
         const { body } = (0, ast_1.parsePgpmHeader)(raw);
         changes.push({
             name: change.name,
-            dependencies: (change.dependencies ?? []).map(dep => resolveDependencyToken(dep, plan, warnings, change.name)),
+            dependencies: (change.dependencies ?? []).map(dep => resolveDependencyToken(dep, plan, warnings, change.name, options.crossPackagePlans)),
             deploy: (0, exports.stripTransactionWrapper)(body)
         });
     }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pgpmjs/transform",
-  "version": "0.22.0",
+  "version": "0.23.1",
   "author": "Constructive <developers@constructive.io>",
   "description": "Change-aware SQL transformation for PGPM — superset of @pgsql/transform plus change categorization, fixture closure, and bundle transpile/apply drivers",
   "main": "index.js",
@@ -39,16 +39,16 @@
     "plpgsql"
   ],
   "devDependencies": {
-    "@pgpmjs/bundle": "^0.14.1",
+    "@pgpmjs/bundle": "^0.14.2",
     "makage": "^0.3.0"
   },
   "dependencies": {
-    "@pgpmjs/ast": "^0.11.1",
+    "@pgpmjs/ast": "^0.11.2",
     "@pgpmjs/naming-spec": "^0.3.0",
     "@pgsql/scripts": "^18.3.4",
     "@pgsql/semantics": "^18.1.2",
     "@pgsql/transform": "^18.15.2",
     "plpgsql-parser": "^18.5.3"
   },
-  "gitHead": "6da0705f8aa49d62dc2defb8b135b71674611483"
+  "gitHead": "43121004a8bedd2c5fcf7bf33208079d55bad01f"
 }

package/partition-driver.js

@@ -27,6 +27,7 @@ exports.partitionUnits = partitionUnits;
 const naming_spec_1 = require("@pgpmjs/naming-spec");
 const semantics_1 = require("@pgsql/semantics");
 const transform_1 = require("@pgsql/transform");
+const byte_slice_1 = require("./byte-slice");
 /** A partition whose packages require each other in a cycle — unshippable. */
 class PartitionCycleError extends Error {
     packages;
@@ -86,6 +87,7 @@ exports.RESIDUAL_UNIT_PATH = 'misc/statements';
  */
 function buildUnitGraph(sql, style, splitRiders) {
     const facts = (0, semantics_1.classifyStatements)(sql);
+    const sqlBytes = (0, byte_slice_1.sqlSourceBytes)(sql);
     const graph = (0, transform_1.buildStatementGraph)(facts);
     const unitOf = new Array(facts.length).fill(-1);
     const units = [];
@@ -168,7 +170,7 @@ function buildUnitGraph(sql, style, splitRiders) {
         unitOf[i] = unitFor(null, f.kind);
     });
     facts.forEach((f, i) => {
-        const text = sql.slice(f.span.start, f.span.start + f.span.len).trim();
+        const text = (0, byte_slice_1.sliceStatementBytes)(sqlBytes, f.span.start, f.span.len).trim();
         if (!text)
             return;
         const unit = units[unitOf[i]];

package/program.js

@@ -4,6 +4,7 @@ exports.parseSqlProgram = parseSqlProgram;
 exports.emitSqlProgram = emitSqlProgram;
 const semantics_1 = require("@pgsql/semantics");
 const plpgsql_parser_1 = require("plpgsql-parser");
+const byte_slice_1 = require("./byte-slice");
 /**
  * Parse a SQL script into a statement program. Requires `loadModule()` from
  * `plpgsql-parser` to have completed.
@@ -12,6 +13,7 @@ function parseSqlProgram(source) {
     const parseResult = (0, plpgsql_parser_1.parseSql)(source);
     const stmts = parseResult?.stmts ?? [];
     const facts = (0, semantics_1.classifyStatements)(source);
+    const sourceBytes = (0, byte_slice_1.sqlSourceBytes)(source);
     const statements = [];
     for (let i = 0; i < stmts.length; i++) {
         const rawStmt = stmts[i];
@@ -22,7 +24,7 @@ function parseSqlProgram(source) {
             stmt: rawStmt.stmt,
             facts: facts[i],
             span: { start, len },
-            raw: source.slice(start, start + len),
+            raw: (0, byte_slice_1.sliceStatementBytes)(sourceBytes, start, len),
             dirty: false
         });
     }
@@ -38,16 +40,22 @@ function emitSqlProgram(program) {
     const { source, statements } = program;
     if (!statements.some(s => s.dirty))
         return source;
+    // Spans are UTF-8 byte offsets; slice statements and the verbatim gaps
+    // between them on the bytes so a multibyte character never mis-cuts the
+    // reassembly. Statement boundaries fall on character boundaries.
+    const sourceBytes = (0, byte_slice_1.sqlSourceBytes)(source);
     const pieces = [];
     let cursor = 0;
     for (const statement of statements) {
         const { start, len } = statement.span;
         if (start > cursor)
-            pieces.push(source.slice(cursor, start));
-        pieces.push(statement.dirty ? plpgsql_parser_1.Deparser.deparse(statement.stmt) : source.slice(start, start + len));
+            pieces.push((0, byte_slice_1.sliceStatementBytes)(sourceBytes, cursor, start - cursor));
+        pieces.push(statement.dirty
+            ? plpgsql_parser_1.Deparser.deparse(statement.stmt)
+            : (0, byte_slice_1.sliceStatementBytes)(sourceBytes, start, len));
         cursor = start + len;
     }
-    if (cursor < source.length)
-        pieces.push(source.slice(cursor));
+    if (cursor < sourceBytes.length)
+        pieces.push((0, byte_slice_1.sliceStatementBytes)(sourceBytes, cursor));
     return pieces.join('');
 }

package/semantic-diff-driver.js

@@ -36,6 +36,7 @@ const scripts_1 = require("@pgsql/scripts");
 const semantics_1 = require("@pgsql/semantics");
 const transform_1 = require("@pgsql/transform");
 const plpgsql_parser_1 = require("plpgsql-parser");
+const byte_slice_1 = require("./byte-slice");
 const constraint_names_1 = require("./constraint-names");
 const granularity_driver_1 = require("./granularity-driver");
 const identityKey = (id) => `${id.kind}\u0000${id.schema ?? ''}\u0000${id.table ?? ''}\u0000${id.name}`;
@@ -52,8 +53,9 @@ function readSide(sql, warnings, label) {
     const order = [];
     const unitOf = new Array(facts.length).fill(null);
     const byObject = new Map();
+    const sqlBytes = (0, byte_slice_1.sqlSourceBytes)(sql);
     facts.forEach((f, i) => {
-        const text = sql.slice(f.span.start, f.span.start + f.span.len).trim();
+        const text = (0, byte_slice_1.sliceStatementBytes)(sqlBytes, f.span.start, f.span.len).trim();
         if (!text)
             return;
         let identity = groupingIdentity(f);

package/sub-object.js

@@ -3,6 +3,7 @@ Object.defineProperty(exports, "__esModule", { value: true });
 exports.subObjectIdentityOf = subObjectIdentityOf;
 exports.nameUnnamedConstraints = nameUnnamedConstraints;
 const plpgsql_parser_1 = require("plpgsql-parser");
+const byte_slice_1 = require("./byte-slice");
 const constraint_names_1 = require("./constraint-names");
 /** The single AlterTableCmd of a one-command ALTER TABLE, or null. */
 function singleCmd(stmt) {
@@ -50,6 +51,7 @@ function subObjectIdentityOf(facts) {
  */
 function nameUnnamedConstraints(sql) {
     const parsed = (0, plpgsql_parser_1.parseSync)(sql);
+    const sqlBytes = (0, byte_slice_1.sqlSourceBytes)(sql);
     const pieces = [];
     for (const raw of parsed.sql.stmts) {
         const node = raw.stmt;
@@ -75,7 +77,7 @@ function nameUnnamedConstraints(sql) {
         else {
             const start = raw.stmt_location ?? 0;
             const len = raw.stmt_len;
-            const text = len === undefined ? sql.slice(start) : sql.slice(start, start + len);
+            const text = (0, byte_slice_1.sliceStatementBytes)(sqlBytes, start, len);
             const trimmed = text.trim();
             pieces.push(trimmed.endsWith(';') ? trimmed : `${trimmed};`);
         }