Release 0.1.6 by swapnilpaliwal-sd · Pull Request #1712 · AxiomCodeAI/axiomcodegraph

swapnilpaliwal-sd · 2026-09-28T20:48:37Z

Promotes 0.1.6 to dev: 26 fixes merged into 0.1.6.

javascript: a const holding a wrapped handler is registered at its route (javascript: a const holding a wrapped handler is registered at its route #1638)
impact, test-impact: a call inside a mock stub is not a test route, and a request reaches the handler for its path and verb (impact, test-impact: a call inside a mock stub is not a test route, and a request reaches the handler for its path and verb #1646)
typescript: a callback handed to a library higher-order function is reached only from the site that hands it over (typescript: a callback handed to a library higher-order function is reached only from the site that hands it over #1653)
path / impact: a file:line whose partial path matches two files names both and asks which (impact / path: a file:line whose partial path matches two files answers for one of them, the one with the shorter function #1599) (path / impact: a file:line whose partial path matches two files names both and asks which (#1599) #1654)
test-impact: each TypeScript/JavaScript test file gets the command its package runs it with (test-impact: the TypeScript command is npx vitest run on files vitest does not collect #1570) (test-impact: each TypeScript/JavaScript test file gets the command its package runs it with (#1570) #1655)
csharp: a field, const, property or enum member is owned by its declaring type (csharp: impact on Type.Field, Type.Const or Type.Property is answered as an undeclared type name and called local #1445) (csharp: a field, const, property or enum member is owned by its declaring type (#1445) #1670)
parser: 0.1.2 on 0.1.6 (forward-port of parser: 0.1.2, for the parser changes in #1592, #1616, #1621 and #1625 #1668) (parser: 0.1.2 on 0.1.6 (forward-port of #1668) #1672)
impact: a test that only stubs the change is not a test row to run (impact: a test that only stubs the change is not a test row to run #1677)
freshness: a graph built by an older engine, other rules or another IMPACT_VERSION is rebuilt, and the answer says so (python/java/csharp: the first query after an edit answers from the pre-edit graph, because the 10 s wait is shorter than every measured rebuild #1595) (freshness: a graph built by an older engine, other rules or another IMPACT_VERSION is rebuilt, and the answer says so (#1595) #1678)
javascript: path says unknown, not independent, when one side calls through a value (javascript: path calls two functions independent though one makes an unresolved call through a parameter or loop variable #1649) (javascript: path says unknown, not independent, when one side calls through a value (#1649) #1679)
ci: faster engine suites (concurrent cases, an engine cache that hits) (ci: faster engine suites (concurrent cases, an engine cache that hits) #1680)
index: the live graph link is relative, so a copied checkout keeps its own graph (index: the live graph link is relative, so a copied checkout keeps its own graph #1681)
changed, impact: a lambda is named by where it is; file:line takes a field or type line (csharp: impact and path by file:line on a class declaration line answer 'no callable spans' #1426) (changed, impact: a lambda is named by where it is; file:line takes a field or type line (#1426) #1682)
impact: a caller through an interface or base method is a direct caller, and an interface-typed injection receives the bean (java: impact on an interface method omits a caller injected with @Autowired and says the change is local #1542, java: impact says a caller injected through an interface did not get its bean from the container #1384) (impact: a caller through an interface or base method is a direct caller, and an interface-typed injection receives the bean (#1542, #1384) #1683)
typescript: re-bless case 78 for the callback edges typescript: a callback handed to a library higher-order function is reached only from the site that hands it over #1653 adds (typescript: a function held in an array or table is run by the call that iterates or indexes it (#967) #1657) (typescript: re-bless case 78 for the callback edges #1653 adds (#1657) #1684)
javascript: a class field initializer spans its fields, not the class (javascript: a class field initializer spans its fields, not the class #1688)
mcp, cli: a list of sites comes one per line, path:line: code [tag] (mcp, cli: a list of sites comes one per line, path:line: code [tag] #1689)
typescript: a published module's classes, method-reference consts, namespaces, object literals and default arrows are entry points (typescript: a published module's classes, method-reference consts, namespaces, object literals and default arrows are entry points #1692)
Merge 0.1.5 into 0.1.6: the 0.1.4 release fixes (javascript: a declaration on the same line as an arrow is treated as nested in it, so impact and path walk a fake defines hop #1598, the Windows launcher) (Merge 0.1.5 into 0.1.6: the 0.1.4 release fixes (#1598, the Windows launcher) #1695)
typescript: a || / ?? / && default carries the callback's call signature (typescript: a || / ?? / && default carries the callback's call signature #1697)
context: a tie between entry points goes to code a reader can follow (context: a tie between entry points goes to code a reader can follow #1700)
typescript: resolve calls through middleware parameters typed by an indexed access (typescript: resolve calls through middleware parameters typed by an indexed access #1701)
impact: follow a TypeScript structural type to the sites that never name it (impact: follow a TypeScript structural type to the sites that never name it #1702)
Read hook: a call in the text read is not an edge the text does not show (Read hook: a call in the text read is not an edge the text does not show #1703)
typescript: try a merged interface's later declaration block first (typescript: try a merged interface's later declaration block first #1705)
query rules: never run a cached binary built for another machine (query rules: never run a cached binary built for another machine #1707)

Work continues on the next patch branch, cut from 0.1.6.

… keeps its own version) Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>

… both and asks which (#1599) A partial path (a basename, or dir/file) took the smallest span across every file it matched, so the answer was for whichever function at that line was shorter, labelled with what was typed. The file is now settled first: the path written in full, else the one file it is a suffix of; two or more are refused with each file and what spans the line there. Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com> Co-Authored-By: Claude <noreply@anthropic.com>

The vendored parser keeps its own version and moves once per release, only when parser code changed. dev and main ship it at 0.1.1. On 0.1.4, four merged changes touched parser/: #1592 (a concise arrow body is owned by the arrow), #1616 (generated-output detection), #1621 (a package's published API is rooted from package.json) and #1625 (the Java walk and build directories). So 0.1.2, one bump for the release. (cherry picked from commit 51c5e8e) Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>

parser: 0.1.2 on 0.1.6 (forward-port of #1668)

path / impact: a file:line whose partial path matches two files names both and asks which (#1599)

…eached from the site that hands it over, not from every caller of the same signature `xs.map((x) => ...)` stored the arrow in the parameter of a bodiless `map` declaration, so it became a value candidate of one function type shared by every `.map` call: impact through one map callback reached every caller of `.map` in the repository, and a walk down from any caller reached every map callback. Where no project body runs at the call site (a .d.ts member, an ambient declaration, an unresolved callee), the argument no longer flows into the parameter. The site itself gets a `callback_registered` edge to the function it hands over, the tier the JavaScript engine already uses for the same shape. A project higher-order function with a body keeps its dispatch through the parameter. Golden case 78 covers inline, named, local-held and ambient hand-overs, with a project HOF as control. Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>

…nd a request reaches the handler that answers its path and verb A call written inside a mocking library's stub or verification (when(x.m()), verify(x).m(), doReturn(v).when(x).m(), Moq Setup/Verify lambdas, NSubstitute Received()/Returns()) resolved to the real method and counted as a sound test route, though the receiver is a mock and the body never runs. Such a site is now the tier "stub": kept as a [stubs it] dependent, kept out of the closure and of path, and the tests that only stub the change (or reach it only through a type their class holds as a mock) are listed apart instead of counted. test-impact selects them only for a signature change or a removal. A handler's route is now its type's mapping prefix joined to its own mapping, and a route key joins only requests that send the handler's HTTP method; a path a handler registers verbatim is not also aliased to a parameterised sibling. When no test is counted and the change has framework-entered entry points, the answer says so and gives the search that finds those tests.

typescript: a callback handed to a library higher-order function is reached only from the site that hands it over

impact, test-impact: a call inside a mock stub is not a test route, and a request reaches the handler for its path and verb

Case 78 (callable collections, #1657) was blessed on 0.1.5, before the rule that a function handed to a bodiless library method gets a callback_registered edge from the site that hands it over (#1653, on 0.1.6 only). With both on 0.1.6 the case gains three such edges and loses none: registry.push(pushed) module init -> pushed steps.forEach((step) => step(node)) runForEach -> its arrow frozen.reduce((acc, step) => ...) runReduce -> its arrow Each is the shape #1653 labels on purpose, and the JavaScript engine labels the same way. Only the .edges golden moves; the oracle, envelope and entry goldens are unchanged.

typescript: re-bless case 78 for the callback edges #1653 adds (#1657)

impact, path, context and test-impact take --grep (--grep-limit N): the verb runs with --json and ax_grep.py prints each site the way grep does, with the graph's certainty as a trailing tag, capped with the rest counted by kind. Tests fold to one row per test file (+N more in this file), so one file of many tests no longer spends the whole cap. The MCP tools answer this way by default; full=True, and parameters only the prose carries (source, why, delete, page), give the verb's own answer unchanged. Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>

…field or type line (#1426) changed, test-impact and the edit hooks charged an edit inside a lambda to the lambda itself. Every lambda carries the one name <lambda>, so the edit came back as a signature change of <lambda> (or Owner.<lambda> in C#), impact matched every lambda in the graph, and test-impact ran their tests. - changed: an edited line is charged to the narrowest declaration that is not a lambda. An edit inside a lambda in a method is that method's body change; inside a field initializer it is the field's. A lambda nothing encloses is its own body change, named Enclosing.<lambda@L42> and targeted by file:line. Its parameter list is read from the lambda header, never as a method signature. An expression body after => is no longer read as header text. - path, impact: Owner.method.<lambda@L42> resolves to that lambda, and a lambda answer is labelled that way. - impact, path: file:line on a field, property, constant or type header line answers for that declaration instead of "no callable spans"; a callable written on the line still wins. Top-level code at a line is the module body that holds it, not the first class body of the file. - Cases for Python, Java and C# with near-miss controls: two lambdas in one method stay distinct, and an edit to a named method next to a lambda stays that method's.

…checkout keeps its own graph graph.sqlite pointed at its graph by an absolute path. A copy of the repository kept it: the copy answered from the original's graph without its own edits, and once the original moved it found the link dangling and rebuilt itself as "a build was interrupted". - axiomcode-build makes the link from inside .axiomcode/out, relative (typescript/graph.sqlite, .live.sqlite during a rebuild), and resolves a relative link where it reads one. - ax_fresh.relink, run by every query verb, by the build and by `graph`: an absolute link an older build left is re-pointed at this checkout's own graph. Into another checkout it says so (copied, or moved when the original is gone); with no graph here to take, the link is removed and one is built. - "a build was interrupted" now names the dangling target, and is said only for this checkout's own link. Tests (refresh_races.py, copied checkout): a fresh index writes a relative link; a copy reads and rebuilds its own graph; an absolute link into another checkout, still there or gone, is re-pointed with no rebuild. Controls: the original does not see the copy's edit; an absolute link into its own graph is kept silently. Closes #1605 Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>

…e order Every suite was one serial loop over cases/*/, and a case is a separate program whose few seconds are almost all per-process start-up (node, python, the solver), so a 4-CPU runner sat three-quarters idle: the TypeScript loop alone was 6.5 minutes of 79 cases at 4-7 s. graph/test/tools/case-pool.sh runs up to AXIOM_SUITE_JOBS cases at once (default: the CPU count). Each case's stdout and stderr go to its own file and are printed whole, in case order, as soon as every case before it has finished, and the counters the loop body updates (pass/fail and the failed list, under whatever names the suite uses) are summed in case order, so the report, the summary and the exit status are what the serial loop printed. The loop bodies are unchanged: each is wrapped in a one-case `for` so its `continue`s still mean "next case". AXIOM_SUITE_JOBS=1 is the old loop verbatim, nothing captured. --bless, --keep, --oracle and name filters behave as before; a body that calls `exit` still ends the suite with its status; an interrupt stops every running case. What concurrent cases must not share: * the compiled engine: the first case past the filter runs alone, so it compiles or restores the engine once and every later case finds it cached; * staged library facts: run-souffle.sh keys them on module directory names and CSV size+mtime in whole seconds. Every case's library IR lives in a directory of the same name (`$w/libir`), so two cases parsed in the same second could share a key while holding different rows. Serially that cannot happen; side by side it can. Each concurrent case stages into its own AXIOM_LIBFACTS_CACHE, a new knob that moves only those facts (the engine binary stays in the shared cache). Everything else a case touches is its own work directory; the oracles and normalisers write to stdout or into it, and the java agreement classes and C# engine are built before the loop or cached by atomic rename. TypeScript: the seven fixtures (linking, dispatch, typeflow, overloads, specificity, tsconfig-chain, multi-program) were the minute after the last case before the summary. They are now jobs in the same pool, each in its own $WORK-<fixture> as before, so they run beside the cases. The pass/fail counters are also initialised before the source-root check that adds to them; they were set after it, so its failure was counted and then zeroed. Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>

…saved by release branches The engine cache key carried a hash of the runner's CPU model and feature flags, because the driver compiles with -march=native and a binary restored onto a different hosted runner died with SIGILL. Which CPU a job lands on is luck, so the key missed on nearly every run and each language paid a three-minute compile for rules nothing had changed. run-souffle.sh takes AXIOM_ENGINE_MARCH: `native` stays the default, `portable` compiles for the compiler's baseline target, as build-engines.yml builds the published engines; any other value is passed as -march=<value>. The engine job sets `portable`, so any runner can run any runner's binary, and the key names the setting instead of the CPU. The SIGILL guard stays for the native default. A cache saved by a pull request is restorable by that pull request alone; the ones every pull request can restore are its base branch's and the default branch's. Pull requests target release branches (0.1.6, ...), which CI never ran on, so nothing was ever saved where the next pull request could find it. A push to a `0.*` branch now runs the suites (not the platform engine build), which saves the engine for that branch's rules. Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>

actions/cache saves only after a successful job, so a pull request that was red — the one whose next push most needs to be fast — recompiled its engine on every push, and a release branch whose suite was red saved nothing for the pull requests into it. The binary does not depend on the verdict: run-souffle.sh publishes it only whole and verified. The cache step is now a restore before the suite and a save after it that runs whatever the result, except when the job was cancelled or this language's engine is not there to save (a key saved without it would stay taken and could never be filled). Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>

Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>

…hrough a parameter, a loop variable, a local or a computed member (#1649) The engine marks an unresolved call whose callee is a value rather than a name (unresolved_value_callee: parameter, loop_variable, local, computed, dynamic_code). path reads it: when no chain and no by-name lead connects the two, a side whose closure makes such a call now answers that the connection is unknown and names the sites, instead of "independent". A call through an imported binding, a resolved call and a function that calls nothing keep their answers. No call edge changes. Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>

…for its parameter call Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>

…PACT_VERSION is stale The file table now records what built the graph (the engine's version and content hash, the plugin's rules hash, IMPACT_VERSION). `index`, the queries and the refresher compare it with the axiomcode that would build now: a difference rebuilds, with no file changed, and says "graph built by an older axiomcode (<old> -> <new>)". A query answers from the old graph at once with that line while the refresher rebuilds; --fresh waits for it. The same engine, even reinstalled in another directory, with no edit, is still up to date.

…hose engine matches the graph The failed change set is keyed on the files only; the engine difference it failed on is kept beside it and compared only when the caller sees one.

The tests line now also names callables in test code that the count does not credit. A test that only stubs the change on a mock, or reaches it only through a type it holds as a mock, is listed apart as [stubs it] and is not one of them; without this it was told both "not counted" and "treat as a test to run". Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>

…ring type (#1445) The index mapped C# member declarations with no owner and a qualified_name of file#name, although every IR row links its declaring type by csTypeLinkHash. So impact Type.Field, Type.Prop and Type.CONST fell back to a type-name match and called the change local. The owner is now that type row: display Type.Member (Outer.Inner.Member when nested) and qualified_name <type qualified name>.<name>. A property then answers through its accessors, readers and writers resolved; a field or const lists readers in scope and by name, since no C# field access is resolved by the engine yet. New case with a same-named near-miss on another type.

…mespaces, object literals and default arrows are entry points A module named by package.json rooted only its exported functions and arrow consts. The public members of an exported class (and the classes it inherits from), a const bound to a method reference (`instance.m`, `instance.m.bind(x)`, `Klass.staticM`) or holding a function by any holder route, the callables of an exported object literal, the exported functions of an exported namespace, and an anonymous `export default` arrow are now package_export roots too. `private` / `#name` members and non-exported namespace functions stay out. Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>

…s package runs it with (#1570) `npx vitest run <files>` was printed for every TS/JS selection, so a script suite (`src/test/x-tests.ts`, run as `npx tsx src/test/x-tests.ts`) got a command that exits 1 with "No test files found". Each file is now placed with the runner whose include globs collect it (vitest, jest, mocha: config file, package.json key, deps or scripts), run from its package; failing that, the package script, the file's own usage line, or the package README that runs it; failing that, it is named as collected by no runner rather than handed to one. The --json answer carries `not_collected`, and `next:` names every command. Control: a *.test.ts file the package's vitest collects keeps `npx vitest run`, and with nothing configured anywhere the command is unchanged; Python and Java commands are untouched. Closes #1570 Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com> Co-Authored-By: Claude <noreply@anthropic.com>

…s it (#1570) The per-file runner placement read the first `include`/`testMatch` in a config file, so several shapes lost a command that ran or got one that does not: - vitest: the test include is the one under `test:` (not `coverage.include`); each inline project collects on its own; `exclude` takes files back out; vite.config.* is the config when there is no vitest.config.*; a workspace member is read from its own config or with vitest's defaults. - jest: testMatch is relative to <rootDir>, testRegex replaces it, rootDir/roots/ testPathIgnorePatterns are applied, and `jest --config x` in a script is a config too; a glob like `.[jt]s?(x)` is read whole, and a nested e2e/jest.config.ts runs its own files. - playwright (testDir/testMatch), `node --test` / `tsx --test` scripts and react-scripts get their own commands. - a runner with a config outranks one that is only a dependency; with a package.json and no runner nothing is handed to vitest; a lint script naming the file is not its test command; a test script's own glob can run it; paths with spaces are quoted. Control: each shape has a nearby case that keeps its command (a file outside the include, a default-named file under testRegex, a vitest config beside a jest dependency, a lint glob). Closes #1570 Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>

…ommand share one signature command_for takes the graph (for Java/C# concrete test classes) and the repository (for each TS/JS file's package and runner): command_for(lang, files, classes, db=None, repo='.'). The edit hooks pass the repository they located, so a TypeScript body edit gets the command its package runs, and a multi-package plan stays on the hook's one line. Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>

A config was read as the literal keys it spells, so an include behind a const, a shorthand property, a spread of configDefaults, a re-export or a mergeConfig base was taken as the runner's defaults, and a file the runner collects was named as collected by nothing. The config is now evaluated: defineConfig and config functions, consts, shorthand, spreads of the runners' defaults, mergeConfig, re-exports and relative imports. What it still cannot read is unknown, and unknown never excludes a file. Also: vitest test.dir, workspace members configured by vite.config.*, `vitest --config x` scripts, jest `!glob` in testMatch, and mocha --extension/--recursive run with the script's flags. Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>

…auncher) Both branches had bumped IMPACT_VERSION to 33 and EXPORT_VERSION to 3 for different changes, so either number alone would let a cache built by the other code survive. The merge keeps both changes and moves to IMPACT_VERSION 35 (34 is taken by #1683) and EXPORT_VERSION 4.

…ame it Adding a required property to an interface breaks code that is checked against it without writing its name. Three such shapes are now answered: - a union or object type alias over the type: its users break, and they name only the alias (followed through aliases of aliases); a type alias is not listed as an entry point; - an object literal carrying the tag of a discriminated interface, `{ kind: 'LiteralNode', ... }`, reached for the interface and its bases; - an intrinsic JSX element whose attributes the project's IntrinsicElements table checks against the props type. Rows are [by name]. Every TypeScript interface answer also says that other structurally checked sites are not in it and that `tsc --noEmit` lists them. Closes #784 Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>

The compiler lists a merged interface's overloads block by block with each later block ahead of the earlier ones, so a module augmentation's overload, or lib.es2015.symbol.wellknown's String.match/replace/split/search, wins over the original when the argument fits both. The engine committed to the earlier block because the exact-type specificity rule preferred it. - order blocks only where the IR states it: augmentation after its target, same-file blocks in source order, lib.es5 before the other lib.es* files; overload indexes compare only within one block - the later block wins only where every argument is shown to fit it (concrete match, callback, or an argument type declaring every required member), so s.match("x") stays on lib.es5 - an augmentation block sees its file's imports and top-level names, so `find(m: Matcher)` inside `declare module "./text"` has a parameter type - a library argument type that declares every member a client interface requires is no longer rejected against it (prune-only) Case 78-merged-interface-later-block-first pins both orders with controls (Bare against Matcher keeps the first block). Closes #422 Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>

A compiled query program left in the plugin's legacy dl/.cache is named by its rules alone, so a plugin copy synced from a Mac to a Linux machine (or back) carries the other machine's binary under the right name. impact and path then ran it and died with OSError: Exec format error. compiled() now reads the binary's header (ELF, Mach-O, PE and the CPU they name) and skips one built for another OS or CPU; the query falls back to the interpreter and compiles for this machine in the background, as it does when nothing is cached. A binary this machine compiled is still used. Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>

…dler-registration javascript: a const holding a wrapped handler is registered at its route

test-impact: each TypeScript/JavaScript test file gets the command its package runs it with (#1570)

ci: faster engine suites (concurrent cases, an engine cache that hits)

javascript: a class field initializer spans its fields, not the class

javascript: path says unknown, not independent, when one side calls through a value (#1649)

csharp: a field, const, property or enum member is owned by its declaring type (#1445)

impact: a test that only stubs the change is not a test row to run

…hange freshness: a graph built by an older engine, other rules or another IMPACT_VERSION is rebuilt, and the answer says so (#1595)

…le-line changed, impact: a lambda is named by where it is; file:line takes a field or type line (#1426)

@Autowired

…er, and an interface-typed injection receives the bean (#1542, #1384) A service-layer caller holds an interface-typed field and calls through it. impact listed that caller only under "reaches those" for the implementation, and not at all for the interface method when the engine had narrowed the injected field to its one bean, so the answer read "few callers" or "the change is local". - graph_sql.via_base_rows: a call site typed on a base b (its targets are b and overrides of b, or b alone) is a direct row for each override the engine kept there, worded "calls it (via the interface)" or "(via the base class)"; resolved only where nothing else can run there, one of a set otherwise. For b itself a multi_inferred edge is the call the code names: "calls it", resolved. A call narrowed to one override is a caller of b when the receiver read there is a field declared with b's type. An override left out of a site's set gets no row, so a caller whose receiver resolves to a sibling implementation does not become a caller of this one. - graph_sql.bean_injections: an injection point the engine wired to one bean receives that bean whatever type it declares, so an @Autowired interface field or constructor parameter is "on the injected bean". - The rules (dl/impact.dl via_base, via_site, injected_bean, direct_edge) and the SQL port read the same rows; the plain row a via row replaces is suppressed, so a caller is listed once. IMPACT_VERSION 34. - The skill's impact reference names the new wording. - Cases: java/callers-through-the-interface, csharp/callers-through-the-interface, python/callers-through-the-base-class, each with a sibling-implementation near-miss control. java/next-step-reads-the-whole-answer now expects the call line the base-class caller is listed at.

…allers-direct impact: a caller through an interface or base method is a direct caller, and an interface-typed injection receives the bean (#1542, #1384)

context: a tie between entry points goes to code a reader can follow

…oluse-read Read hook: a call in the text read is not an edge the text does not show

query rules: never run a cached binary built for another machine

typescript: a published module's classes, method-reference consts, namespaces, object literals and default arrows are entry points

typescript: a || / ?? / && default carries the callback's call signature

typescript: resolve calls through middleware parameters typed by an indexed access

typescript: try a merged interface's later declaration block first

impact: follow a TypeScript structural type to the sites that never name it

index: the live graph link is relative, so a copied checkout keeps its own graph

mcp, cli: a list of sites comes one per line, path:line: code [tag]

swapnilpaliwal-sd and others added 30 commits September 28, 2026 09:22

release 0.1.6: every package and plugin manifest at 0.1.6 (the parser…

cbf73f9

… keeps its own version) Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>

Merge pull request #1672 from AxiomCodeAI/release/0.1.6-parser-0.1.2

cbd42b2

parser: 0.1.2 on 0.1.6 (forward-port of #1668)

Merge pull request #1654 from AxiomCodeAI/loop/ja-is1599

39451e0

path / impact: a file:line whose partial path matches two files names both and asks which (#1599)

Merge pull request #1653 from AxiomCodeAI/loop/ty-hof-callback-fanout

017c3a0

typescript: a callback handed to a library higher-order function is reached only from the site that hands it over

Merge pull request #1646 from AxiomCodeAI/fix/test-route-mock-stub

f7fc0d1

impact, test-impact: a call inside a mock stub is not a test route, and a request reaches the handler for its path and verb

Merge pull request #1684 from AxiomCodeAI/fix/1657-case78-rebless

aaf4267

typescript: re-bless case 78 for the callback edges #1653 adds (#1657)

tests: each suite's header names AXIOM_SUITE_JOBS

06e8b75

Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>

tests: 64-memoize-wrapper-result expects the value_callee diagnostic …

f930fce

…for its parameter call Co-authored-by: axiomcode-bot[bot] <334110751+axiomcode-bot[bot]@users.noreply.github.com>

freshness: a failed rebuild is still reported as failed to a caller w…

623fd1c

…hose engine matches the graph The failed change set is keyed on the files only; the engine difference it failed on is kept beside it and compared only when the caller sees one.

swapnilpaliwal-sd and others added 24 commits September 28, 2026 12:49

Merge pull request #1638 from AxiomCodeAI/loop/ja-express-wrapped-han…

6ce528c

…dler-registration javascript: a const holding a wrapped handler is registered at its route

Merge pull request #1655 from AxiomCodeAI/loop/to-t7

e2dfa59

test-impact: each TypeScript/JavaScript test file gets the command its package runs it with (#1570)

Merge pull request #1680 from AxiomCodeAI/ci/faster-engine-suites

8372620

ci: faster engine suites (concurrent cases, an engine cache that hits)

Merge pull request #1688 from AxiomCodeAI/loop/ja-is1663

6b6ff58

javascript: a class field initializer spans its fields, not the class

Merge pull request #1679 from AxiomCodeAI/loop/ja-is1649

e2c3ce0

javascript: path says unknown, not independent, when one side calls through a value (#1649)

Merge pull request #1670 from AxiomCodeAI/fix/1445-csharp-field-owner

ecc2dcb

csharp: a field, const, property or enum member is owned by its declaring type (#1445)

Merge pull request #1677 from AxiomCodeAI/fix/1646-stub-tests-not-to-run

fe5f94e

impact: a test that only stubs the change is not a test row to run

Merge pull request #1678 from AxiomCodeAI/fix/index-stale-on-engine-c…

d89b631

…hange freshness: a graph built by an older engine, other rules or another IMPACT_VERSION is rebuilt, and the answer says so (#1595)

Merge pull request #1682 from AxiomCodeAI/fix/1426-lambda-identity-fi…

4bbcd10

…le-line changed, impact: a lambda is named by where it is; file:line takes a field or type line (#1426)

Merge pull request #1683 from AxiomCodeAI/fix/1542-impact-interface-c…

0a9a240

…allers-direct impact: a caller through an interface or base method is a direct caller, and an interface-typed injection receives the bean (#1542, #1384)

Merge pull request #1700 from AxiomCodeAI/loop/to-regress-cli-context

ba505ad

context: a tie between entry points goes to code a reader can follow

Merge pull request #1703 from AxiomCodeAI/loop/to-regress-hook-postto…

87fb94c

…oluse-read Read hook: a call in the text read is not an edge the text does not show

Merge pull request #1707 from AxiomCodeAI/loop/to-regress-crashes

c15b8d3

query rules: never run a cached binary built for another machine

Merge pull request #1692 from AxiomCodeAI/loop/ty-is1626

f32502c

typescript: a published module's classes, method-reference consts, namespaces, object literals and default arrows are entry points

Merge pull request #1697 from AxiomCodeAI/loop/ty-is146

ee1e6bd

typescript: a || / ?? / && default carries the callback's call signature

Merge pull request #1701 from AxiomCodeAI/loop/ty-is838

cbc8ced

typescript: resolve calls through middleware parameters typed by an indexed access

Merge pull request #1705 from AxiomCodeAI/loop/ty-is422

19c28bc

typescript: try a merged interface's later declaration block first

Merge pull request #1702 from AxiomCodeAI/loop/ty-is784

f75f2ed

impact: follow a TypeScript structural type to the sites that never name it

Merge pull request #1681 from AxiomCodeAI/loop/ty-is1605

add194e

index: the live graph link is relative, so a copied checkout keeps its own graph

Merge pull request #1689 from AxiomCodeAI/loop/to-grep-shaped-output

405494d

mcp, cli: a list of sites comes one per line, path:line: code [tag]

swapnilpaliwal-sd requested review from JaredHLZhang, Whua689 and suyashpaliwal26 as code owners September 28, 2026 20:48

swapnilpaliwal-sd merged commit 18b83af into dev Sep 28, 2026
24 checks passed

swapnilpaliwal-sd deleted the 0.1.6 branch September 28, 2026 21:05

swapnilpaliwal-sd mentioned this pull request Sep 28, 2026

Merge dev (the 0.1.6 release) into 0.1.7 #1714

Merged

Navigation Menu

Repository navigation

Release 0.1.6 - #1712

Release 0.1.6#1712
swapnilpaliwal-sd merged 64 commits into
devfrom
0.1.6

swapnilpaliwal-sd commented Sep 28, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

swapnilpaliwal-sd commented Sep 28, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant