@sdsrs/code-graph 0.178.0 → 0.179.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
@@ -4,7 +4,7 @@
4
4
  "author": {
5
5
  "name": "sdsrs"
6
6
  },
7
- "version": "0.178.0",
7
+ "version": "0.179.0",
8
8
  "keywords": [
9
9
  "code-graph",
10
10
  "ast",
@@ -28,6 +28,7 @@ const { formatCoveringTests } = require('./covering-tests');
28
28
  const { emitPreToolContext } = require('./hook-emit');
29
29
  const { hidden } = require('./proc-opts');
30
30
  const { remainingMs } = require('./hook-fail-open');
31
+ const { definitionHeaders } = require('./stop-impact');
31
32
 
32
33
  // v0.49 — walk up from the shell cwd (subdir-cwd fix). The per-cwd index.db
33
34
  // gate kept this hook dark for entire sessions after `cd backend/` — daagu
@@ -135,19 +136,21 @@ const scanned = oldStr.length > 8192 ? oldStr.slice(0, 8192) : oldStr;
135
136
  // matches — 0.163.0's rule. Three replacements (the earliest match, then an
136
137
  // allow-list of what may precede a method name, then keyword-anchored arms
137
138
  // first) each fixed shapes and broke others, and three review rounds found a
138
- // regression in each (pre-tag review 2026-09-29), so the released rule stays
139
- // until a per-language design with a corpus for each language replaces it.
139
+ // regression in each (pre-tag review 2026-09-29), so this rule still names the
140
+ // pick; the Stop check's per-language reader decides below, where it reads the
141
+ // language, which definition the edit changed.
140
142
  //
141
143
  // Where the definition starts is its name's line: `// TODO: make this
142
144
  // async\nfn f(` matches from the comment line, and the same-file pick below
143
145
  // counts lines from here. That moves no name, only the line.
144
146
  let symbol = null;
145
147
  let symbolAt = -1;
148
+ let nameAt = -1;
146
149
  for (const pat of fnPatterns) {
147
150
  const m = scanned.match(new RegExp(pat.source, `${pat.flags}d`));
148
151
  if (m) {
149
152
  symbol = m[1] || m[2];
150
- const nameAt = (m.indices[1] || m.indices[2])[0];
153
+ nameAt = (m.indices[1] || m.indices[2])[0];
151
154
  symbolAt = Math.max(m.index, scanned.lastIndexOf('\n', nameAt - 1) + 1);
152
155
  break;
153
156
  }
@@ -179,7 +182,107 @@ function headerAt(text, at) {
179
182
  .filter((i) => i >= 0);
180
183
  return ends.length ? rest.slice(0, Math.min(...ends)) : rest.split('\n', 1)[0];
181
184
  }
182
- if (newStr.includes(headerAt(scanned, symbolAt))) process.exit(0);
185
+ const bodyOnly = newStr.includes(headerAt(scanned, symbolAt));
186
+
187
+ // Which definition the edit changed, where the Stop check's per-language reader
188
+ // (stop-impact.js LANGS) reads this file: a definition of the hunk whose
189
+ // signature key the new text no longer has. The pick above sent 35 of its 103
190
+ // impact queries for names that are no definition there (a Rust `match f(…)
191
+ // {`, a comment's "function already", a string) and missed 24 changed headers
192
+ // (a later definition, a `{` inside the parameters, an old header still inside
193
+ // the new one), measured on every Edit hunk in this machine's transcripts
194
+ // (tasks/specs/pre-edit-pick-by-reader.md).
195
+ // - a changed definition of the picked name → it, placed where it starts;
196
+ // - the pick's header changed as the first-match rule reads it, and the pick
197
+ // is not proven no definition → the pick, as before. The reader may not
198
+ // see it (a method whose parameters span lines, `#[inline] fn` on one
199
+ // line); where it reads it unchanged, the old reading can have run into
200
+ // the next definition, which 0.178.0 did too;
201
+ // - else the first changed definition, logged for the Stop check too;
202
+ // - else none.
203
+ // The pick is no definition when its line opens a comment that has not closed
204
+ // before it, or, in Rust and Python, when its name does not follow `fn` /
205
+ // `def`. A comment here is not stop-impact.js's COMMENT_LINE, which takes any
206
+ // line opening with `*` or `#`: `*entries(prefix) {` is a JS generator method,
207
+ // `#[inline] pub fn` a Rust definition; a doc comment's `*` is followed by a
208
+ // space, a `/` or the line's end.
209
+ //
210
+ // Only in the extensions of the reader's languages that corpus had hunks in
211
+ // (not C: 2 hunks). Elsewhere the reader misread shapes no hunk there had
212
+ // (pre-tag review of 0.179.0: Kotlin's `fun Config.validate(` reads as
213
+ // `Config`, PHP's `function &name(` has no keyword before the name, a C++
214
+ // statement `Type name(args);` reads as a definition), so the pick keeps its
215
+ // own verdict, as in 0.178.0; so does a hunk the reader has no reading of (a
216
+ // header past its cap).
217
+ const READER_MEASURED = new Set(['.rs', '.py', '.js', '.mjs', '.cjs', '.ts']);
218
+ const OPENS_A_COMMENT = /^\s*(?:\/\/|\/\*|\*(?=\s|\/|$)|#(?!\[)|--|;)/;
219
+ function onComment(line, nameCol) {
220
+ const closed = /^\s*(?:\/\*.*?\*\/\s*)+/.exec(line.slice(0, nameCol));
221
+ return OPENS_A_COMMENT.test(closed ? line.slice(closed[0].length) : line);
222
+ }
223
+ function changedDefinition() {
224
+ const ext = path.extname(editedAbs).toLowerCase();
225
+ const before = READER_MEASURED.has(ext) ? definitionHeaders(oldStr, ext) : null;
226
+ const after = before && definitionHeaders(newStr, ext);
227
+ if (!before || !after) return bodyOnly ? null : { name: symbol, at: symbolAt };
228
+ const count = (defs) => {
229
+ const m = new Map();
230
+ for (const d of defs) m.set(`${d.name}\0${d.key}`, (m.get(`${d.name}\0${d.key}`) || 0) + 1);
231
+ return m;
232
+ };
233
+ const has = count(after.defs);
234
+ const left = new Map(has);
235
+ let changed = [];
236
+ for (const d of before.defs) {
237
+ const k = `${d.name}\0${d.key}`;
238
+ if (left.get(k) > 0) left.set(k, left.get(k) - 1);
239
+ else changed.push(d);
240
+ }
241
+ // The count matches headers, not definitions: of two definitions of one name
242
+ // with one old header, one kept, it marks the later, whichever the edit
243
+ // changed. Such a name's definitions pair in order when the hunk keeps their
244
+ // number and pairing finds as many changed as the count; otherwise which one
245
+ // changed is not in the text (one went or came, or the hunk moved them), and
246
+ // the reader names none of that name (the rule below can still keep the
247
+ // pick).
248
+ const twinned = new Set(changed
249
+ .filter((c) => (has.get(`${c.name}\0${c.key}`) || 0) > 0)
250
+ .map((c) => c.name));
251
+ if (twinned.size) {
252
+ const counted = new Map();
253
+ for (const d of changed) if (twinned.has(d.name)) counted.set(d.name, (counted.get(d.name) || 0) + 1);
254
+ const byName = (defs) => {
255
+ const m = new Map();
256
+ for (const d of defs) if (twinned.has(d.name)) (m.get(d.name) || m.set(d.name, []).get(d.name)).push(d);
257
+ return m;
258
+ };
259
+ const olds = byName(before.defs);
260
+ const news = byName(after.defs);
261
+ changed = changed.filter((d) => !twinned.has(d.name));
262
+ for (const [name, o] of olds) {
263
+ const n = news.get(name) || [];
264
+ const paired = o.length === n.length ? o.filter((d, i) => d.key !== n[i].key) : [];
265
+ if (paired.length === counted.get(name)) changed.push(...paired);
266
+ }
267
+ }
268
+ changed.sort((x, y) => x.at - y.at);
269
+ const same = changed.find((d) => d.name === symbol);
270
+ if (same) return { name: symbol, at: same.at };
271
+ const lineStart = scanned.lastIndexOf('\n', nameAt - 1) + 1;
272
+ const lineEnd = scanned.indexOf('\n', nameAt);
273
+ const pickLine = scanned.slice(lineStart, lineEnd === -1 ? scanned.length : lineEnd);
274
+ const notADefinition = onComment(pickLine, nameAt - lineStart)
275
+ || (before.keywordOnly && !before.keywordLeads(scanned.slice(lineStart, nameAt)));
276
+ if (!bodyOnly && !notADefinition) return { name: symbol, at: symbolAt };
277
+ return changed.length ? { name: changed[0].name, at: changed[0].at } : null;
278
+ }
279
+ const edited = changedDefinition();
280
+ if (!edited) process.exit(0);
281
+ if (edited.name !== symbol) {
282
+ symbol = edited.name;
283
+ logEdit(symbol);
284
+ }
285
+ symbolAt = edited.at;
183
286
 
184
287
  if (process.env.CODE_GRAPH_QUIET_HOOKS === '1') process.exit(0);
185
288
 
@@ -89,12 +89,17 @@ const NOT_A_TYPE = new Set([
89
89
  // initializers, C# `: base(…)` — body, not signature)
90
90
  // closeParen the parameter list's `)` ends it (Ruby, Lua: no return types,
91
91
  // and a one-line body may follow on the same line)
92
+ // keywordOnly every function definition's name follows `keyword`, so a
93
+ // name without it is none (pre-edit-guide reads that; Rust and
94
+ // Python only, the languages it was measured on. These rules do not
95
+ // find every such definition: a line opening with `#[` or `/* … */`,
96
+ // a second one on a line)
92
97
  const LANGS = new Map();
93
98
  function defineLang(exts, profile) {
94
99
  for (const e of exts) LANGS.set(e, profile);
95
100
  }
96
- defineLang(['.rs'], { keyword: 'fn', ends: '{;' });
97
- defineLang(['.py', '.pyi'], { keyword: 'def', ends: ':' });
101
+ defineLang(['.rs'], { keyword: 'fn', ends: '{;', keywordOnly: true });
102
+ defineLang(['.py', '.pyi'], { keyword: 'def', ends: ':', keywordOnly: true });
98
103
  defineLang(['.go'], { keyword: 'func', receiver: true, ends: '{;\n' });
99
104
  defineLang(['.js', '.jsx', '.mjs', '.cjs', '.ts', '.tsx', '.mts', '.cts'],
100
105
  { keyword: 'function\\*?', jsForms: true, ends: '{;', arrowEnds: true, angles: true });
@@ -110,12 +115,18 @@ function escapeRe(s) {
110
115
  return String(s).replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
111
116
  }
112
117
 
118
+ // What comes before a definition's name: the keyword, Go's receiver, Kotlin's
119
+ // type parameters.
120
+ function keywordLead(lang) {
121
+ const receiver = lang.receiver ? '(?:\\([^)\\n]*\\)\\s*)?' : '';
122
+ const typeParams = lang.typeParams ? '(?:<[^>\\n]*>\\s*)?' : '';
123
+ return `\\b(?:${lang.keyword})\\s+${receiver}${typeParams}`;
124
+ }
125
+
113
126
  function definitionPatterns(S, lang) {
114
127
  const out = [];
115
128
  if (lang.keyword) {
116
- const receiver = lang.receiver ? '(?:\\([^)\\n]*\\)\\s*)?' : '';
117
- const typeParams = lang.typeParams ? '(?:<[^>\\n]*>\\s*)?' : '';
118
- out.push({ re: new RegExp(`\\b(?:${lang.keyword})\\s+${receiver}${typeParams}${S}\\b`, 'g') });
129
+ out.push({ re: new RegExp(`${keywordLead(lang)}${S}\\b`, 'g') });
119
130
  }
120
131
  if (lang.jsForms) {
121
132
  out.push({ re: new RegExp(`\\b(?:const|let|var)\\s+${S}\\s*=\\s*(?:async\\s+)?(?:function\\b|\\([^)\\n]*\\)\\s*=>|[A-Za-z_$][\\w$]*\\s*=>)`, 'g') });
@@ -183,6 +194,40 @@ function allSignatures(text, ext = '') {
183
194
  return out;
184
195
  }
185
196
 
197
+ /**
198
+ * Every definition in `text`, in text order, with the line it starts on and
199
+ * its signature key (as allSignatures reads it) — for pre-edit-guide, which
200
+ * must tell which of a hunk's definitions an edit changed and where it sits.
201
+ * `keywordOnly` (LANGS) with `keywordLeads`, which tests the text before a
202
+ * name on its line: where both hold, a name it is false for is no definition.
203
+ * null when there is no exact reading of the whole text: the language is not
204
+ * in LANGS, a definition sits on an over-long line, or any header has no
205
+ * reading.
206
+ * @returns {{ defs: {name:string, at:number, key:string}[], keywordOnly: boolean,
207
+ * keywordLeads: (before: string) => boolean }|null}
208
+ */
209
+ function definitionHeaders(text, ext = '') {
210
+ const lang = LANGS.get(String(ext).toLowerCase());
211
+ if (!lang || typeof text !== 'string') return null;
212
+ const starts = definitionStarts(text, lang, ANY_NAME, null);
213
+ if (starts === null) return null;
214
+ const defs = [];
215
+ for (const [name, lineStarts] of starts) {
216
+ for (const at of lineStarts) {
217
+ const header = readHeader(text, at, lang);
218
+ if (header === null) return null;
219
+ defs.push({ name, at, key: signatureKey(header) });
220
+ }
221
+ }
222
+ defs.sort((x, y) => x.at - y.at);
223
+ const lead = lang.keyword ? new RegExp(`${keywordLead(lang)}$`) : null;
224
+ return {
225
+ defs,
226
+ keywordOnly: lang.keywordOnly === true,
227
+ keywordLeads: (before) => lead !== null && lead.test(before),
228
+ };
229
+ }
230
+
186
231
  /**
187
232
  * Where definitions start: Map name → Set of line-start offsets. With `symbol`
188
233
  * the patterns carry it escaped in `S`; without, `S` is ANY_NAME and the name
@@ -670,6 +715,6 @@ function runMain() {
670
715
  if (require.main === module) runMain();
671
716
 
672
717
  module.exports = {
673
- extractSignatures, allSignatures, signatureChanged, findCallSiteLine, computeStopReport, formatStopContext,
718
+ extractSignatures, allSignatures, definitionHeaders, signatureChanged, findCallSiteLine, computeStopReport, formatStopContext,
674
719
  followUpOf, reportRecords, MAX_SYMBOLS_CHECKED, MAX_CALLERS_LISTED, MAX_FOLLOWUP_FILES,
675
720
  };
@@ -41,7 +41,7 @@ jobs:
41
41
  # refuses a snapshot that has no checksum sidecar.
42
42
  - name: Build snapshot
43
43
  run: |
44
- npx -y -p @sdsrs/code-graph@0.178.0 code-graph-mcp snapshot create --out "code-graph-snapshot-${GITHUB_SHA:0:7}.db.zst"
44
+ npx -y -p @sdsrs/code-graph@0.179.0 code-graph-mcp snapshot create --out "code-graph-snapshot-${GITHUB_SHA:0:7}.db.zst"
45
45
  - name: Upload to release
46
46
  env:
47
47
  GH_TOKEN: ${{ github.token }}
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@sdsrs/code-graph",
3
- "version": "0.178.0",
3
+ "version": "0.179.0",
4
4
  "description": "MCP server that indexes codebases into an AST knowledge graph with semantic search, call graph traversal, and HTTP route tracing",
5
5
  "license": "MIT",
6
6
  "repository": {
@@ -35,10 +35,10 @@
35
35
  "node": ">=16"
36
36
  },
37
37
  "optionalDependencies": {
38
- "@sdsrs/code-graph-linux-x64": "0.178.0",
39
- "@sdsrs/code-graph-linux-arm64": "0.178.0",
40
- "@sdsrs/code-graph-darwin-x64": "0.178.0",
41
- "@sdsrs/code-graph-darwin-arm64": "0.178.0",
42
- "@sdsrs/code-graph-win32-x64": "0.178.0"
38
+ "@sdsrs/code-graph-linux-x64": "0.179.0",
39
+ "@sdsrs/code-graph-linux-arm64": "0.179.0",
40
+ "@sdsrs/code-graph-darwin-x64": "0.179.0",
41
+ "@sdsrs/code-graph-darwin-arm64": "0.179.0",
42
+ "@sdsrs/code-graph-win32-x64": "0.179.0"
43
43
  }
44
44
  }