@smeltjs/core 0.5.0 → 0.6.0

package/dist/plan/diff.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"diff.d.ts","sourceRoot":"","sources":["../../src/plan/diff.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,WAAW,EAAiB,SAAS,EAAkB,OAAO,EAAE,MAAM,aAAa,CAAC;AAKlG,eAAO,MAAM,eAAe,YAAY,CAAC;AAEzC;;;GAGG;AACH,eAAO,MAAM,kBAAkB,kBAAkB,CAAC;AAClD,eAAO,MAAM,kBAAkB,kBAAkB,CAAC;AAClD,eAAO,MAAM,gBAAgB,gBAAgB,CAAC;AAQ9C,MAAM,WAAW,kBAAkB;IACjC,gEAAgE;IAChE,QAAQ,CAAC,aAAa,CAAC,EAAE,OAAO,CAAC;CAClC;AAED;;;;;;;;;;;;;;;;GAgBG;AACH,qBAAa,WAAY,YAAW,OAAO;;IACzC,QAAQ,CAAC,EAAE,aAAmB;IAG9B,YAAY,OAAO,GAAE,kBAAuB,EAE3C;IAED,IAAI,CAAC,KAAK,EAAE,SAAS,GAAG,OAAO,CAAC,WAAW,CAAC,CAE3C;CACF;AA2BD;;;;;;GAMG;AACH,wBAAgB,QAAQ,CAAC,KAAK,EAAE,SAAS,EAAE,OAAO,GAAE,kBAAuB,GAAG,WAAW,CA8FxF","sourcesContent":["import { ContentKindError, MissingMarkerPricingError } from '../errors.ts';\nimport type { ElisionPlan, MarkerPricing, PlanInput, PlannedElision, Planner } from '../types.ts';\n\nimport { predictOutputBytes, savingBytes } from './budget.ts';\nimport { probeKind } from './kind.ts';\n\nexport const DIFF_PLANNER_ID = 'diff/v1';\n\n/**\n * The three rules, coarsest first: whole files the focus never touches; hunks inside a\n * file it does; and, inside a hunk it does, the line runs no match sits near.\n */\nexport const FILE_COLLAPSE_RULE = 'file-collapse';\nexport const HUNK_COLLAPSE_RULE = 'hunk-collapse';\nexport const HUNK_WINDOW_RULE = 'hunk-window';\n\n/** Context lines kept either side of a match inside a matched hunk, tried in order. */\nconst WINDOW_LADDER: readonly number[] = [4, 3, 2, 1, 0];\n\n/** Never collapse a run inside a hunk shorter than this. */\nconst MIN_RUN_LINES = 3;\n\nexport interface DiffPlannerOptions {\n  /** Focus matching is substring, case-insensitive by default. */\n  readonly caseSensitive?: boolean;\n}\n\n/**\n * The diff planner: **files and hunks are the units, not lines.**\n *\n * A unified diff has a structure a line planner cannot see: files, each with a header\n * and hunks. Under a focus, a file none of whose hunks carry a term collapses whole —\n * one marker per run of such files, the outline naming their paths — and inside a file\n * that does match, the hunks that do not collapse as a run while the header survives\n * verbatim — and inside a hunk that does match, the line runs no match sits near\n * collapse as a window (the lexical planner's move, confined to the hunk, with the same\n * context ladder under budget pressure), so the planner never keeps more of a hunk than\n * a line planner would. With no focus every file header is kept and each file's hunks\n * collapse to one marker, so the survivor is the diff's table of contents. Measured on\n * the bench's real diff (`git-diff`): every hunk mentioned the focus term, and without\n * the window rule this planner cut nothing where lexical cut to 1516 B.\n *\n * Refuses text without a unified-diff header shape ({@link ContentKindError}).\n */\nexport class DiffPlanner implements Planner {\n  readonly id = DIFF_PLANNER_ID;\n  readonly #options: DiffPlannerOptions;\n\n  constructor(options: DiffPlannerOptions = {}) {\n    this.#options = options;\n  }\n\n  plan(input: PlanInput): Promise<ElisionPlan> {\n    return Promise.resolve(planDiff(input, this.#options));\n  }\n}\n\ninterface Line {\n  readonly start: number;\n  /** One past the last byte, excluding the newline. */\n  readonly end: number;\n  readonly text: string;\n}\n\ninterface Hunk {\n  readonly start: number;\n  readonly end: number;\n  /** The `@@ … @@` header, without any trailing function context. */\n  readonly header: string;\n  readonly text: string;\n  /** The body lines after the header, for the window rule. */\n  readonly lines: readonly Line[];\n}\n\ninterface FileDiff {\n  readonly start: number;\n  readonly end: number;\n  readonly path: string;\n  readonly headerText: string;\n  readonly hunks: readonly Hunk[];\n}\n\n/**\n * The synchronous core. Deterministic; every candidate priced through the input's\n * {@link MarkerPricing}. Byte offsets come straight off the UTF-8 line split, so no\n * conversion is needed.\n *\n * @throws {ContentKindError} when the text carries no unified-diff header.\n */\nexport function planDiff(input: PlanInput, options: DiffPlannerOptions = {}): ElisionPlan {\n  const pricing = requirePricing(input);\n  if (probeKind(input.text) !== 'diff') {\n    throw new ContentKindError(\n      `smelt: the diff planner was asked to plan text with no unified-diff header ` +\n        `(a \"diff --git\" line, or \"--- \"/\"+++ \" lines followed by a hunk). It refuses ` +\n        `rather than approximating — output labelled ${DIFF_PLANNER_ID} that was ` +\n        `really line windows would be undetectable from outside. Use \"lexical\", or ` +\n        `\"auto\" to pick by content.`,\n    );\n  }\n  const files = parseFiles(splitLines(input.text));\n  const caseSensitive = options.caseSensitive ?? false;\n  const focus = (input.focus ?? []).filter((term) => term.length > 0);\n  const needles = caseSensitive ? focus : focus.map((term) => term.toLowerCase());\n  const matches = (text: string): boolean => {\n    const haystack = caseSensitive ? text : text.toLowerCase();\n    return needles.some((needle) => haystack.includes(needle));\n  };\n\n  /** The file and hunk collapses — the same at every rung of the window ladder. */\n  const fixed: PlannedElision[] = [];\n  /** The hunks the focus reached, where the window rule runs. */\n  const matched: { readonly file: FileDiff; readonly hunk: Hunk }[] = [];\n  const push = (candidate: PlannedElision): void => {\n    if (savingBytes(candidate, pricing) > 0) fixed.push(candidate);\n  };\n\n  const collapseFiles = (run: readonly FileDiff[]): void => {\n    if (run.length === 0) return;\n    const hunks = run.reduce((n, file) => n + file.hunks.length, 0);\n    push({\n      range: { start: run[0]!.start, end: run[run.length - 1]!.end },\n      reason: {\n        rule: FILE_COLLAPSE_RULE,\n        explanation:\n          `collapsed ${String(run.length)} ${run.length === 1 ? 'file' : 'files'} ` +\n          `(${String(hunks)} ${hunks === 1 ? 'hunk' : 'hunks'})`,\n      },\n      names: run.map((file) => file.path),\n    });\n  };\n\n  const collapseHunks = (file: FileDiff, run: readonly Hunk[]): void => {\n    if (run.length === 0) return;\n    push({\n      range: { start: run[0]!.start, end: run[run.length - 1]!.end },\n      reason: {\n        rule: HUNK_COLLAPSE_RULE,\n        explanation: `collapsed ${String(run.length)} ${run.length === 1 ? 'hunk' : 'hunks'} of ${file.path}`,\n      },\n      names: run.map((hunk) => hunk.header),\n    });\n  };\n\n  let fileRun: FileDiff[] = [];\n  for (const file of files) {\n    const fileKept =\n      needles.length === 0 ||\n      matches(file.headerText) ||\n      file.hunks.some((hunk) => matches(hunk.text));\n    if (!fileKept) {\n      fileRun.push(file);\n      continue;\n    }\n    collapseFiles(fileRun);\n    fileRun = [];\n    let hunkRun: Hunk[] = [];\n    for (const hunk of file.hunks) {\n      if (needles.length > 0 && matches(hunk.text)) {\n        collapseHunks(file, hunkRun);\n        hunkRun = [];\n        matched.push({ file, hunk });\n      } else {\n        hunkRun.push(hunk);\n      }\n    }\n    collapseHunks(file, hunkRun);\n  }\n  collapseFiles(fileRun);\n\n  // The window rule inside every matched hunk, tried with less context at each rung\n  // until the plan fits — the lexical ladder, confined to hunks. The first rung that\n  // fits wins; if none does, the tightest is returned over budget, as it came back.\n  const inputBytes = Buffer.byteLength(input.text, 'utf8');\n  const attempts = WINDOW_LADDER.map((context) => [\n    ...fixed,\n    ...matched.flatMap(({ file, hunk }) => windowsIn(file, hunk, context, matches, pricing)),\n  ]);\n  const elisions =\n    attempts.find((plan) => predictOutputBytes(inputBytes, plan, pricing) <= input.budgetBytes) ??\n    attempts[attempts.length - 1]!;\n\n  return { planner: DIFF_PLANNER_ID, language: input.language, elisions };\n}\n\n/**\n * Inside one matched hunk: keep every line within `context` of a matching line, and\n * collapse each run of the rest that is at least {@link MIN_RUN_LINES} long and pays\n * for its marker. The header line is never part of a run — it is what makes the\n * survivor still read as a hunk.\n */\nfunction windowsIn(\n  file: FileDiff,\n  hunk: Hunk,\n  context: number,\n  matches: (text: string) => boolean,\n  pricing: MarkerPricing,\n): readonly PlannedElision[] {\n  const lines = hunk.lines;\n  const keep: boolean[] = lines.map(() => false);\n  for (let i = 0; i < lines.length; i += 1) {\n    if (!matches(lines[i]!.text)) continue;\n    for (let j = Math.max(0, i - context); j <= Math.min(lines.length - 1, i + context); j += 1) {\n      keep[j] = true;\n    }\n  }\n  const out: PlannedElision[] = [];\n  let runStart = -1;\n  const flush = (endExclusive: number): void => {\n    if (runStart < 0) return;\n    const count = endExclusive - runStart;\n    const range = { start: lines[runStart]!.start, end: lines[endExclusive - 1]!.end };\n    runStart = -1;\n    if (count < MIN_RUN_LINES) return;\n    const candidate: PlannedElision = {\n      range,\n      reason: {\n        rule: HUNK_WINDOW_RULE,\n        explanation: `collapsed ${String(count)} lines of a hunk of ${file.path}`,\n      },\n    };\n    if (savingBytes(candidate, pricing) > 0) out.push(candidate);\n  };\n  for (let i = 0; i < lines.length; i += 1) {\n    if (keep[i] === true) flush(i);\n    else if (runStart < 0) runStart = i;\n  }\n  flush(lines.length);\n  return out;\n}\n\nfunction requirePricing(input: PlanInput): MarkerPricing {\n  const pricing: MarkerPricing | undefined = input.pricing;\n  if (pricing === undefined) throw new MissingMarkerPricingError(DIFF_PLANNER_ID);\n  return pricing;\n}\n\nfunction splitLines(text: string): readonly Line[] {\n  const lines: Line[] = [];\n  let start = 0;\n  for (const content of text.split('\\n')) {\n    const end = start + Buffer.byteLength(content, 'utf8');\n    lines.push({ start, end, text: content });\n    start = end + 1;\n  }\n  return lines;\n}\n\n/** True where a file starts: a `diff --git` line, or a `---` line followed by `+++`. */\nfunction startsFile(lines: readonly Line[], i: number): boolean {\n  const line = lines[i]!.text;\n  if (line.startsWith('diff --git ')) return true;\n  return (\n    line.startsWith('--- ') &&\n    lines[i + 1]?.text.startsWith('+++ ') === true &&\n    !(i > 0 && lines[i - 1]!.text.startsWith('diff --git '))\n  );\n}\n\nfunction parseFiles(lines: readonly Line[]): readonly FileDiff[] {\n  const starts: number[] = [];\n  for (let i = 0; i < lines.length; i += 1) {\n    if (startsFile(lines, i)) {\n      // A `---` line directly under a `diff --git` line belongs to that file's header.\n      if (lines[i]!.text.startsWith('--- ') && starts.at(-1) !== undefined) {\n        const previous = starts.at(-1)!;\n        const between = lines.slice(previous, i).map((line) => line.text);\n        if (\n          between[0]?.startsWith('diff --git ') === true &&\n          !between.some((l) => l.startsWith('@@ '))\n        )\n          continue;\n      }\n      starts.push(i);\n    }\n  }\n  return starts.map((from, index) => {\n    const to = index + 1 < starts.length ? starts[index + 1]! : lastContentLine(lines) + 1;\n    const fileLines = lines.slice(from, to);\n    const firstHunk = fileLines.findIndex((line) => line.text.startsWith('@@ '));\n    const headerLines = firstHunk === -1 ? fileLines : fileLines.slice(0, firstHunk);\n    const hunks: Hunk[] = [];\n    if (firstHunk !== -1) {\n      let hunkStart = firstHunk;\n      for (let i = firstHunk + 1; i <= fileLines.length; i += 1) {\n        if (i === fileLines.length || fileLines[i]!.text.startsWith('@@ ')) {\n          const body = fileLines.slice(hunkStart, i);\n          hunks.push({\n            start: body[0]!.start,\n            end: body[body.length - 1]!.end,\n            header: hunkHeader(body[0]!.text),\n            text: body.map((line) => line.text).join('\\n'),\n            lines: body.slice(1),\n          });\n          hunkStart = i;\n        }\n      }\n    }\n    return {\n      start: fileLines[0]!.start,\n      end: fileLines[fileLines.length - 1]!.end,\n      path: pathOf(headerLines.map((line) => line.text)),\n      headerText: headerLines.map((line) => line.text).join('\\n'),\n      hunks,\n    };\n  });\n}\n\n/** The index of the last line with content, so a trailing newline is never claimed. */\nfunction lastContentLine(lines: readonly Line[]): number {\n  let i = lines.length - 1;\n  while (i > 0 && lines[i]!.text === '') i -= 1;\n  return i;\n}\n\nfunction hunkHeader(line: string): string {\n  const close = line.indexOf('@@', 2);\n  return close === -1 ? line : line.slice(0, close + 2);\n}\n\n/** The path a file header names: the `+++` side first, else the `diff --git` b-side. */\nfunction pathOf(header: readonly string[]): string {\n  const plus = header.find((line) => line.startsWith('+++ '));\n  if (plus !== undefined && plus !== '+++ /dev/null') return stripPrefix(plus.slice(4));\n  const minus = header.find((line) => line.startsWith('--- '));\n  if (minus !== undefined && minus !== '--- /dev/null') return stripPrefix(minus.slice(4));\n  const git = header.find((line) => line.startsWith('diff --git '));\n  if (git !== undefined) {\n    const parts = git.slice('diff --git '.length).split(' b/');\n    return parts[parts.length - 1] ?? git;\n  }\n  return '<unknown>';\n}\n\nfunction stripPrefix(path: string): string {\n  const trimmed = path.split('\\t')[0]!;\n  return trimmed.startsWith('a/') || trimmed.startsWith('b/') ? trimmed.slice(2) : trimmed;\n}\n"]}

package/dist/plan/diff.js

@@ -0,0 +1,284 @@
+import { ContentKindError, MissingMarkerPricingError } from '../errors.js';
+import { predictOutputBytes, savingBytes } from './budget.js';
+import { probeKind } from './kind.js';
+export const DIFF_PLANNER_ID = 'diff/v1';
+/**
+ * The three rules, coarsest first: whole files the focus never touches; hunks inside a
+ * file it does; and, inside a hunk it does, the line runs no match sits near.
+ */
+export const FILE_COLLAPSE_RULE = 'file-collapse';
+export const HUNK_COLLAPSE_RULE = 'hunk-collapse';
+export const HUNK_WINDOW_RULE = 'hunk-window';
+/** Context lines kept either side of a match inside a matched hunk, tried in order. */
+const WINDOW_LADDER = [4, 3, 2, 1, 0];
+/** Never collapse a run inside a hunk shorter than this. */
+const MIN_RUN_LINES = 3;
+/**
+ * The diff planner: **files and hunks are the units, not lines.**
+ *
+ * A unified diff has a structure a line planner cannot see: files, each with a header
+ * and hunks. Under a focus, a file none of whose hunks carry a term collapses whole —
+ * one marker per run of such files, the outline naming their paths — and inside a file
+ * that does match, the hunks that do not collapse as a run while the header survives
+ * verbatim — and inside a hunk that does match, the line runs no match sits near
+ * collapse as a window (the lexical planner's move, confined to the hunk, with the same
+ * context ladder under budget pressure), so the planner never keeps more of a hunk than
+ * a line planner would. With no focus every file header is kept and each file's hunks
+ * collapse to one marker, so the survivor is the diff's table of contents. Measured on
+ * the bench's real diff (`git-diff`): every hunk mentioned the focus term, and without
+ * the window rule this planner cut nothing where lexical cut to 1516 B.
+ *
+ * Refuses text without a unified-diff header shape ({@link ContentKindError}).
+ */
+export class DiffPlanner {
+    id = DIFF_PLANNER_ID;
+    #options;
+    constructor(options = {}) {
+        this.#options = options;
+    }
+    plan(input) {
+        return Promise.resolve(planDiff(input, this.#options));
+    }
+}
+/**
+ * The synchronous core. Deterministic; every candidate priced through the input's
+ * {@link MarkerPricing}. Byte offsets come straight off the UTF-8 line split, so no
+ * conversion is needed.
+ *
+ * @throws {ContentKindError} when the text carries no unified-diff header.
+ */
+export function planDiff(input, options = {}) {
+    const pricing = requirePricing(input);
+    if (probeKind(input.text) !== 'diff') {
+        throw new ContentKindError(`smelt: the diff planner was asked to plan text with no unified-diff header ` +
+            `(a "diff --git" line, or "--- "/"+++ " lines followed by a hunk). It refuses ` +
+            `rather than approximating — output labelled ${DIFF_PLANNER_ID} that was ` +
+            `really line windows would be undetectable from outside. Use "lexical", or ` +
+            `"auto" to pick by content.`);
+    }
+    const files = parseFiles(splitLines(input.text));
+    const caseSensitive = options.caseSensitive ?? false;
+    const focus = (input.focus ?? []).filter((term) => term.length > 0);
+    const needles = caseSensitive ? focus : focus.map((term) => term.toLowerCase());
+    const matches = (text) => {
+        const haystack = caseSensitive ? text : text.toLowerCase();
+        return needles.some((needle) => haystack.includes(needle));
+    };
+    /** The file and hunk collapses — the same at every rung of the window ladder. */
+    const fixed = [];
+    /** The hunks the focus reached, where the window rule runs. */
+    const matched = [];
+    const push = (candidate) => {
+        if (savingBytes(candidate, pricing) > 0)
+            fixed.push(candidate);
+    };
+    const collapseFiles = (run) => {
+        if (run.length === 0)
+            return;
+        const hunks = run.reduce((n, file) => n + file.hunks.length, 0);
+        push({
+            range: { start: run[0].start, end: run[run.length - 1].end },
+            reason: {
+                rule: FILE_COLLAPSE_RULE,
+                explanation: `collapsed ${String(run.length)} ${run.length === 1 ? 'file' : 'files'} ` +
+                    `(${String(hunks)} ${hunks === 1 ? 'hunk' : 'hunks'})`,
+            },
+            names: run.map((file) => file.path),
+        });
+    };
+    const collapseHunks = (file, run) => {
+        if (run.length === 0)
+            return;
+        push({
+            range: { start: run[0].start, end: run[run.length - 1].end },
+            reason: {
+                rule: HUNK_COLLAPSE_RULE,
+                explanation: `collapsed ${String(run.length)} ${run.length === 1 ? 'hunk' : 'hunks'} of ${file.path}`,
+            },
+            names: run.map((hunk) => hunk.header),
+        });
+    };
+    let fileRun = [];
+    for (const file of files) {
+        const fileKept = needles.length === 0 ||
+            matches(file.headerText) ||
+            file.hunks.some((hunk) => matches(hunk.text));
+        if (!fileKept) {
+            fileRun.push(file);
+            continue;
+        }
+        collapseFiles(fileRun);
+        fileRun = [];
+        let hunkRun = [];
+        for (const hunk of file.hunks) {
+            if (needles.length > 0 && matches(hunk.text)) {
+                collapseHunks(file, hunkRun);
+                hunkRun = [];
+                matched.push({ file, hunk });
+            }
+            else {
+                hunkRun.push(hunk);
+            }
+        }
+        collapseHunks(file, hunkRun);
+    }
+    collapseFiles(fileRun);
+    // The window rule inside every matched hunk, tried with less context at each rung
+    // until the plan fits — the lexical ladder, confined to hunks. The first rung that
+    // fits wins; if none does, the tightest is returned over budget, as it came back.
+    const inputBytes = Buffer.byteLength(input.text, 'utf8');
+    const attempts = WINDOW_LADDER.map((context) => [
+        ...fixed,
+        ...matched.flatMap(({ file, hunk }) => windowsIn(file, hunk, context, matches, pricing)),
+    ]);
+    const elisions = attempts.find((plan) => predictOutputBytes(inputBytes, plan, pricing) <= input.budgetBytes) ??
+        attempts[attempts.length - 1];
+    return { planner: DIFF_PLANNER_ID, language: input.language, elisions };
+}
+/**
+ * Inside one matched hunk: keep every line within `context` of a matching line, and
+ * collapse each run of the rest that is at least {@link MIN_RUN_LINES} long and pays
+ * for its marker. The header line is never part of a run — it is what makes the
+ * survivor still read as a hunk.
+ */
+function windowsIn(file, hunk, context, matches, pricing) {
+    const lines = hunk.lines;
+    const keep = lines.map(() => false);
+    for (let i = 0; i < lines.length; i += 1) {
+        if (!matches(lines[i].text))
+            continue;
+        for (let j = Math.max(0, i - context); j <= Math.min(lines.length - 1, i + context); j += 1) {
+            keep[j] = true;
+        }
+    }
+    const out = [];
+    let runStart = -1;
+    const flush = (endExclusive) => {
+        if (runStart < 0)
+            return;
+        const count = endExclusive - runStart;
+        const range = { start: lines[runStart].start, end: lines[endExclusive - 1].end };
+        runStart = -1;
+        if (count < MIN_RUN_LINES)
+            return;
+        const candidate = {
+            range,
+            reason: {
+                rule: HUNK_WINDOW_RULE,
+                explanation: `collapsed ${String(count)} lines of a hunk of ${file.path}`,
+            },
+        };
+        if (savingBytes(candidate, pricing) > 0)
+            out.push(candidate);
+    };
+    for (let i = 0; i < lines.length; i += 1) {
+        if (keep[i] === true)
+            flush(i);
+        else if (runStart < 0)
+            runStart = i;
+    }
+    flush(lines.length);
+    return out;
+}
+function requirePricing(input) {
+    const pricing = input.pricing;
+    if (pricing === undefined)
+        throw new MissingMarkerPricingError(DIFF_PLANNER_ID);
+    return pricing;
+}
+function splitLines(text) {
+    const lines = [];
+    let start = 0;
+    for (const content of text.split('\n')) {
+        const end = start + Buffer.byteLength(content, 'utf8');
+        lines.push({ start, end, text: content });
+        start = end + 1;
+    }
+    return lines;
+}
+/** True where a file starts: a `diff --git` line, or a `---` line followed by `+++`. */
+function startsFile(lines, i) {
+    const line = lines[i].text;
+    if (line.startsWith('diff --git '))
+        return true;
+    return (line.startsWith('--- ') &&
+        lines[i + 1]?.text.startsWith('+++ ') === true &&
+        !(i > 0 && lines[i - 1].text.startsWith('diff --git ')));
+}
+function parseFiles(lines) {
+    const starts = [];
+    for (let i = 0; i < lines.length; i += 1) {
+        if (startsFile(lines, i)) {
+            // A `---` line directly under a `diff --git` line belongs to that file's header.
+            if (lines[i].text.startsWith('--- ') && starts.at(-1) !== undefined) {
+                const previous = starts.at(-1);
+                const between = lines.slice(previous, i).map((line) => line.text);
+                if (between[0]?.startsWith('diff --git ') === true &&
+                    !between.some((l) => l.startsWith('@@ ')))
+                    continue;
+            }
+            starts.push(i);
+        }
+    }
+    return starts.map((from, index) => {
+        const to = index + 1 < starts.length ? starts[index + 1] : lastContentLine(lines) + 1;
+        const fileLines = lines.slice(from, to);
+        const firstHunk = fileLines.findIndex((line) => line.text.startsWith('@@ '));
+        const headerLines = firstHunk === -1 ? fileLines : fileLines.slice(0, firstHunk);
+        const hunks = [];
+        if (firstHunk !== -1) {
+            let hunkStart = firstHunk;
+            for (let i = firstHunk + 1; i <= fileLines.length; i += 1) {
+                if (i === fileLines.length || fileLines[i].text.startsWith('@@ ')) {
+                    const body = fileLines.slice(hunkStart, i);
+                    hunks.push({
+                        start: body[0].start,
+                        end: body[body.length - 1].end,
+                        header: hunkHeader(body[0].text),
+                        text: body.map((line) => line.text).join('\n'),
+                        lines: body.slice(1),
+                    });
+                    hunkStart = i;
+                }
+            }
+        }
+        return {
+            start: fileLines[0].start,
+            end: fileLines[fileLines.length - 1].end,
+            path: pathOf(headerLines.map((line) => line.text)),
+            headerText: headerLines.map((line) => line.text).join('\n'),
+            hunks,
+        };
+    });
+}
+/** The index of the last line with content, so a trailing newline is never claimed. */
+function lastContentLine(lines) {
+    let i = lines.length - 1;
+    while (i > 0 && lines[i].text === '')
+        i -= 1;
+    return i;
+}
+function hunkHeader(line) {
+    const close = line.indexOf('@@', 2);
+    return close === -1 ? line : line.slice(0, close + 2);
+}
+/** The path a file header names: the `+++` side first, else the `diff --git` b-side. */
+function pathOf(header) {
+    const plus = header.find((line) => line.startsWith('+++ '));
+    if (plus !== undefined && plus !== '+++ /dev/null')
+        return stripPrefix(plus.slice(4));
+    const minus = header.find((line) => line.startsWith('--- '));
+    if (minus !== undefined && minus !== '--- /dev/null')
+        return stripPrefix(minus.slice(4));
+    const git = header.find((line) => line.startsWith('diff --git '));
+    if (git !== undefined) {
+        const parts = git.slice('diff --git '.length).split(' b/');
+        return parts[parts.length - 1] ?? git;
+    }
+    return '<unknown>';
+}
+function stripPrefix(path) {
+    const trimmed = path.split('\t')[0];
+    return trimmed.startsWith('a/') || trimmed.startsWith('b/') ? trimmed.slice(2) : trimmed;
+}
+//# sourceMappingURL=diff.js.map

package/dist/plan/diff.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"diff.js","sourceRoot":"","sources":["../../src/plan/diff.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,gBAAgB,EAAE,yBAAyB,EAAE,MAAM,cAAc,CAAC;AAG3E,OAAO,EAAE,kBAAkB,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AAC9D,OAAO,EAAE,SAAS,EAAE,MAAM,WAAW,CAAC;AAEtC,MAAM,CAAC,MAAM,eAAe,GAAG,SAAS,CAAC;AAEzC;;;GAGG;AACH,MAAM,CAAC,MAAM,kBAAkB,GAAG,eAAe,CAAC;AAClD,MAAM,CAAC,MAAM,kBAAkB,GAAG,eAAe,CAAC;AAClD,MAAM,CAAC,MAAM,gBAAgB,GAAG,aAAa,CAAC;AAE9C,uFAAuF;AACvF,MAAM,aAAa,GAAsB,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC,CAAC;AAEzD,4DAA4D;AAC5D,MAAM,aAAa,GAAG,CAAC,CAAC;AAOxB;;;;;;;;;;;;;;;;GAgBG;AACH,MAAM,OAAO,WAAW;IACb,EAAE,GAAG,eAAe,CAAC;IACrB,QAAQ,CAAqB;IAEtC,YAAY,OAAO,GAAuB,EAAE;QAC1C,IAAI,CAAC,QAAQ,GAAG,OAAO,CAAC;IAC1B,CAAC;IAED,IAAI,CAAC,KAAgB;QACnB,OAAO,OAAO,CAAC,OAAO,CAAC,QAAQ,CAAC,KAAK,EAAE,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC;IACzD,CAAC;CACF;AA2BD;;;;;;GAMG;AACH,MAAM,UAAU,QAAQ,CAAC,KAAgB,EAAE,OAAO,GAAuB,EAAE;IACzE,MAAM,OAAO,GAAG,cAAc,CAAC,KAAK,CAAC,CAAC;IACtC,IAAI,SAAS,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,MAAM,EAAE,CAAC;QACrC,MAAM,IAAI,gBAAgB,CACxB,6EAA6E;YAC3E,+EAA+E;YAC/E,+CAA+C,eAAe,YAAY;YAC1E,4EAA4E;YAC5E,4BAA4B,CAC/B,CAAC;IACJ,CAAC;IACD,MAAM,KAAK,GAAG,UAAU,CAAC,UAAU,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC;IACjD,MAAM,aAAa,GAAG,OAAO,CAAC,aAAa,IAAI,KAAK,CAAC;IACrD,MAAM,KAAK,GAAG,CAAC,KAAK,CAAC,KAAK,IAAI,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;IACpE,MAAM,OAAO,GAAG,aAAa,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,WAAW,EAAE,CAAC,CAAC;IAChF,MAAM,OAAO,GAAG,CAAC,IAAY,EAAW,EAAE;QACxC,MAAM,QAAQ,GAAG,aAAa,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,WAAW,EAAE,CAAC;QAC3D,OAAO,OAAO,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,QAAQ,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC;IAC7D,CAAC,CAAC;IAEF,iFAAiF;IACjF,MAAM,KAAK,GAAqB,EAAE,CAAC;IACnC,+DAA+D;IAC/D,MAAM,OAAO,GAAuD,EAAE,CAAC;IACvE,MAAM,IAAI,GAAG,CAAC,SAAyB,EAAQ,EAAE;QAC/C,IAAI,WAAW,CAAC,SAAS,EAAE,OAAO,CAAC,GAAG,CAAC;YAAE,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;IACjE,CAAC,CAAC;IAEF,MAAM,aAAa,GAAG,CAAC,GAAwB,EAAQ,EAAE;QACvD,IAAI,GAAG,CAAC,MAAM,KAAK,CAAC;YAAE,OAAO;QAC7B,MAAM,KAAK,GAAG,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,EAAE,CAAC,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;QAChE,IAAI,CAAC;YACH,KAAK,EAAE,EAAE,KAAK,EAAE,GAAG,CAAC,CAAC,CAAE,CAAC,KAAK,EAAE,GAAG,EAAE,GAAG,CAAC,GAAG,CAAC,MAAM,GAAG,CAAC,CAAE,CAAC,GAAG,EAAE;YAC9D,MAAM,EAAE;gBACN,IAAI,EAAE,kBAAkB;gBACxB,WAAW,EACT,aAAa,MAAM,CAAC,GAAG,CAAC,MAAM,CAAC,IAAI,GAAG,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,OAAO,GAAG;oBACzE,IAAI,MAAM,CAAC,KAAK,CAAC,IAAI,KAAK,KAAK,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,OAAO,GAAG;aACzD;YACD,KAAK,EAAE,GAAG,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC;SACpC,CAAC,CAAC;IACL,CAAC,CAAC;IAEF,MAAM,aAAa,GAAG,CAAC,IAAc,EAAE,GAAoB,EAAQ,EAAE;QACnE,IAAI,GAAG,CAAC,MAAM,KAAK,CAAC;YAAE,OAAO;QAC7B,IAAI,CAAC;YACH,KAAK,EAAE,EAAE,KAAK,EAAE,GAAG,CAAC,CAAC,CAAE,CAAC,KAAK,EAAE,GAAG,EAAE,GAAG,CAAC,GAAG,CAAC,MAAM,GAAG,CAAC,CAAE,CAAC,GAAG,EAAE;YAC9D,MAAM,EAAE;gBACN,IAAI,EAAE,kBAAkB;gBACxB,WAAW,EAAE,aAAa,MAAM,CAAC,GAAG,CAAC,MAAM,CAAC,IAAI,GAAG,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,OAAO,OAAO,IAAI,CAAC,IAAI,EAAE;aACtG;YACD,KAAK,EAAE,GAAG,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,MAAM,CAAC;SACtC,CAAC,CAAC;IACL,CAAC,CAAC;IAEF,IAAI,OAAO,GAAe,EAAE,CAAC;IAC7B,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,MAAM,QAAQ,GACZ,OAAO,CAAC,MAAM,KAAK,CAAC;YACpB,OAAO,CAAC,IAAI,CAAC,UAAU,CAAC;YACxB,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC;QAChD,IAAI,CAAC,QAAQ,EAAE,CAAC;YACd,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YACnB,SAAS;QACX,CAAC;QACD,aAAa,CAAC,OAAO,CAAC,CAAC;QACvB,OAAO,GAAG,EAAE,CAAC;QACb,IAAI,OAAO,GAAW,EAAE,CAAC;QACzB,KAAK,MAAM,IAAI,IAAI,IAAI,CAAC,KAAK,EAAE,CAAC;YAC9B,IAAI,OAAO,CAAC,MAAM,GAAG,CAAC,IAAI,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;gBAC7C,aAAa,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;gBAC7B,OAAO,GAAG,EAAE,CAAC;gBACb,OAAO,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC;YAC/B,CAAC;iBAAM,CAAC;gBACN,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YACrB,CAAC;QACH,CAAC;QACD,aAAa,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;IAC/B,CAAC;IACD,aAAa,CAAC,OAAO,CAAC,CAAC;IAEvB,kFAAkF;IAClF,mFAAmF;IACnF,kFAAkF;IAClF,MAAM,UAAU,GAAG,MAAM,CAAC,UAAU,CAAC,KAAK,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;IACzD,MAAM,QAAQ,GAAG,aAAa,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC;QAC9C,GAAG,KAAK;QACR,GAAG,OAAO,CAAC,OAAO,CAAC,CAAC,EAAE,IAAI,EAAE,IAAI,EAAE,EAAE,EAAE,CAAC,SAAS,CAAC,IAAI,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC;KACzF,CAAC,CAAC;IACH,MAAM,QAAQ,GACZ,QAAQ,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,kBAAkB,CAAC,UAAU,EAAE,IAAI,EAAE,OAAO,CAAC,IAAI,KAAK,CAAC,WAAW,CAAC;QAC3F,QAAQ,CAAC,QAAQ,CAAC,MAAM,GAAG,CAAC,CAAE,CAAC;IAEjC,OAAO,EAAE,OAAO,EAAE,eAAe,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ,EAAE,QAAQ,EAAE,CAAC;AAC1E,CAAC;AAED;;;;;GAKG;AACH,SAAS,SAAS,CAChB,IAAc,EACd,IAAU,EACV,OAAe,EACf,OAAkC,EAClC,OAAsB;IAEtB,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC;IACzB,MAAM,IAAI,GAAc,KAAK,CAAC,GAAG,CAAC,GAAG,EAAE,CAAC,KAAK,CAAC,CAAC;IAC/C,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC;QACzC,IAAI,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,CAAE,CAAC,IAAI,CAAC;YAAE,SAAS;QACvC,KAAK,IAAI,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,CAAC,GAAG,OAAO,CAAC,EAAE,CAAC,IAAI,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC,GAAG,OAAO,CAAC,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC;YAC5F,IAAI,CAAC,CAAC,CAAC,GAAG,IAAI,CAAC;QACjB,CAAC;IACH,CAAC;IACD,MAAM,GAAG,GAAqB,EAAE,CAAC;IACjC,IAAI,QAAQ,GAAG,CAAC,CAAC,CAAC;IAClB,MAAM,KAAK,GAAG,CAAC,YAAoB,EAAQ,EAAE;QAC3C,IAAI,QAAQ,GAAG,CAAC;YAAE,OAAO;QACzB,MAAM,KAAK,GAAG,YAAY,GAAG,QAAQ,CAAC;QACtC,MAAM,KAAK,GAAG,EAAE,KAAK,EAAE,KAAK,CAAC,QAAQ,CAAE,CAAC,KAAK,EAAE,GAAG,EAAE,KAAK,CAAC,YAAY,GAAG,CAAC,CAAE,CAAC,GAAG,EAAE,CAAC;QACnF,QAAQ,GAAG,CAAC,CAAC,CAAC;QACd,IAAI,KAAK,GAAG,aAAa;YAAE,OAAO;QAClC,MAAM,SAAS,GAAmB;YAChC,KAAK;YACL,MAAM,EAAE;gBACN,IAAI,EAAE,gBAAgB;gBACtB,WAAW,EAAE,aAAa,MAAM,CAAC,KAAK,CAAC,uBAAuB,IAAI,CAAC,IAAI,EAAE;aAC1E;SACF,CAAC;QACF,IAAI,WAAW,CAAC,SAAS,EAAE,OAAO,CAAC,GAAG,CAAC;YAAE,GAAG,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;IAC/D,CAAC,CAAC;IACF,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC;QACzC,IAAI,IAAI,CAAC,CAAC,CAAC,KAAK,IAAI;YAAE,KAAK,CAAC,CAAC,CAAC,CAAC;aAC1B,IAAI,QAAQ,GAAG,CAAC;YAAE,QAAQ,GAAG,CAAC,CAAC;IACtC,CAAC;IACD,KAAK,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;IACpB,OAAO,GAAG,CAAC;AACb,CAAC;AAED,SAAS,cAAc,CAAC,KAAgB;IACtC,MAAM,OAAO,GAA8B,KAAK,CAAC,OAAO,CAAC;IACzD,IAAI,OAAO,KAAK,SAAS;QAAE,MAAM,IAAI,yBAAyB,CAAC,eAAe,CAAC,CAAC;IAChF,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,SAAS,UAAU,CAAC,IAAY;IAC9B,MAAM,KAAK,GAAW,EAAE,CAAC;IACzB,IAAI,KAAK,GAAG,CAAC,CAAC;IACd,KAAK,MAAM,OAAO,IAAI,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC;QACvC,MAAM,GAAG,GAAG,KAAK,GAAG,MAAM,CAAC,UAAU,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;QACvD,KAAK,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,GAAG,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC;QAC1C,KAAK,GAAG,GAAG,GAAG,CAAC,CAAC;IAClB,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC;AAED,wFAAwF;AACxF,SAAS,UAAU,CAAC,KAAsB,EAAE,CAAS;IACnD,MAAM,IAAI,GAAG,KAAK,CAAC,CAAC,CAAE,CAAC,IAAI,CAAC;IAC5B,IAAI,IAAI,CAAC,UAAU,CAAC,aAAa,CAAC;QAAE,OAAO,IAAI,CAAC;IAChD,OAAO,CACL,IAAI,CAAC,UAAU,CAAC,MAAM,CAAC;QACvB,KAAK,CAAC,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,UAAU,CAAC,MAAM,CAAC,KAAK,IAAI;QAC9C,CAAC,CAAC,CAAC,GAAG,CAAC,IAAI,KAAK,CAAC,CAAC,GAAG,CAAC,CAAE,CAAC,IAAI,CAAC,UAAU,CAAC,aAAa,CAAC,CAAC,CACzD,CAAC;AACJ,CAAC;AAED,SAAS,UAAU,CAAC,KAAsB;IACxC,MAAM,MAAM,GAAa,EAAE,CAAC;IAC5B,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC;QACzC,IAAI,UAAU,CAAC,KAAK,EAAE,CAAC,CAAC,EAAE,CAAC;YACzB,iFAAiF;YACjF,IAAI,KAAK,CAAC,CAAC,CAAE,CAAC,IAAI,CAAC,UAAU,CAAC,MAAM,CAAC,IAAI,MAAM,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,KAAK,SAAS,EAAE,CAAC;gBACrE,MAAM,QAAQ,GAAG,MAAM,CAAC,EAAE,CAAC,CAAC,CAAC,CAAE,CAAC;gBAChC,MAAM,OAAO,GAAG,KAAK,CAAC,KAAK,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;gBAClE,IACE,OAAO,CAAC,CAAC,CAAC,EAAE,UAAU,CAAC,aAAa,CAAC,KAAK,IAAI;oBAC9C,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,UAAU,CAAC,KAAK,CAAC,CAAC;oBAEzC,SAAS;YACb,CAAC;YACD,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACjB,CAAC;IACH,CAAC;IACD,OAAO,MAAM,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE;QAChC,MAAM,EAAE,GAAG,KAAK,GAAG,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,GAAG,CAAC,CAAE,CAAC,CAAC,CAAC,eAAe,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;QACvF,MAAM,SAAS,GAAG,KAAK,CAAC,KAAK,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC;QACxC,MAAM,SAAS,GAAG,SAAS,CAAC,SAAS,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,KAAK,CAAC,CAAC,CAAC;QAC7E,MAAM,WAAW,GAAG,SAAS,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,EAAE,SAAS,CAAC,CAAC;QACjF,MAAM,KAAK,GAAW,EAAE,CAAC;QACzB,IAAI,SAAS,KAAK,CAAC,CAAC,EAAE,CAAC;YACrB,IAAI,SAAS,GAAG,SAAS,CAAC;YAC1B,KAAK,IAAI,CAAC,GAAG,SAAS,GAAG,CAAC,EAAE,CAAC,IAAI,SAAS,CAAC,MAAM,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC;gBAC1D,IAAI,CAAC,KAAK,SAAS,CAAC,MAAM,IAAI,SAAS,CAAC,CAAC,CAAE,CAAC,IAAI,CAAC,UAAU,CAAC,KAAK,CAAC,EAAE,CAAC;oBACnE,MAAM,IAAI,GAAG,SAAS,CAAC,KAAK,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC;oBAC3C,KAAK,CAAC,IAAI,CAAC;wBACT,KAAK,EAAE,IAAI,CAAC,CAAC,CAAE,CAAC,KAAK;wBACrB,GAAG,EAAE,IAAI,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,CAAE,CAAC,GAAG;wBAC/B,MAAM,EAAE,UAAU,CAAC,IAAI,CAAC,CAAC,CAAE,CAAC,IAAI,CAAC;wBACjC,IAAI,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC;wBAC9C,KAAK,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC;qBACrB,CAAC,CAAC;oBACH,SAAS,GAAG,CAAC,CAAC;gBAChB,CAAC;YACH,CAAC;QACH,CAAC;QACD,OAAO;YACL,KAAK,EAAE,SAAS,CAAC,CAAC,CAAE,CAAC,KAAK;YAC1B,GAAG,EAAE,SAAS,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,CAAE,CAAC,GAAG;YACzC,IAAI,EAAE,MAAM,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YAClD,UAAU,EAAE,WAAW,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC;YAC3D,KAAK;SACN,CAAC;IACJ,CAAC,CAAC,CAAC;AACL,CAAC;AAED,uFAAuF;AACvF,SAAS,eAAe,CAAC,KAAsB;IAC7C,IAAI,CAAC,GAAG,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC;IACzB,OAAO,CAAC,GAAG,CAAC,IAAI,KAAK,CAAC,CAAC,CAAE,CAAC,IAAI,KAAK,EAAE;QAAE,CAAC,IAAI,CAAC,CAAC;IAC9C,OAAO,CAAC,CAAC;AACX,CAAC;AAED,SAAS,UAAU,CAAC,IAAY;IAC9B,MAAM,KAAK,GAAG,IAAI,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC;IACpC,OAAO,KAAK,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,KAAK,GAAG,CAAC,CAAC,CAAC;AACxD,CAAC;AAED,wFAAwF;AACxF,SAAS,MAAM,CAAC,MAAyB;IACvC,MAAM,IAAI,GAAG,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC,CAAC;IAC5D,IAAI,IAAI,KAAK,SAAS,IAAI,IAAI,KAAK,eAAe;QAAE,OAAO,WAAW,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;IACtF,MAAM,KAAK,GAAG,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC,CAAC;IAC7D,IAAI,KAAK,KAAK,SAAS,IAAI,KAAK,KAAK,eAAe;QAAE,OAAO,WAAW,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;IACzF,MAAM,GAAG,GAAG,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,UAAU,CAAC,aAAa,CAAC,CAAC,CAAC;IAClE,IAAI,GAAG,KAAK,SAAS,EAAE,CAAC;QACtB,MAAM,KAAK,GAAG,GAAG,CAAC,KAAK,CAAC,aAAa,CAAC,MAAM,CAAC,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC;QAC3D,OAAO,KAAK,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,IAAI,GAAG,CAAC;IACxC,CAAC;IACD,OAAO,WAAW,CAAC;AACrB,CAAC;AAED,SAAS,WAAW,CAAC,IAAY;IAC/B,MAAM,OAAO,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,CAAE,CAAC;IACrC,OAAO,OAAO,CAAC,UAAU,CAAC,IAAI,CAAC,IAAI,OAAO,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC;AAC3F,CAAC","sourcesContent":["import { ContentKindError, MissingMarkerPricingError } from '../errors.ts';\nimport type { ElisionPlan, MarkerPricing, PlanInput, PlannedElision, Planner } from '../types.ts';\n\nimport { predictOutputBytes, savingBytes } from './budget.ts';\nimport { probeKind } from './kind.ts';\n\nexport const DIFF_PLANNER_ID = 'diff/v1';\n\n/**\n * The three rules, coarsest first: whole files the focus never touches; hunks inside a\n * file it does; and, inside a hunk it does, the line runs no match sits near.\n */\nexport const FILE_COLLAPSE_RULE = 'file-collapse';\nexport const HUNK_COLLAPSE_RULE = 'hunk-collapse';\nexport const HUNK_WINDOW_RULE = 'hunk-window';\n\n/** Context lines kept either side of a match inside a matched hunk, tried in order. */\nconst WINDOW_LADDER: readonly number[] = [4, 3, 2, 1, 0];\n\n/** Never collapse a run inside a hunk shorter than this. */\nconst MIN_RUN_LINES = 3;\n\nexport interface DiffPlannerOptions {\n  /** Focus matching is substring, case-insensitive by default. */\n  readonly caseSensitive?: boolean;\n}\n\n/**\n * The diff planner: **files and hunks are the units, not lines.**\n *\n * A unified diff has a structure a line planner cannot see: files, each with a header\n * and hunks. Under a focus, a file none of whose hunks carry a term collapses whole —\n * one marker per run of such files, the outline naming their paths — and inside a file\n * that does match, the hunks that do not collapse as a run while the header survives\n * verbatim — and inside a hunk that does match, the line runs no match sits near\n * collapse as a window (the lexical planner's move, confined to the hunk, with the same\n * context ladder under budget pressure), so the planner never keeps more of a hunk than\n * a line planner would. With no focus every file header is kept and each file's hunks\n * collapse to one marker, so the survivor is the diff's table of contents. Measured on\n * the bench's real diff (`git-diff`): every hunk mentioned the focus term, and without\n * the window rule this planner cut nothing where lexical cut to 1516 B.\n *\n * Refuses text without a unified-diff header shape ({@link ContentKindError}).\n */\nexport class DiffPlanner implements Planner {\n  readonly id = DIFF_PLANNER_ID;\n  readonly #options: DiffPlannerOptions;\n\n  constructor(options: DiffPlannerOptions = {}) {\n    this.#options = options;\n  }\n\n  plan(input: PlanInput): Promise<ElisionPlan> {\n    return Promise.resolve(planDiff(input, this.#options));\n  }\n}\n\ninterface Line {\n  readonly start: number;\n  /** One past the last byte, excluding the newline. */\n  readonly end: number;\n  readonly text: string;\n}\n\ninterface Hunk {\n  readonly start: number;\n  readonly end: number;\n  /** The `@@ … @@` header, without any trailing function context. */\n  readonly header: string;\n  readonly text: string;\n  /** The body lines after the header, for the window rule. */\n  readonly lines: readonly Line[];\n}\n\ninterface FileDiff {\n  readonly start: number;\n  readonly end: number;\n  readonly path: string;\n  readonly headerText: string;\n  readonly hunks: readonly Hunk[];\n}\n\n/**\n * The synchronous core. Deterministic; every candidate priced through the input's\n * {@link MarkerPricing}. Byte offsets come straight off the UTF-8 line split, so no\n * conversion is needed.\n *\n * @throws {ContentKindError} when the text carries no unified-diff header.\n */\nexport function planDiff(input: PlanInput, options: DiffPlannerOptions = {}): ElisionPlan {\n  const pricing = requirePricing(input);\n  if (probeKind(input.text) !== 'diff') {\n    throw new ContentKindError(\n      `smelt: the diff planner was asked to plan text with no unified-diff header ` +\n        `(a \"diff --git\" line, or \"--- \"/\"+++ \" lines followed by a hunk). It refuses ` +\n        `rather than approximating — output labelled ${DIFF_PLANNER_ID} that was ` +\n        `really line windows would be undetectable from outside. Use \"lexical\", or ` +\n        `\"auto\" to pick by content.`,\n    );\n  }\n  const files = parseFiles(splitLines(input.text));\n  const caseSensitive = options.caseSensitive ?? false;\n  const focus = (input.focus ?? []).filter((term) => term.length > 0);\n  const needles = caseSensitive ? focus : focus.map((term) => term.toLowerCase());\n  const matches = (text: string): boolean => {\n    const haystack = caseSensitive ? text : text.toLowerCase();\n    return needles.some((needle) => haystack.includes(needle));\n  };\n\n  /** The file and hunk collapses — the same at every rung of the window ladder. */\n  const fixed: PlannedElision[] = [];\n  /** The hunks the focus reached, where the window rule runs. */\n  const matched: { readonly file: FileDiff; readonly hunk: Hunk }[] = [];\n  const push = (candidate: PlannedElision): void => {\n    if (savingBytes(candidate, pricing) > 0) fixed.push(candidate);\n  };\n\n  const collapseFiles = (run: readonly FileDiff[]): void => {\n    if (run.length === 0) return;\n    const hunks = run.reduce((n, file) => n + file.hunks.length, 0);\n    push({\n      range: { start: run[0]!.start, end: run[run.length - 1]!.end },\n      reason: {\n        rule: FILE_COLLAPSE_RULE,\n        explanation:\n          `collapsed ${String(run.length)} ${run.length === 1 ? 'file' : 'files'} ` +\n          `(${String(hunks)} ${hunks === 1 ? 'hunk' : 'hunks'})`,\n      },\n      names: run.map((file) => file.path),\n    });\n  };\n\n  const collapseHunks = (file: FileDiff, run: readonly Hunk[]): void => {\n    if (run.length === 0) return;\n    push({\n      range: { start: run[0]!.start, end: run[run.length - 1]!.end },\n      reason: {\n        rule: HUNK_COLLAPSE_RULE,\n        explanation: `collapsed ${String(run.length)} ${run.length === 1 ? 'hunk' : 'hunks'} of ${file.path}`,\n      },\n      names: run.map((hunk) => hunk.header),\n    });\n  };\n\n  let fileRun: FileDiff[] = [];\n  for (const file of files) {\n    const fileKept =\n      needles.length === 0 ||\n      matches(file.headerText) ||\n      file.hunks.some((hunk) => matches(hunk.text));\n    if (!fileKept) {\n      fileRun.push(file);\n      continue;\n    }\n    collapseFiles(fileRun);\n    fileRun = [];\n    let hunkRun: Hunk[] = [];\n    for (const hunk of file.hunks) {\n      if (needles.length > 0 && matches(hunk.text)) {\n        collapseHunks(file, hunkRun);\n        hunkRun = [];\n        matched.push({ file, hunk });\n      } else {\n        hunkRun.push(hunk);\n      }\n    }\n    collapseHunks(file, hunkRun);\n  }\n  collapseFiles(fileRun);\n\n  // The window rule inside every matched hunk, tried with less context at each rung\n  // until the plan fits — the lexical ladder, confined to hunks. The first rung that\n  // fits wins; if none does, the tightest is returned over budget, as it came back.\n  const inputBytes = Buffer.byteLength(input.text, 'utf8');\n  const attempts = WINDOW_LADDER.map((context) => [\n    ...fixed,\n    ...matched.flatMap(({ file, hunk }) => windowsIn(file, hunk, context, matches, pricing)),\n  ]);\n  const elisions =\n    attempts.find((plan) => predictOutputBytes(inputBytes, plan, pricing) <= input.budgetBytes) ??\n    attempts[attempts.length - 1]!;\n\n  return { planner: DIFF_PLANNER_ID, language: input.language, elisions };\n}\n\n/**\n * Inside one matched hunk: keep every line within `context` of a matching line, and\n * collapse each run of the rest that is at least {@link MIN_RUN_LINES} long and pays\n * for its marker. The header line is never part of a run — it is what makes the\n * survivor still read as a hunk.\n */\nfunction windowsIn(\n  file: FileDiff,\n  hunk: Hunk,\n  context: number,\n  matches: (text: string) => boolean,\n  pricing: MarkerPricing,\n): readonly PlannedElision[] {\n  const lines = hunk.lines;\n  const keep: boolean[] = lines.map(() => false);\n  for (let i = 0; i < lines.length; i += 1) {\n    if (!matches(lines[i]!.text)) continue;\n    for (let j = Math.max(0, i - context); j <= Math.min(lines.length - 1, i + context); j += 1) {\n      keep[j] = true;\n    }\n  }\n  const out: PlannedElision[] = [];\n  let runStart = -1;\n  const flush = (endExclusive: number): void => {\n    if (runStart < 0) return;\n    const count = endExclusive - runStart;\n    const range = { start: lines[runStart]!.start, end: lines[endExclusive - 1]!.end };\n    runStart = -1;\n    if (count < MIN_RUN_LINES) return;\n    const candidate: PlannedElision = {\n      range,\n      reason: {\n        rule: HUNK_WINDOW_RULE,\n        explanation: `collapsed ${String(count)} lines of a hunk of ${file.path}`,\n      },\n    };\n    if (savingBytes(candidate, pricing) > 0) out.push(candidate);\n  };\n  for (let i = 0; i < lines.length; i += 1) {\n    if (keep[i] === true) flush(i);\n    else if (runStart < 0) runStart = i;\n  }\n  flush(lines.length);\n  return out;\n}\n\nfunction requirePricing(input: PlanInput): MarkerPricing {\n  const pricing: MarkerPricing | undefined = input.pricing;\n  if (pricing === undefined) throw new MissingMarkerPricingError(DIFF_PLANNER_ID);\n  return pricing;\n}\n\nfunction splitLines(text: string): readonly Line[] {\n  const lines: Line[] = [];\n  let start = 0;\n  for (const content of text.split('\\n')) {\n    const end = start + Buffer.byteLength(content, 'utf8');\n    lines.push({ start, end, text: content });\n    start = end + 1;\n  }\n  return lines;\n}\n\n/** True where a file starts: a `diff --git` line, or a `---` line followed by `+++`. */\nfunction startsFile(lines: readonly Line[], i: number): boolean {\n  const line = lines[i]!.text;\n  if (line.startsWith('diff --git ')) return true;\n  return (\n    line.startsWith('--- ') &&\n    lines[i + 1]?.text.startsWith('+++ ') === true &&\n    !(i > 0 && lines[i - 1]!.text.startsWith('diff --git '))\n  );\n}\n\nfunction parseFiles(lines: readonly Line[]): readonly FileDiff[] {\n  const starts: number[] = [];\n  for (let i = 0; i < lines.length; i += 1) {\n    if (startsFile(lines, i)) {\n      // A `---` line directly under a `diff --git` line belongs to that file's header.\n      if (lines[i]!.text.startsWith('--- ') && starts.at(-1) !== undefined) {\n        const previous = starts.at(-1)!;\n        const between = lines.slice(previous, i).map((line) => line.text);\n        if (\n          between[0]?.startsWith('diff --git ') === true &&\n          !between.some((l) => l.startsWith('@@ '))\n        )\n          continue;\n      }\n      starts.push(i);\n    }\n  }\n  return starts.map((from, index) => {\n    const to = index + 1 < starts.length ? starts[index + 1]! : lastContentLine(lines) + 1;\n    const fileLines = lines.slice(from, to);\n    const firstHunk = fileLines.findIndex((line) => line.text.startsWith('@@ '));\n    const headerLines = firstHunk === -1 ? fileLines : fileLines.slice(0, firstHunk);\n    const hunks: Hunk[] = [];\n    if (firstHunk !== -1) {\n      let hunkStart = firstHunk;\n      for (let i = firstHunk + 1; i <= fileLines.length; i += 1) {\n        if (i === fileLines.length || fileLines[i]!.text.startsWith('@@ ')) {\n          const body = fileLines.slice(hunkStart, i);\n          hunks.push({\n            start: body[0]!.start,\n            end: body[body.length - 1]!.end,\n            header: hunkHeader(body[0]!.text),\n            text: body.map((line) => line.text).join('\\n'),\n            lines: body.slice(1),\n          });\n          hunkStart = i;\n        }\n      }\n    }\n    return {\n      start: fileLines[0]!.start,\n      end: fileLines[fileLines.length - 1]!.end,\n      path: pathOf(headerLines.map((line) => line.text)),\n      headerText: headerLines.map((line) => line.text).join('\\n'),\n      hunks,\n    };\n  });\n}\n\n/** The index of the last line with content, so a trailing newline is never claimed. */\nfunction lastContentLine(lines: readonly Line[]): number {\n  let i = lines.length - 1;\n  while (i > 0 && lines[i]!.text === '') i -= 1;\n  return i;\n}\n\nfunction hunkHeader(line: string): string {\n  const close = line.indexOf('@@', 2);\n  return close === -1 ? line : line.slice(0, close + 2);\n}\n\n/** The path a file header names: the `+++` side first, else the `diff --git` b-side. */\nfunction pathOf(header: readonly string[]): string {\n  const plus = header.find((line) => line.startsWith('+++ '));\n  if (plus !== undefined && plus !== '+++ /dev/null') return stripPrefix(plus.slice(4));\n  const minus = header.find((line) => line.startsWith('--- '));\n  if (minus !== undefined && minus !== '--- /dev/null') return stripPrefix(minus.slice(4));\n  const git = header.find((line) => line.startsWith('diff --git '));\n  if (git !== undefined) {\n    const parts = git.slice('diff --git '.length).split(' b/');\n    return parts[parts.length - 1] ?? git;\n  }\n  return '<unknown>';\n}\n\nfunction stripPrefix(path: string): string {\n  const trimmed = path.split('\\t')[0]!;\n  return trimmed.startsWith('a/') || trimmed.startsWith('b/') ? trimmed.slice(2) : trimmed;\n}\n"]}

package/dist/plan/json.d.ts

@@ -0,0 +1,37 @@
+import type { ElisionPlan, PlanInput, Planner } from '../types.ts';
+export declare const JSON_PLANNER_ID = "json/v1";
+/** The one rule this planner has. */
+export declare const MEMBER_COLLAPSE_RULE = "member-collapse";
+export interface JsonPlannerOptions {
+    /** Focus matching is substring, case-insensitive by default. */
+    readonly caseSensitive?: boolean;
+}
+/**
+ * The JSON planner: **members are the units, not lines.**
+ *
+ * A pretty-printed tool result — a test report, an API response, a bench log — puts
+ * its biggest values on single lines, and a line planner cannot cut a line. This one
+ * scans the document once with positions, so each object member (`"key": value`) and
+ * array element is a unit with a byte range, and collapses runs of sibling units the
+ * focus does not touch — exactly the structural planner's move, with members for
+ * declarations. A matched container is descended into, so the cut lands as deep as
+ * the focus reaches; with no focus the root's members are kept as a skeleton and each
+ * container beneath collapses to one marker whose outline names its keys.
+ *
+ * It refuses text that is not JSON ({@link ContentKindError}) rather than
+ * approximating — the structural planner's rule, for the same reason.
+ */
+export declare class JsonPlanner implements Planner {
+    #private;
+    readonly id = "json/v1";
+    constructor(options?: JsonPlannerOptions);
+    plan(input: PlanInput): Promise<ElisionPlan>;
+}
+/**
+ * The synchronous core, exported like {@link planLexical}. Deterministic; every
+ * candidate is priced through the input's {@link MarkerPricing}, never estimated.
+ *
+ * @throws {ContentKindError} when the text does not parse as a JSON object or array.
+ */
+export declare function planJson(input: PlanInput, options?: JsonPlannerOptions): ElisionPlan;
+//# sourceMappingURL=json.d.ts.map

package/dist/plan/json.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"json.d.ts","sourceRoot":"","sources":["../../src/plan/json.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,WAAW,EAAiB,SAAS,EAAkB,OAAO,EAAE,MAAM,aAAa,CAAC;AAMlG,eAAO,MAAM,eAAe,YAAY,CAAC;AAEzC,qCAAqC;AACrC,eAAO,MAAM,oBAAoB,oBAAoB,CAAC;AAEtD,MAAM,WAAW,kBAAkB;IACjC,gEAAgE;IAChE,QAAQ,CAAC,aAAa,CAAC,EAAE,OAAO,CAAC;CAClC;AAED;;;;;;;;;;;;;;GAcG;AACH,qBAAa,WAAY,YAAW,OAAO;;IACzC,QAAQ,CAAC,EAAE,aAAmB;IAG9B,YAAY,OAAO,GAAE,kBAAuB,EAE3C;IAED,IAAI,CAAC,KAAK,EAAE,SAAS,GAAG,OAAO,CAAC,WAAW,CAAC,CAE3C;CACF;AAYD;;;;;GAKG;AACH,wBAAgB,QAAQ,CAAC,KAAK,EAAE,SAAS,EAAE,OAAO,GAAE,kBAAuB,GAAG,WAAW,CAyDxF","sourcesContent":["import { ContentKindError, MissingMarkerPricingError } from '../errors.ts';\nimport type { ElisionPlan, MarkerPricing, PlanInput, PlannedElision, Planner } from '../types.ts';\n\nimport { savingBytes } from './budget.ts';\nimport { probeKind } from './kind.ts';\nimport { utf8OffsetIndex } from './offsets.ts';\n\nexport const JSON_PLANNER_ID = 'json/v1';\n\n/** The one rule this planner has. */\nexport const MEMBER_COLLAPSE_RULE = 'member-collapse';\n\nexport interface JsonPlannerOptions {\n  /** Focus matching is substring, case-insensitive by default. */\n  readonly caseSensitive?: boolean;\n}\n\n/**\n * The JSON planner: **members are the units, not lines.**\n *\n * A pretty-printed tool result — a test report, an API response, a bench log — puts\n * its biggest values on single lines, and a line planner cannot cut a line. This one\n * scans the document once with positions, so each object member (`\"key\": value`) and\n * array element is a unit with a byte range, and collapses runs of sibling units the\n * focus does not touch — exactly the structural planner's move, with members for\n * declarations. A matched container is descended into, so the cut lands as deep as\n * the focus reaches; with no focus the root's members are kept as a skeleton and each\n * container beneath collapses to one marker whose outline names its keys.\n *\n * It refuses text that is not JSON ({@link ContentKindError}) rather than\n * approximating — the structural planner's rule, for the same reason.\n */\nexport class JsonPlanner implements Planner {\n  readonly id = JSON_PLANNER_ID;\n  readonly #options: JsonPlannerOptions;\n\n  constructor(options: JsonPlannerOptions = {}) {\n    this.#options = options;\n  }\n\n  plan(input: PlanInput): Promise<ElisionPlan> {\n    return Promise.resolve(planJson(input, this.#options));\n  }\n}\n\n/** One member or element: where it sits, what to call it, and what it contains. */\ninterface Unit {\n  readonly start: number;\n  readonly end: number;\n  /** The object key, or `[index]` for an array element. */\n  readonly label: string;\n  /** The members inside, when the value is an object or array. */\n  readonly children?: readonly Unit[];\n}\n\n/**\n * The synchronous core, exported like {@link planLexical}. Deterministic; every\n * candidate is priced through the input's {@link MarkerPricing}, never estimated.\n *\n * @throws {ContentKindError} when the text does not parse as a JSON object or array.\n */\nexport function planJson(input: PlanInput, options: JsonPlannerOptions = {}): ElisionPlan {\n  const pricing = requirePricing(input);\n  if (probeKind(input.text) !== 'json') {\n    throw new ContentKindError(\n      `smelt: the json planner was asked to plan text that does not parse as a JSON ` +\n        `object or array. It refuses rather than approximating — output labelled ` +\n        `${JSON_PLANNER_ID} that was really line windows would be undetectable from ` +\n        `outside. Use \"lexical\", or \"auto\" to pick by content.`,\n    );\n  }\n  const root = scanValue(input.text, skipWs(input.text, 0));\n  const units = root.children ?? [];\n  const boundaries = [...flatten(units)].flatMap((unit) => [unit.start, unit.end]);\n  const toByte = utf8OffsetIndex(input.text, boundaries);\n  const caseSensitive = options.caseSensitive ?? false;\n  const focus = (input.focus ?? []).filter((term) => term.length > 0);\n  const needles = caseSensitive ? focus : focus.map((term) => term.toLowerCase());\n\n  const matches = (unit: Unit): boolean => {\n    const raw = input.text.slice(unit.start, unit.end);\n    const haystack = caseSensitive ? raw : raw.toLowerCase();\n    return needles.some((needle) => haystack.includes(needle));\n  };\n\n  const elisions: PlannedElision[] = [];\n  const collapse = (run: readonly Unit[]): void => {\n    if (run.length === 0) return;\n    const candidate: PlannedElision = {\n      range: { start: toByte.get(run[0]!.start)!, end: toByte.get(run[run.length - 1]!.end)! },\n      reason: {\n        rule: MEMBER_COLLAPSE_RULE,\n        explanation: `collapsed ${String(run.length)} sibling ${run.length === 1 ? 'member' : 'members'}`,\n      },\n      names: run.map((unit) => unit.label),\n    };\n    if (savingBytes(candidate, pricing) > 0) elisions.push(candidate);\n  };\n\n  const walk = (siblings: readonly Unit[], depth: number): void => {\n    let run: Unit[] = [];\n    for (const unit of siblings) {\n      // With a focus, a unit survives when its bytes carry a term; with none, the root\n      // is kept as a skeleton and everything beneath it is fair game.\n      const kept = needles.length > 0 ? matches(unit) : depth === 0;\n      if (!kept) {\n        run.push(unit);\n        continue;\n      }\n      collapse(run);\n      run = [];\n      if (unit.children !== undefined) walk(unit.children, depth + 1);\n    }\n    collapse(run);\n  };\n  walk(units, 0);\n\n  return { planner: JSON_PLANNER_ID, language: input.language, elisions };\n}\n\nfunction* flatten(units: readonly Unit[]): Generator<Unit> {\n  for (const unit of units) {\n    yield unit;\n    if (unit.children !== undefined) yield* flatten(unit.children);\n  }\n}\n\nfunction requirePricing(input: PlanInput): MarkerPricing {\n  const pricing: MarkerPricing | undefined = input.pricing;\n  if (pricing === undefined) throw new MissingMarkerPricingError(JSON_PLANNER_ID);\n  return pricing;\n}\n\n/* ------------------------------------------------------------------------------------\n * A positional scan of already-validated JSON: `probeKind` proved `JSON.parse` accepts\n * the text, so this only has to find where each value starts and ends.\n * ---------------------------------------------------------------------------------- */\n\ninterface Scanned {\n  readonly end: number;\n  readonly children?: readonly Unit[];\n}\n\nfunction skipWs(text: string, i: number): number {\n  let j = i;\n  while (j < text.length && ' \\t\\n\\r'.includes(text[j]!)) j += 1;\n  return j;\n}\n\nfunction scanValue(text: string, i: number): Scanned {\n  const ch = text[i];\n  if (ch === '{') return scanObject(text, i);\n  if (ch === '[') return scanArray(text, i);\n  if (ch === '\"') return { end: scanString(text, i) };\n  let j = i;\n  while (j < text.length && !',]} \\t\\n\\r'.includes(text[j]!)) j += 1;\n  return { end: j };\n}\n\nfunction scanString(text: string, i: number): number {\n  let j = i + 1;\n  while (j < text.length) {\n    const ch = text[j]!;\n    if (ch === '\\\\') {\n      j += 2;\n      continue;\n    }\n    if (ch === '\"') return j + 1;\n    j += 1;\n  }\n  return j;\n}\n\nfunction scanObject(text: string, open: number): Scanned {\n  const children: Unit[] = [];\n  let i = skipWs(text, open + 1);\n  while (i < text.length && text[i] !== '}') {\n    const keyStart = i;\n    const keyEnd = scanString(text, i);\n    const label = JSON.parse(text.slice(keyStart, keyEnd)) as string;\n    i = skipWs(text, keyEnd);\n    i = skipWs(text, i + 1); // ':'\n    const value = scanValue(text, i);\n    children.push({\n      start: keyStart,\n      end: value.end,\n      label,\n      ...(value.children === undefined ? {} : { children: value.children }),\n    });\n    i = skipWs(text, value.end);\n    if (text[i] === ',') i = skipWs(text, i + 1);\n  }\n  return { end: i + 1, children };\n}\n\nfunction scanArray(text: string, open: number): Scanned {\n  const children: Unit[] = [];\n  let i = skipWs(text, open + 1);\n  let index = 0;\n  while (i < text.length && text[i] !== ']') {\n    const value = scanValue(text, i);\n    children.push({\n      start: i,\n      end: value.end,\n      label: `[${String(index)}]`,\n      ...(value.children === undefined ? {} : { children: value.children }),\n    });\n    index += 1;\n    i = skipWs(text, value.end);\n    if (text[i] === ',') i = skipWs(text, i + 1);\n  }\n  return { end: i + 1, children };\n}\n"]}

package/dist/plan/json.js

@@ -0,0 +1,181 @@
+import { ContentKindError, MissingMarkerPricingError } from '../errors.js';
+import { savingBytes } from './budget.js';
+import { probeKind } from './kind.js';
+import { utf8OffsetIndex } from './offsets.js';
+export const JSON_PLANNER_ID = 'json/v1';
+/** The one rule this planner has. */
+export const MEMBER_COLLAPSE_RULE = 'member-collapse';
+/**
+ * The JSON planner: **members are the units, not lines.**
+ *
+ * A pretty-printed tool result — a test report, an API response, a bench log — puts
+ * its biggest values on single lines, and a line planner cannot cut a line. This one
+ * scans the document once with positions, so each object member (`"key": value`) and
+ * array element is a unit with a byte range, and collapses runs of sibling units the
+ * focus does not touch — exactly the structural planner's move, with members for
+ * declarations. A matched container is descended into, so the cut lands as deep as
+ * the focus reaches; with no focus the root's members are kept as a skeleton and each
+ * container beneath collapses to one marker whose outline names its keys.
+ *
+ * It refuses text that is not JSON ({@link ContentKindError}) rather than
+ * approximating — the structural planner's rule, for the same reason.
+ */
+export class JsonPlanner {
+    id = JSON_PLANNER_ID;
+    #options;
+    constructor(options = {}) {
+        this.#options = options;
+    }
+    plan(input) {
+        return Promise.resolve(planJson(input, this.#options));
+    }
+}
+/**
+ * The synchronous core, exported like {@link planLexical}. Deterministic; every
+ * candidate is priced through the input's {@link MarkerPricing}, never estimated.
+ *
+ * @throws {ContentKindError} when the text does not parse as a JSON object or array.
+ */
+export function planJson(input, options = {}) {
+    const pricing = requirePricing(input);
+    if (probeKind(input.text) !== 'json') {
+        throw new ContentKindError(`smelt: the json planner was asked to plan text that does not parse as a JSON ` +
+            `object or array. It refuses rather than approximating — output labelled ` +
+            `${JSON_PLANNER_ID} that was really line windows would be undetectable from ` +
+            `outside. Use "lexical", or "auto" to pick by content.`);
+    }
+    const root = scanValue(input.text, skipWs(input.text, 0));
+    const units = root.children ?? [];
+    const boundaries = [...flatten(units)].flatMap((unit) => [unit.start, unit.end]);
+    const toByte = utf8OffsetIndex(input.text, boundaries);
+    const caseSensitive = options.caseSensitive ?? false;
+    const focus = (input.focus ?? []).filter((term) => term.length > 0);
+    const needles = caseSensitive ? focus : focus.map((term) => term.toLowerCase());
+    const matches = (unit) => {
+        const raw = input.text.slice(unit.start, unit.end);
+        const haystack = caseSensitive ? raw : raw.toLowerCase();
+        return needles.some((needle) => haystack.includes(needle));
+    };
+    const elisions = [];
+    const collapse = (run) => {
+        if (run.length === 0)
+            return;
+        const candidate = {
+            range: { start: toByte.get(run[0].start), end: toByte.get(run[run.length - 1].end) },
+            reason: {
+                rule: MEMBER_COLLAPSE_RULE,
+                explanation: `collapsed ${String(run.length)} sibling ${run.length === 1 ? 'member' : 'members'}`,
+            },
+            names: run.map((unit) => unit.label),
+        };
+        if (savingBytes(candidate, pricing) > 0)
+            elisions.push(candidate);
+    };
+    const walk = (siblings, depth) => {
+        let run = [];
+        for (const unit of siblings) {
+            // With a focus, a unit survives when its bytes carry a term; with none, the root
+            // is kept as a skeleton and everything beneath it is fair game.
+            const kept = needles.length > 0 ? matches(unit) : depth === 0;
+            if (!kept) {
+                run.push(unit);
+                continue;
+            }
+            collapse(run);
+            run = [];
+            if (unit.children !== undefined)
+                walk(unit.children, depth + 1);
+        }
+        collapse(run);
+    };
+    walk(units, 0);
+    return { planner: JSON_PLANNER_ID, language: input.language, elisions };
+}
+function* flatten(units) {
+    for (const unit of units) {
+        yield unit;
+        if (unit.children !== undefined)
+            yield* flatten(unit.children);
+    }
+}
+function requirePricing(input) {
+    const pricing = input.pricing;
+    if (pricing === undefined)
+        throw new MissingMarkerPricingError(JSON_PLANNER_ID);
+    return pricing;
+}
+function skipWs(text, i) {
+    let j = i;
+    while (j < text.length && ' \t\n\r'.includes(text[j]))
+        j += 1;
+    return j;
+}
+function scanValue(text, i) {
+    const ch = text[i];
+    if (ch === '{')
+        return scanObject(text, i);
+    if (ch === '[')
+        return scanArray(text, i);
+    if (ch === '"')
+        return { end: scanString(text, i) };
+    let j = i;
+    while (j < text.length && !',]} \t\n\r'.includes(text[j]))
+        j += 1;
+    return { end: j };
+}
+function scanString(text, i) {
+    let j = i + 1;
+    while (j < text.length) {
+        const ch = text[j];
+        if (ch === '\\') {
+            j += 2;
+            continue;
+        }
+        if (ch === '"')
+            return j + 1;
+        j += 1;
+    }
+    return j;
+}
+function scanObject(text, open) {
+    const children = [];
+    let i = skipWs(text, open + 1);
+    while (i < text.length && text[i] !== '}') {
+        const keyStart = i;
+        const keyEnd = scanString(text, i);
+        const label = JSON.parse(text.slice(keyStart, keyEnd));
+        i = skipWs(text, keyEnd);
+        i = skipWs(text, i + 1); // ':'
+        const value = scanValue(text, i);
+        children.push({
+            start: keyStart,
+            end: value.end,
+            label,
+            ...(value.children === undefined ? {} : { children: value.children }),
+        });
+        i = skipWs(text, value.end);
+        if (text[i] === ',')
+            i = skipWs(text, i + 1);
+    }
+    return { end: i + 1, children };
+}
+function scanArray(text, open) {
+    const children = [];
+    let i = skipWs(text, open + 1);
+    let index = 0;
+    while (i < text.length && text[i] !== ']') {
+        const value = scanValue(text, i);
+        children.push({
+            start: i,
+            end: value.end,
+            label: `[${String(index)}]`,
+            ...(value.children === undefined ? {} : { children: value.children }),
+        });
+        index += 1;
+        i = skipWs(text, value.end);
+        if (text[i] === ',')
+            i = skipWs(text, i + 1);
+    }
+    return { end: i + 1, children };
+}
+//# sourceMappingURL=json.js.map

package/dist/plan/json.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"json.js","sourceRoot":"","sources":["../../src/plan/json.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,gBAAgB,EAAE,yBAAyB,EAAE,MAAM,cAAc,CAAC;AAG3E,OAAO,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AAC1C,OAAO,EAAE,SAAS,EAAE,MAAM,WAAW,CAAC;AACtC,OAAO,EAAE,eAAe,EAAE,MAAM,cAAc,CAAC;AAE/C,MAAM,CAAC,MAAM,eAAe,GAAG,SAAS,CAAC;AAEzC,qCAAqC;AACrC,MAAM,CAAC,MAAM,oBAAoB,GAAG,iBAAiB,CAAC;AAOtD;;;;;;;;;;;;;;GAcG;AACH,MAAM,OAAO,WAAW;IACb,EAAE,GAAG,eAAe,CAAC;IACrB,QAAQ,CAAqB;IAEtC,YAAY,OAAO,GAAuB,EAAE;QAC1C,IAAI,CAAC,QAAQ,GAAG,OAAO,CAAC;IAC1B,CAAC;IAED,IAAI,CAAC,KAAgB;QACnB,OAAO,OAAO,CAAC,OAAO,CAAC,QAAQ,CAAC,KAAK,EAAE,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC;IACzD,CAAC;CACF;AAYD;;;;;GAKG;AACH,MAAM,UAAU,QAAQ,CAAC,KAAgB,EAAE,OAAO,GAAuB,EAAE;IACzE,MAAM,OAAO,GAAG,cAAc,CAAC,KAAK,CAAC,CAAC;IACtC,IAAI,SAAS,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,MAAM,EAAE,CAAC;QACrC,MAAM,IAAI,gBAAgB,CACxB,+EAA+E;YAC7E,0EAA0E;YAC1E,GAAG,eAAe,2DAA2D;YAC7E,uDAAuD,CAC1D,CAAC;IACJ,CAAC;IACD,MAAM,IAAI,GAAG,SAAS,CAAC,KAAK,CAAC,IAAI,EAAE,MAAM,CAAC,KAAK,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;IAC1D,MAAM,KAAK,GAAG,IAAI,CAAC,QAAQ,IAAI,EAAE,CAAC;IAClC,MAAM,UAAU,GAAG,CAAC,GAAG,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,IAAI,CAAC,KAAK,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC;IACjF,MAAM,MAAM,GAAG,eAAe,CAAC,KAAK,CAAC,IAAI,EAAE,UAAU,CAAC,CAAC;IACvD,MAAM,aAAa,GAAG,OAAO,CAAC,aAAa,IAAI,KAAK,CAAC;IACrD,MAAM,KAAK,GAAG,CAAC,KAAK,CAAC,KAAK,IAAI,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;IACpE,MAAM,OAAO,GAAG,aAAa,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,WAAW,EAAE,CAAC,CAAC;IAEhF,MAAM,OAAO,GAAG,CAAC,IAAU,EAAW,EAAE;QACtC,MAAM,GAAG,GAAG,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC;QACnD,MAAM,QAAQ,GAAG,aAAa,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC,WAAW,EAAE,CAAC;QACzD,OAAO,OAAO,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,QAAQ,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC;IAC7D,CAAC,CAAC;IAEF,MAAM,QAAQ,GAAqB,EAAE,CAAC;IACtC,MAAM,QAAQ,GAAG,CAAC,GAAoB,EAAQ,EAAE;QAC9C,IAAI,GAAG,CAAC,MAAM,KAAK,CAAC;YAAE,OAAO;QAC7B,MAAM,SAAS,GAAmB;YAChC,KAAK,EAAE,EAAE,KAAK,EAAE,MAAM,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAE,CAAC,KAAK,CAAE,EAAE,GAAG,EAAE,MAAM,CAAC,GAAG,CAAC,GAAG,CAAC,GAAG,CAAC,MAAM,GAAG,CAAC,CAAE,CAAC,GAAG,CAAE,EAAE;YACxF,MAAM,EAAE;gBACN,IAAI,EAAE,oBAAoB;gBAC1B,WAAW,EAAE,aAAa,MAAM,CAAC,GAAG,CAAC,MAAM,CAAC,YAAY,GAAG,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,SAAS,EAAE;aAClG;YACD,KAAK,EAAE,GAAG,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,KAAK,CAAC;SACrC,CAAC;QACF,IAAI,WAAW,CAAC,SAAS,EAAE,OAAO,CAAC,GAAG,CAAC;YAAE,QAAQ,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;IACpE,CAAC,CAAC;IAEF,MAAM,IAAI,GAAG,CAAC,QAAyB,EAAE,KAAa,EAAQ,EAAE;QAC9D,IAAI,GAAG,GAAW,EAAE,CAAC;QACrB,KAAK,MAAM,IAAI,IAAI,QAAQ,EAAE,CAAC;YAC5B,iFAAiF;YACjF,gEAAgE;YAChE,MAAM,IAAI,GAAG,OAAO,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,KAAK,KAAK,CAAC,CAAC;YAC9D,IAAI,CAAC,IAAI,EAAE,CAAC;gBACV,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;gBACf,SAAS;YACX,CAAC;YACD,QAAQ,CAAC,GAAG,CAAC,CAAC;YACd,GAAG,GAAG,EAAE,CAAC;YACT,IAAI,IAAI,CAAC,QAAQ,KAAK,SAAS;gBAAE,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,KAAK,GAAG,CAAC,CAAC,CAAC;QAClE,CAAC;QACD,QAAQ,CAAC,GAAG,CAAC,CAAC;IAChB,CAAC,CAAC;IACF,IAAI,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC;IAEf,OAAO,EAAE,OAAO,EAAE,eAAe,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ,EAAE,QAAQ,EAAE,CAAC;AAC1E,CAAC;AAED,QAAQ,CAAC,CAAC,OAAO,CAAC,KAAsB;IACtC,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,MAAM,IAAI,CAAC;QACX,IAAI,IAAI,CAAC,QAAQ,KAAK,SAAS;YAAE,KAAK,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;IACjE,CAAC;AACH,CAAC;AAED,SAAS,cAAc,CAAC,KAAgB;IACtC,MAAM,OAAO,GAA8B,KAAK,CAAC,OAAO,CAAC;IACzD,IAAI,OAAO,KAAK,SAAS;QAAE,MAAM,IAAI,yBAAyB,CAAC,eAAe,CAAC,CAAC;IAChF,OAAO,OAAO,CAAC;AACjB,CAAC;AAYD,SAAS,MAAM,CAAC,IAAY,EAAE,CAAS;IACrC,IAAI,CAAC,GAAG,CAAC,CAAC;IACV,OAAO,CAAC,GAAG,IAAI,CAAC,MAAM,IAAI,SAAS,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAE,CAAC;QAAE,CAAC,IAAI,CAAC,CAAC;IAC/D,OAAO,CAAC,CAAC;AACX,CAAC;AAED,SAAS,SAAS,CAAC,IAAY,EAAE,CAAS;IACxC,MAAM,EAAE,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC;IACnB,IAAI,EAAE,KAAK,GAAG;QAAE,OAAO,UAAU,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC;IAC3C,IAAI,EAAE,KAAK,GAAG;QAAE,OAAO,SAAS,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC;IAC1C,IAAI,EAAE,KAAK,GAAG;QAAE,OAAO,EAAE,GAAG,EAAE,UAAU,CAAC,IAAI,EAAE,CAAC,CAAC,EAAE,CAAC;IACpD,IAAI,CAAC,GAAG,CAAC,CAAC;IACV,OAAO,CAAC,GAAG,IAAI,CAAC,MAAM,IAAI,CAAC,YAAY,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAE,CAAC;QAAE,CAAC,IAAI,CAAC,CAAC;IACnE,OAAO,EAAE,GAAG,EAAE,CAAC,EAAE,CAAC;AACpB,CAAC;AAED,SAAS,UAAU,CAAC,IAAY,EAAE,CAAS;IACzC,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;IACd,OAAO,CAAC,GAAG,IAAI,CAAC,MAAM,EAAE,CAAC;QACvB,MAAM,EAAE,GAAG,IAAI,CAAC,CAAC,CAAE,CAAC;QACpB,IAAI,EAAE,KAAK,IAAI,EAAE,CAAC;YAChB,CAAC,IAAI,CAAC,CAAC;YACP,SAAS;QACX,CAAC;QACD,IAAI,EAAE,KAAK,GAAG;YAAE,OAAO,CAAC,GAAG,CAAC,CAAC;QAC7B,CAAC,IAAI,CAAC,CAAC;IACT,CAAC;IACD,OAAO,CAAC,CAAC;AACX,CAAC;AAED,SAAS,UAAU,CAAC,IAAY,EAAE,IAAY;IAC5C,MAAM,QAAQ,GAAW,EAAE,CAAC;IAC5B,IAAI,CAAC,GAAG,MAAM,CAAC,IAAI,EAAE,IAAI,GAAG,CAAC,CAAC,CAAC;IAC/B,OAAO,CAAC,GAAG,IAAI,CAAC,MAAM,IAAI,IAAI,CAAC,CAAC,CAAC,KAAK,GAAG,EAAE,CAAC;QAC1C,MAAM,QAAQ,GAAG,CAAC,CAAC;QACnB,MAAM,MAAM,GAAG,UAAU,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC;QACnC,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,QAAQ,EAAE,MAAM,CAAC,CAAW,CAAC;QACjE,CAAC,GAAG,MAAM,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzB,CAAC,GAAG,MAAM,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,MAAM;QAC/B,MAAM,KAAK,GAAG,SAAS,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC;QACjC,QAAQ,CAAC,IAAI,CAAC;YACZ,KAAK,EAAE,QAAQ;YACf,GAAG,EAAE,KAAK,CAAC,GAAG;YACd,KAAK;YACL,GAAG,CAAC,KAAK,CAAC,QAAQ,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ,EAAE,CAAC;SACtE,CAAC,CAAC;QACH,CAAC,GAAG,MAAM,CAAC,IAAI,EAAE,KAAK,CAAC,GAAG,CAAC,CAAC;QAC5B,IAAI,IAAI,CAAC,CAAC,CAAC,KAAK,GAAG;YAAE,CAAC,GAAG,MAAM,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAC/C,CAAC;IACD,OAAO,EAAE,GAAG,EAAE,CAAC,GAAG,CAAC,EAAE,QAAQ,EAAE,CAAC;AAClC,CAAC;AAED,SAAS,SAAS,CAAC,IAAY,EAAE,IAAY;IAC3C,MAAM,QAAQ,GAAW,EAAE,CAAC;IAC5B,IAAI,CAAC,GAAG,MAAM,CAAC,IAAI,EAAE,IAAI,GAAG,CAAC,CAAC,CAAC;IAC/B,IAAI,KAAK,GAAG,CAAC,CAAC;IACd,OAAO,CAAC,GAAG,IAAI,CAAC,MAAM,IAAI,IAAI,CAAC,CAAC,CAAC,KAAK,GAAG,EAAE,CAAC;QAC1C,MAAM,KAAK,GAAG,SAAS,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC;QACjC,QAAQ,CAAC,IAAI,CAAC;YACZ,KAAK,EAAE,CAAC;YACR,GAAG,EAAE,KAAK,CAAC,GAAG;YACd,KAAK,EAAE,IAAI,MAAM,CAAC,KAAK,CAAC,GAAG;YAC3B,GAAG,CAAC,KAAK,CAAC,QAAQ,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ,EAAE,CAAC;SACtE,CAAC,CAAC;QACH,KAAK,IAAI,CAAC,CAAC;QACX,CAAC,GAAG,MAAM,CAAC,IAAI,EAAE,KAAK,CAAC,GAAG,CAAC,CAAC;QAC5B,IAAI,IAAI,CAAC,CAAC,CAAC,KAAK,GAAG;YAAE,CAAC,GAAG,MAAM,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAC/C,CAAC;IACD,OAAO,EAAE,GAAG,EAAE,CAAC,GAAG,CAAC,EAAE,QAAQ,EAAE,CAAC;AAClC,CAAC","sourcesContent":["import { ContentKindError, MissingMarkerPricingError } from '../errors.ts';\nimport type { ElisionPlan, MarkerPricing, PlanInput, PlannedElision, Planner } from '../types.ts';\n\nimport { savingBytes } from './budget.ts';\nimport { probeKind } from './kind.ts';\nimport { utf8OffsetIndex } from './offsets.ts';\n\nexport const JSON_PLANNER_ID = 'json/v1';\n\n/** The one rule this planner has. */\nexport const MEMBER_COLLAPSE_RULE = 'member-collapse';\n\nexport interface JsonPlannerOptions {\n  /** Focus matching is substring, case-insensitive by default. */\n  readonly caseSensitive?: boolean;\n}\n\n/**\n * The JSON planner: **members are the units, not lines.**\n *\n * A pretty-printed tool result — a test report, an API response, a bench log — puts\n * its biggest values on single lines, and a line planner cannot cut a line. This one\n * scans the document once with positions, so each object member (`\"key\": value`) and\n * array element is a unit with a byte range, and collapses runs of sibling units the\n * focus does not touch — exactly the structural planner's move, with members for\n * declarations. A matched container is descended into, so the cut lands as deep as\n * the focus reaches; with no focus the root's members are kept as a skeleton and each\n * container beneath collapses to one marker whose outline names its keys.\n *\n * It refuses text that is not JSON ({@link ContentKindError}) rather than\n * approximating — the structural planner's rule, for the same reason.\n */\nexport class JsonPlanner implements Planner {\n  readonly id = JSON_PLANNER_ID;\n  readonly #options: JsonPlannerOptions;\n\n  constructor(options: JsonPlannerOptions = {}) {\n    this.#options = options;\n  }\n\n  plan(input: PlanInput): Promise<ElisionPlan> {\n    return Promise.resolve(planJson(input, this.#options));\n  }\n}\n\n/** One member or element: where it sits, what to call it, and what it contains. */\ninterface Unit {\n  readonly start: number;\n  readonly end: number;\n  /** The object key, or `[index]` for an array element. */\n  readonly label: string;\n  /** The members inside, when the value is an object or array. */\n  readonly children?: readonly Unit[];\n}\n\n/**\n * The synchronous core, exported like {@link planLexical}. Deterministic; every\n * candidate is priced through the input's {@link MarkerPricing}, never estimated.\n *\n * @throws {ContentKindError} when the text does not parse as a JSON object or array.\n */\nexport function planJson(input: PlanInput, options: JsonPlannerOptions = {}): ElisionPlan {\n  const pricing = requirePricing(input);\n  if (probeKind(input.text) !== 'json') {\n    throw new ContentKindError(\n      `smelt: the json planner was asked to plan text that does not parse as a JSON ` +\n        `object or array. It refuses rather than approximating — output labelled ` +\n        `${JSON_PLANNER_ID} that was really line windows would be undetectable from ` +\n        `outside. Use \"lexical\", or \"auto\" to pick by content.`,\n    );\n  }\n  const root = scanValue(input.text, skipWs(input.text, 0));\n  const units = root.children ?? [];\n  const boundaries = [...flatten(units)].flatMap((unit) => [unit.start, unit.end]);\n  const toByte = utf8OffsetIndex(input.text, boundaries);\n  const caseSensitive = options.caseSensitive ?? false;\n  const focus = (input.focus ?? []).filter((term) => term.length > 0);\n  const needles = caseSensitive ? focus : focus.map((term) => term.toLowerCase());\n\n  const matches = (unit: Unit): boolean => {\n    const raw = input.text.slice(unit.start, unit.end);\n    const haystack = caseSensitive ? raw : raw.toLowerCase();\n    return needles.some((needle) => haystack.includes(needle));\n  };\n\n  const elisions: PlannedElision[] = [];\n  const collapse = (run: readonly Unit[]): void => {\n    if (run.length === 0) return;\n    const candidate: PlannedElision = {\n      range: { start: toByte.get(run[0]!.start)!, end: toByte.get(run[run.length - 1]!.end)! },\n      reason: {\n        rule: MEMBER_COLLAPSE_RULE,\n        explanation: `collapsed ${String(run.length)} sibling ${run.length === 1 ? 'member' : 'members'}`,\n      },\n      names: run.map((unit) => unit.label),\n    };\n    if (savingBytes(candidate, pricing) > 0) elisions.push(candidate);\n  };\n\n  const walk = (siblings: readonly Unit[], depth: number): void => {\n    let run: Unit[] = [];\n    for (const unit of siblings) {\n      // With a focus, a unit survives when its bytes carry a term; with none, the root\n      // is kept as a skeleton and everything beneath it is fair game.\n      const kept = needles.length > 0 ? matches(unit) : depth === 0;\n      if (!kept) {\n        run.push(unit);\n        continue;\n      }\n      collapse(run);\n      run = [];\n      if (unit.children !== undefined) walk(unit.children, depth + 1);\n    }\n    collapse(run);\n  };\n  walk(units, 0);\n\n  return { planner: JSON_PLANNER_ID, language: input.language, elisions };\n}\n\nfunction* flatten(units: readonly Unit[]): Generator<Unit> {\n  for (const unit of units) {\n    yield unit;\n    if (unit.children !== undefined) yield* flatten(unit.children);\n  }\n}\n\nfunction requirePricing(input: PlanInput): MarkerPricing {\n  const pricing: MarkerPricing | undefined = input.pricing;\n  if (pricing === undefined) throw new MissingMarkerPricingError(JSON_PLANNER_ID);\n  return pricing;\n}\n\n/* ------------------------------------------------------------------------------------\n * A positional scan of already-validated JSON: `probeKind` proved `JSON.parse` accepts\n * the text, so this only has to find where each value starts and ends.\n * ---------------------------------------------------------------------------------- */\n\ninterface Scanned {\n  readonly end: number;\n  readonly children?: readonly Unit[];\n}\n\nfunction skipWs(text: string, i: number): number {\n  let j = i;\n  while (j < text.length && ' \\t\\n\\r'.includes(text[j]!)) j += 1;\n  return j;\n}\n\nfunction scanValue(text: string, i: number): Scanned {\n  const ch = text[i];\n  if (ch === '{') return scanObject(text, i);\n  if (ch === '[') return scanArray(text, i);\n  if (ch === '\"') return { end: scanString(text, i) };\n  let j = i;\n  while (j < text.length && !',]} \\t\\n\\r'.includes(text[j]!)) j += 1;\n  return { end: j };\n}\n\nfunction scanString(text: string, i: number): number {\n  let j = i + 1;\n  while (j < text.length) {\n    const ch = text[j]!;\n    if (ch === '\\\\') {\n      j += 2;\n      continue;\n    }\n    if (ch === '\"') return j + 1;\n    j += 1;\n  }\n  return j;\n}\n\nfunction scanObject(text: string, open: number): Scanned {\n  const children: Unit[] = [];\n  let i = skipWs(text, open + 1);\n  while (i < text.length && text[i] !== '}') {\n    const keyStart = i;\n    const keyEnd = scanString(text, i);\n    const label = JSON.parse(text.slice(keyStart, keyEnd)) as string;\n    i = skipWs(text, keyEnd);\n    i = skipWs(text, i + 1); // ':'\n    const value = scanValue(text, i);\n    children.push({\n      start: keyStart,\n      end: value.end,\n      label,\n      ...(value.children === undefined ? {} : { children: value.children }),\n    });\n    i = skipWs(text, value.end);\n    if (text[i] === ',') i = skipWs(text, i + 1);\n  }\n  return { end: i + 1, children };\n}\n\nfunction scanArray(text: string, open: number): Scanned {\n  const children: Unit[] = [];\n  let i = skipWs(text, open + 1);\n  let index = 0;\n  while (i < text.length && text[i] !== ']') {\n    const value = scanValue(text, i);\n    children.push({\n      start: i,\n      end: value.end,\n      label: `[${String(index)}]`,\n      ...(value.children === undefined ? {} : { children: value.children }),\n    });\n    index += 1;\n    i = skipWs(text, value.end);\n    if (text[i] === ',') i = skipWs(text, i + 1);\n  }\n  return { end: i + 1, children };\n}\n"]}