Repository navigation
0.1.7 Release - #1764
Merged
Merged
0.1.7 Release#1764
Conversation
… function stored on this, in a module variable, in a tested options member, or through an expression (#1675) unresolved_value_callee gains five reasons, each an unresolved call whose callee is a value the graph does not hold: - module_variable: `let handler = null; ... handler()` - field: `this.cb()` where cb is a field, not a method - parameter_member: `opts.onReady()` when the code treats the member as an optional function (typeof test, `&&`/`||`/`??` guard, if condition, `?.()`) - expression: `(c ? a : b)()`, `(0, cb)()`, `make()()` - dynamic_code: `new Function(s)()` A bare member call on an untyped parameter (`res.send()`, `s.trim()`) is not counted: without a type it reads the same as a library method, and counting it turned 10 of 12 independent pairs on a real Express app into unknown. No call edge changes. Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>
…iased this field, a fallback field and new this.Ctor() (#1675) The field reason of unresolved_value_callee now covers: - a field holding only null until a setter runs (`cb = null`, `this.cb = null`, `static cb = null`): the null literal no longer counts as the known value - a static `this.cb = f` / `this.cb()` pair, known by the assignment, not a field declaration - `self.cb()` with `const self = this`: any receiver valued as the instance or class - `new this.Ctor()` - `this.cb = cb || function () {}`: the call resolves to the fallback, but the parameter it falls back from is a function no caller in the graph supplies A module variable initialised from a call and never reassigned (`const log = require('debug')('app')`) is no longer a value callee: the library made it. One handed a function (`memo(Inner)`) stays open. `Model.tag2()` (a known-missing member of case 49) now names the field. No call edge changes. Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>
A word the code only writes (a local `const labelOf = new Map()`, a parameter,
`process.env.Path`, `HttpStatus.OK`, an object key `{ Bucket: … }`) was
answered as "type X (not declared here …)", its mentions counted as exact
callers with a passing verified: line. The undeclared-type fallback now needs
evidence the name is a type: `new X`, a library method of X, a type position,
a reference bound to a type, or an unbound, capitalised, unqualified name
(`Path.Combine(…)`). References the parser binds to a value are dropped.
Without that evidence the answer says nothing is declared and lists the
methods that write the word as [by name] leads, so a multi-graph answer no
longer carries a phantom section beside the graph that does declare it.
Closes #1602
Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>
…ling The resolver rewrote '#' to '.' for everything but methods, so Vault.#secret became Vault..secret and #secret became secret: no spelling reached the field. A '#' name is now looked up verbatim as a field before the rewrite. JavaScript class fields were also indexed with no owner, so even a public field's readers were only 'alongside' rows; the field now takes its class from ownerTypeLinkHash (and its own id), as TypeScript's already do. Golden cases for JS and TS, with controls: a private method, a public field, and a same-named private field in another class. Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com> Co-Authored-By: Claude <noreply@anthropic.com>
…pact --in answers for the declarations under it A scope matched any file whose path contained it, so --in tests also took parser/src/test/*-tests.ts, and impact <name> --in <dir> answered for every declaration of the name, in every language graph, all outside the scope. A scope that names a path of the repository now matches by prefix (one that names no path still matches anywhere); impact narrows its targets to the declarations under the scope, and the fan-out drops a graph whose declarations all lie outside it when another graph has one inside. Closes #1584 Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>
… keeps its own version) Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>
…er, so a fixture's Array no longer answers production calls (#1574) A test fixture that is its own program (its own tsconfig, often noLib with a hand-written interface Array) declared globals the engine treated as repository-wide, and the parser keys a declaration group by name, so the fixture's Array also merged into every other program's Array. A production xs.map(f) resolved to the fixture's map. - A client global is visible to a module its tsconfig governs or that lies in that tsconfig's directory tree; library globals and globals with no governing tsconfig stay repository-wide. - A declaration joins another's merged entity only when its program sees that declaration's file. - Readers with no module in hand (Object/Function prototype members, typeof a global, a namespace callee) read the unscoped globals plus those of a program that covers every module. New golden case 79 (nested fixture program) with a control: the fixture's own calls keep their edges. Closes #1574 Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com> Co-Authored-By: Claude <noreply@anthropic.com>
context/impact: --in matches by path prefix and scopes impact's declarations
impact: resolve a private class field (Owner.#name) in JS and TS
javascript: path says unknown, not independent, for a callee stored on this, in a module variable, in a tested options member, or computed by an expression (#1675)
path/impact: a name no declaration carries is not a type (#1602)
typescript: a global declared in one program is not in scope in another (#1574)
Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>
Merge dev (the 0.1.6 release) into 0.1.7
…ten function A call through a wrapper's closure took its value from the site's argument alone, even when the wrapper reassigned that parameter first. The wrapper's own writes to the parameter now feed the per-site answer: the written function's result always, the site's argument unless a plain, unbranched write that does not read the parameter replaced it. Writes inside the closure or a nested function do not count. Closes #1706 Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>
…r a traceback A graph.sqlite that is not a readable database (a disk that filled, a copy cut short) made every verb, from the shell and through the MCP server, die in a Python sqlite3 traceback. Each verb's door (axiomcode-path G, axiomcode-graph) now probes the graph (schema read, quick_check only when that fails or when an sqlite error escapes mid-answer), says "graph is corrupt", moves it aside with a marker that stops the build calling unchanged files up to date, rebuilds it, and answers from the last good (baseline) graph meanwhile. Another language's corrupt graph is named and left out; the main language still answers. tests/corrupt_graph.py: garbage, truncated, bad page, marker, MCP and other-language cases, with controls (sound graph, empty graph, up to date). Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com> Co-Authored-By: Claude <noreply@anthropic.com>
…port let a setter reassigns (#1708) An ES import is a live binding. A named import or a namespace member call on a module that exports a let and reassigns it in a setter runs whatever the setter stored. The engine now marks such an unresolved call as a value callee (imported_variable) through a named import, a namespace, an export specifier, a renaming re-export and export star, so path answers unknown and names the site instead of independent. An export the module never reassigns, and an exported const that resolves, keep their answers. No call edge changes. Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>
javascript: a wrapper that writes its parameter answers with the written function
plugin: a corrupt graph.sqlite is said, moved aside and rebuilt, never a traceback
…sts remote hops TypeScript had no framework-behavior phase, so no TS bundle held an ext_remote_edge row and impact/path stopped at the send. destinations.dl joins an HTTP client (fetch, an axios-style client or its create()d instance) to the route that serves it (router calls, @Controller/@get), reading the path from a literal, a template, a constant (local or imported) or a const config object. Verb must agree; the route with the fewest differing segments wins. Unjoined halves are reported as remote_unserved / remote_unsent / remote_undetermined. Framework names live in config-resolution/knobs.dl. context now prints ext_remote_edge rows as their own tier (transport, destination, confidence), never merged into call hops. impact's test closure walks a remote hop from sender to handler on its own `remote` rung, so test-impact selects the test that drives the client. Part of #1293 Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com> Co-Authored-By: Claude <noreply@anthropic.com>
typescript: cross-process remote_edge for HTTP routes; context lists remote hops
javascript: path says unknown when a call goes through an imported export let a setter reassigns (#1708)
Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>
…goldens tests: TS cases 68 and 69 expect #1710's remote_unsent section
… and reuse it Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>
…uage per answer What was wrong 1. test-impact printed `pytest <files>` for every Python selection. A Django project with no pytest in its requirements, config or CI runs `python manage.py test <dotted labels>`, and a unittest project runs `python -m unittest`; the printed command and the `next:` line built from it did not run there. The edit hook's body line reuses the same command. 2. When a changed file outside the index (a fixture) was named by test files in their text, the runner line was chosen by the majority extension of those files and listed all of them. A Python answer could print `pytest` over .cs, .java, .ts and .js fixture files that merely contain path words, and a Java answer could add a .py basename to -Dtest=. 3. The by-name tier treated __init__.py as a module name. 4. `impact` and `test-impact` answered "tests: 0" when the graph was built with --src and the tests live in a sibling tree, without saying the test tree was never read. The change - command_for hands each runner only files of its own language. - A Python selection is grouped by the nearest project directory and run with that project's runner, read in the order a project states it: a pytest config section, pytest.ini or a conftest.py; the command its CI workflow, tox, nox or Makefile runs; a declared pytest dependency; the test file's own pytest import; a Django manage.py (dotted labels, run from the manage.py directory); TestCase modules with nothing else (`python -m unittest`). With nothing to go on it stays pytest. A manage.py that is not Django's (a Flask CLI) does not count. - A selected Python file that declares no test is not handed to a runner: one that runs as a program (a __main__ guard, a shebang, a top-level sys.exit) gets `python <file>`, and a fixture or helper module under a test tree is named apart instead of being put on a pytest line that would collect nothing from it. - The text tier reads only the graph's language's files, before choosing the most specific needle, and prefers a test module the runner collects over a helper or fixture beside it. Test files of another language that name the file are counted in one line, not listed: that language's graph answers for them. No cross-language link is made. - The by-name tier no longer matches a stem every package repeats (__init__, __main__, conftest): on a --src graph it listed 15 __init__.py files as the tests "named after" an edited __init__.py. - `next:` recognises `python manage.py test` and `python -m unittest`. - impact prints a NOT SEEN line (and --json `tests_outside_src`) when no test is counted and test files of the graph's language lie outside the --src it was indexed with; test-impact repeats it under "no test". Tests - tests/test_command.py: Django, unittest and pytest runner cases, each with a near miss (pytest-django dependency, a [tool.pytest.ini_options] section, a conftest.py, a Flask manage.py, a test importing pytest, a bare def test_), a script test and a fixture module, __init__ and __main__ stems (control: pricing.py still names test_pricing.py), language filtering for Python, Java and C#, and the next: line. Each positive check fails on the old code; the near misses pass on both. - tests/changed_range.py: the text tier's command holds only the graph's language's collected test modules, with a Java and a C# file and a Python helper naming the same fixture. - tests/cases/python/tests-outside-src: --src app with tests/ outside it says NOT SEEN; the near miss (a test inside --src) is counted and says nothing. 2 of 3 checks fail on the old code, the control passes. - Suites: tests/test_command.py 107 ok; tests/fastpath.py 4 of 4; tests/changed_range.py all ok (5 of its checks fail on the old code); tests/run.py --lang python 149 of 154, the 5 failures all in lambda-is-named-by-its-place, a TypeError from the duplicate G.enclosing already on the release branch (fixed separately), not touched here; tests/run.py abstract-test-base-command 2 of 2; tests/enrich_lines.py 36 of 36; tests/hook_languages.py 7 of 7. Smoke (before is the installed build, after is this change, same graph) - A Django project (requirements and CI run manage.py test, no pytest), body edit of one model method: 126 tests in 23 files both times. Before: `pytest <app>/tests/test_x.py ...`; after: `python manage.py test <app>.tests.test_x ...`. - A library whose tox runs `-m unittest discover`, body edit of one helper: 102 tests in 7 files both times. Before: `pytest tests/...`; after: `python -m unittest tests.test_x ...`, and that command resolved all 7 modules and ran 209 tests (104 errored on this machine's missing optional dependencies, the same on the unedited tree). - A Flask project (pytest in its dependencies, a manage.py that is a Flask CLI), body edit: 683 tests in 96 files, and the pytest line is byte-identical before and after (the control). - The same project indexed with --src on its package dir, impact of the same function: before "tests: 0 of 3" and nothing else; after the same count plus "NOT SEEN: ... 104 test file(s) lie outside it". The whole repository indexed finds 683 tests for that edit. - This repository's own tests (run as scripts), test-impact on this change: before a pytest line over script tests, fixture modules and 32 files of other languages; after `pytest` for the pytest modules and `python tests/<x>.py` for each script, 3 fixture modules named apart, and "32 test file(s) in another language name them too". - Java and C#: tests/run.py abstract-test-base-command (mvn -Dtest line) and tests/hook_languages.py (dotnet test --filter line) pass unchanged. Not yet measured on the full corpus or on held-out projects; the smoke numbers above are the only corpus numbers.
…y their key, class expressions named by their binding The JavaScript index adapter dropped a field's owner (`impact Store.items` answered `field items`), left a computed-key member with an empty name (`Counter.`), and displayed an anonymous class expression as `<anonymous-class>` with its variable as a second declaration. Fields now take their class by ownerTypeLinkHash, computed keys are named `[<key as written>]` (and impact keeps them one segment), and a class expression takes the variable, field or property it is bound to, so a nested member reads `Outer.Inner.ping`. INDEX_VERSION 5 re-indexes older graphs. Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com> Co-Authored-By: Claude <noreply@anthropic.com>
…argument of its own reads the parameter (#695) The per-site answer for memoize/once/lazy wrappers dropped every value of the closure's carrying return, not only the parameter's call. `fn() || fallback`, `override || (cached = make())` and `let result = defaults; result = fn() || result` lost the fallback, the override and the default. Now only the parameter's call is answered per site; every other operand keeps its ordinary value (an `undefined` cache placeholder aside). Sites that do not pass the parameter themselves fall back instead of answering nothing: a missing argument reads the parameter's default, a spread argument reads the merged parameter as before, a wrapped getter passed to a wrapper (`memoize(memoize(f))`, `once(getAlpha)`) answers what a call through the inner site returns, and the closure called inside its own wrapper reads that invocation's parameter. Case 64 gains these shapes; its first getters and controls are unchanged. Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>
…answers per site (#695) `function once(fn) { let r; function inner() { if (!r) r = fn(); return r; } return inner; }` returns its closure by name, and a named function declaration was not recognised as the wrapper's closure (only a returned function expression or a const bound to one). So `getter().method()` read the merged parameter: a superset, or nothing once the wrapper was over the fan cap. A returned binding of a function declared inside the wrapper is now its closure too. Case 64 gains declared.js: two instance getters and a wrapped memoized getter over a 27-caller declared wrapper, each reaching its own module; controls (a declared closure that returns something else, a function declared outside the wrapper) are unchanged. Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>
…on the declared-closure fixture Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>
…swers unknown A method calling this.hook() where neither the class, an ancestor nor any write the graph follows gives the instance a hook runs a function stored from outside (e.hook = fn, Object.assign(this, opts)). It is now an unresolved value callee (instance_member), so path says unknown and names the site instead of independent. Classes with an unresolved or dynamic base are left out, as are ambient members (hasOwnProperty). Closes #1709 Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>
impact, path, context: when the graph has no answer, give the text matches, labelled [text]
…-mislabelled impact: label each link of a chained route registration with its own verb and path
…ed-key javascript: a call through an object-literal table with a computed key reaches the table's functions
javascript: this.x() with no declared x answers unknown in path
A call through an object literal resolved only when the receiver was the bare const in its own file. An imported const (`api.fetch()`), a nested literal (`api.users.list()`) and a factory returning a literal (`makeStore().inc()`) all stayed unresolved, for methods and for arrow- or function-valued properties alike. expr_objlit follows each of those roads to the literal the receiver evaluates to; an annotated variable or return type is left to the type it names. Case 80 covers the shapes, with annotated and non-literal controls. Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>
…w resolve Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>
…that passes it
`xs.forEach(this.handle, this)`, `el.addEventListener('click', this.onClick)` and
`setTimeout(this.onHover.bind(this))` hand an instance method over without calling it, and the
passer had no row for it, where a free function passed at the same position had one.
method_value reads the member off the receiver's type (or shape) the way an instance call does,
sees through `.bind`, and leaves getters, setters, data fields and `.call`/`.apply` alone.
Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>
… page 2 `impact --page 2` was rated not useful in 10 of 11 recorded calls: - Page 2 was page 1's view cut further along. The default view caps the users at 25 rows and 5 per file, and collapses a [by name] flood, so page 2 printed the few rows past page 1's budget and then the same "... +54 more" and "+9 [by name]" summaries. The rows those summaries stood for were on no page, and `--page all` still hid 63 of them. - A one-page answer asked for page 2 printed page 1 again, with no word that there was no page 2. - A section continued on a later page repeated the whole answer's counts in its heading, so the numbers did not match what the page showed. The entry-point rows were moved under a "low-certainty rows" heading and left their own sub-heading behind, empty. - The collapsed [by name] rows came with a search for the target's own method name, which for a type or a property is a compiler-made name: grep "<init>", "<classbody>", "<lambda>". - Through MCP the footer read "page=3 (MCP) or page=3 (CLI), or all of it with page all", and `page` took only an integer, so page="all" could not be sent. The change: - ax_pages: page 1 is the default view, unchanged. Pages 2 and on hold every row page 1 did not print, in the order of the full answer, with no summary lines. impact marks the rows its default view hides and the lines that summarise them, so the pager builds both views from one run. `--page all` prints every row. - A page past the end of a one-page answer says "page 2 does not exist: this answer has 1 page". - A continued section's heading counts the rows on that page, by rung; entry-point rows keep their sub-heading and are headed as entry points. A section with no rows on a later page is not printed. - The [by name] search names what the collapsed rows call, or the target's own name (the type, the field, the method); never a name starting with "<" or a dunder. - MCP: `page` takes a number or "all"; `--page all` is written page="all" and the footer names one spelling per surface. The SDK-free server advertises the union as anyOf. The impact description says tests=True is the CLI's --tests-only. - axiomcode-path: two methods named G.enclosing (one per lambda, one per file and line) were merged into one class, and the later one replaced the earlier, so every lambda lookup raised a TypeError (test-impact on named files, the lambda case). The file-and-line one is enclosing_at. Tests: a Python case with a near-miss control for each behaviour (page 2 lists the three hidden callers and no summary; a one-page answer has no page 2, and without --page is printed whole; a type's search names the type, a method's keeps `.value(`), the Java paging case's footer, and tests/mcp.py for page="all", page=1 and the footer wording. The enclosing fix turns 5 of lambda-is-named-by-its-place's 7 failures on the base green; its other 2 (a Python control, a C# field initializer) and member-owner-is-its-type's stale pending marker fail on the base too, with the same output, and are not touched here. Smoke, same graphs, installed build vs this change: - Java REST app, impact on an entity type with 138 users, page 2: 9 new rows and 2 summary lines before; 15 new rows and none after, headed "15 row(s): 15 resolved". Pages 1 to 7 together print 120 of the 120 rows, none twice. - One-page answers asked for page 2 (one Java, one Python, one C# target): page 1 repeated byte for byte before, 3 of 3; "does not exist" after, 3 of 3, and page 1 unchanged. - Type targets: the search was grep "<classbody>" (Python) and a compiler-made name (C#); now grep -rnw on the type's own name, both. A C# property: the names of unrelated methods before, the property's own name after; a method keeps its `.name(` search. Not yet measured on the full corpus or on held-out projects; the smoke numbers above are the only corpus numbers.
… Record/index-signature elements
A call's result was untyped when its type came from:
- a function-typed parameter with its own type parameter, bound from the
argument: `pick: <T>(x: T) => T; pick(u).save()`. Function-type parameters
never linked their declared type, so T had nothing to bind from;
- a signature's own `<T>` (function type, method/call/construct signature in a
type literal), which was not in scope for its return, so `T` referenced nothing;
- a `Record<K, V>` or inline `{ [k: string]: V }` read by key, including a value
that is itself a function type: `factories[kind](n).area()`.
Case 80 covers each shape plus controls that must stay unresolved
(`get<any>()`, an index signature over `any`).
Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>
…h site A method returning `new this.constructor(...)`, directly or through a const it initialises, was read as one merged value: every instantiated subclass of the declaring class, so each call on a clone fanned to all their overrides. Like a `this`-return, it is now substituted per site from the call's receiver. A `let` local and `new this()` in a static keep their merged value. Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com> Co-Authored-By: Claude <noreply@anthropic.com>
…s length (#1601) Line 0, or a line past the last, spans no callable and fell through to the module node, so it was answered as the file's top-level code as if that line existed. Both verbs resolve file:line in G._resolve; it now refuses a line below 1 and, reading the file on disk, a line past its end. Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com> Co-Authored-By: Claude <noreply@anthropic.com>
…ler reports Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>
… own convention A same-named file under a test directory is not a test: fixture inputs (test/fixtures/*/index.js), another language's test and the changed file itself were offered as the tests of src/index.js, with a runner command. A by-name candidate is now a test only by its ecosystem's runner default (x.test.ts, test_x.py, XTest.java, XTests.cs, or directly in a mocha test/ or __tests__/ dir) in the changed file's language family; any other same-named file is counted on a 'not a test' line with no command. Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>
…eceiver typescript: type receivers from a generic signature's T and from Record/index-signature elements
… its own command A test-tree file run as a program with no framework (a runner calling runCase at top level, a tests/check.py with a main guard) was walked to and labelled [test], then counted "0 of 0", and test-impact said no test reaches the change. Its module is now in test_method, from one rule (graph_sql.script_tests) that the exporter and the hooks' fast path both read, so impact counts it and test-impact selects it. The command is the one the project runs it by: its package script, its usage line, else the interpreter; never a guessed framework. Helpers that export, setup and config files, and test data (cases/, fixture projects under src/, tools/) are not scripts. The impact facts version moves to 40 so a cached export is rebuilt. Also: axiomcode-path declared G.enclosing twice, the second shadowing the first, so `changed <file>` without git crashed on a top-level lambda. The file:line form is now G.enclosing_at. Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>
…ds its own entry An unguarded script test that no package script or usage line names was handed to node or npx tsx, which undid #1570 (a file nothing runs is said so). It still counts and is selected; only a file with require.main === module, import.meta.main or a __main__ guard gets its interpreter as the command. Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>
impact, test-impact: a script-style test counts as a test and runs by its own command
…-callback typescript: a method passed by reference registers with the function that passes it
…embers typescript: resolve calls on members of an unannotated object literal
javascript: new this.constructor() builds the receiver's class at each site
path / impact: a file:line outside the file is refused with the file's length
…matches A file:line whose partial path (a basename, or dir/file) matches several indexed files was refused before --in was applied, so the scope could not pick among them. G.line_files now narrows the matches by --in: one file under the scope is the answer, labelled with its full path; two or more are the refusal's list; none keeps the whole list and says so. impact's field lookup for file:line matched the line in every file the path ends, so test.js:6 among 17 test.js answered for one file's const; it now resolves the file the same way and refuses when the path is ambiguous. Closes #1666 Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>
impact / path: --in chooses among the files a partial file:line path matches
test-impact: by-name fallback offers only real tests of the changed file's language
Release 0.1.7
swapnilpaliwal-sd
requested review from
JaredHLZhang,
Whua689 and
suyashpaliwal26
as code owners
September 29, 2026 03:54
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Improvements
Bugs
export letthat a setter reassigns; it now says unknown javascript