Repository navigation
Windows: the query cases answer as on Linux, and the release gate runs them (0.2.0) - #1949
Merged
swapnilpaliwal-sd merged 3 commits intoOct 14, 2026
Merged
Conversation
…s them 0.1.10 passed its release e2e on Windows while 43 of the query cases' 1,844 checks failed there: the e2e is one case per language. Every cause is how Windows writes paths and command lines, or how it holds files; none is platform-specific code. Paths: repository-relative paths are written with `/`, as the graph stores them. - parser: the Python and C# analyzers recorded `pkg\changed.py` in the IR, so rules joining a module's directory to a string (importlib's SourceFileLoader, Django's include()) matched nothing, and C# derived test attributes were missed. - query scripts: os.path joined and normalised graph paths with `\` (runner setupFiles/globalSetup, spawned scripts, the JS/TS import resolver, the freshness snapshot), so tests reached through a file were not linked. posixpath now, the idiom axiomcode-test-impact already used. - path/impact: an absolute file:line under the repository was refused as outside it (a `root + '/'` prefix test); commonpath on normcased paths. Arguments: a native program cannot hand Git Bash an argument holding ' or ": msys parses the command line by its own rules, so `context "the user's cart"` lost every argument after the apostrophe and a quoted string arrived as \Key\. On Windows the CLI, the MCP server and every build Python starts pass them in AXIOMCODE_ARGV (ax_exec.bash_argv); the bash entry points turn them back into "$@". Elsewhere the command line is used as before. Build lock: on Windows it locked nothing (Git Bash's fd 9 is not a descriptor in a native python), so a refresh and a second build of one repository could run at once. A helper (ax_fresh.py hold) now holds a real lock for the build's lifetime; building() sees it and a second build waits. POSIX keeps flock. Tests and the gate: - tests/run.py starts the CLI as the product does (Git Bash found by find-bash.js, MSYS=noglob, arguments via bash_argv), reads and prints UTF-8, uses the sqlite3 module, and waits for a case's background refresh before removing its graph. - tests/windows_argv.py: the argument hand-over through both entry points, with the command line as control. - no_symlink / no_exec start Git Bash and UTF-8 Python as the product does. - the release e2e on win32-x64 now runs every query case against the installed packages, and the Windows-only tests (.github/scripts/e2e-cases.sh). Verified: Windows VM, every language's query cases (1,844 checks), the Windows-only tests and the gate script; repeated asserted-links runs 25/25. Linux: the five engine suites, the query cases and the standalone tests identical to dev (51 of 51 results). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
swapnilpaliwal-sd
requested review from
JaredHLZhang,
Whua689 and
suyashpaliwal26
as code owners
October 14, 2026 06:23
relpath raises across drives on Windows, so a --library whose compiled IR sits in the user cache on C: while the repository is on D: (as on GitHub's runner) failed the whole index. axiomcode-index and axiomcode-graph read it as outside the repository, as a `..` path is; a link or a changed file given absolutely on another drive is kept as given. no_exec's control starts Git Bash, not a bare `bash` (WSL on GitHub's image). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ightly runs every case Every query case on Windows took 37 of the e2e job's 50 minutes, against 2-6 for the smoke test on every other platform. A release now runs the 39 case names that have failed on Windows (.github/e2e/windows-cases.txt: 63 of 356 case directories, every language), about 7 minutes; the nightly (fresh) still runs every case. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
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.
Why
0.1.10 passed its release e2e on Windows, but that gate indexes one case per language. Run against the published
packages, the query cases fail 43 of 1,844 checks on Windows (all pass on Linux). Every cause is how Windows
writes paths and command lines, or how it holds files. None of it is platform-specific code.
What changes
--tests/test-impactmiss tests that reach the change through a file (spawned script, runnersetupFiles/globalSetup, an import chain)os.path→\posixpathfor repository-relative paths (the idiomaxiomcode-test-impactalready uses)include()routes unresolved; C# derived test attributes missedpkg\changed.pyinto the IR/file:lineunder the repo is "outside the indexed repository"root + '/'prefix testcommonpathon normcased paths'or"is mangled (context "the user's cart"loses every argument after the apostrophe;"Key"arrives as\Key\), CLI and MCP alikeAXIOMCODE_ARGV(ax_exec.bash_argv); the bash entry points rebuild"$@".progress/base-commit: No such file)ax_fresh.py holdholds a real lock for the build's lifetime;building()sees itPOSIX behaviour is unchanged by construction:
path.sepis already/,AXIOMCODE_ARGVis used only on Windows,and the lock keeps
flockoutside Git Bash.The gate
Windows-only tests (
.github/scripts/e2e-cases.sh). This is the check that would have caught 0.1.10.tests/run.pystarts the CLI as the product does (Git Bash found byfind-bash.js,MSYS=noglob, arguments viabash_argv), reads and prints UTF-8, and waits for a case's background refresh before removing its graph.tests/windows_argv.py: the argument hand-over through both entry points, with the command line as the control.It fails 6 of 10 checks against
dev's entry scripts.Port
The same change as #1948 (0.1.11), cherry-picked onto
apps/integration-0.2.0, without the 0.1.11 version commit:this branch is already at the untagged 0.2.0, and the version gate passes. One conflict, in
ax_fresh.py: 0.2.0 renamedwatched()towatched_from()on the lines that gained.replace(os.sep, '/'); resolved by keeping both. The web layer0.2.0 adds already writes its paths with
/.On this branch:
windows_argv10/10; Python cases 445/445 in three runs, as on 0.2.0 itself. The results below are#1948's; the Windows VM pass on this branch follows in a comment.
Verified
Windows (Windows Server 2022 VM; the published
engine-win32-x64@0.1.10engines; Soufflé hidden):mixed_separators/no_symlink/no_exec/windows_argvasserted-links*, 5 rounds × 5 languages back to backe2e-cases.shas the release job runs itLinux (
devand this branch, same VM, CI's flags): the five engine suites (--oraclewhere CI uses it), the querycases and the standalone tests give identical results on all 51 (exit codes and pass/fail counts).
macOS: the query cases for every language pass, as on
dev; parsercsharp-tests,python-tests,discovery-testsandgradle-testspass.