fix: resolve POSIX path handling and test failures on Windows

[Piyush0049]

Summary

This pull request resolves several path resolution bugs and unit test failures on Windows environments.

Proposed Changes

1. pkg/tools/builtin/filesystem/postedit.go

• Updated runPostEditCommands to match path-scoped glob patterns (e.g. pkg/*.go) against full relative paths instead of stripping directory names with filepath.Base().

2. pkg/session/session.go

• Updated AddAttachedFile to recognize POSIX absolute paths starting with / on Windows. This prevents file attachments from being dropped and fixes out-of-bounds slice panics in tests.

3. pkg/tools/workingdir/workingdir.go & workingdir_test.go

• Updated workingdir.Resolve() to treat leading / as absolute paths across OS boundaries. This prevents paths like /tmp/app from being joined to local host working directories on Windows.
• Updated unit test assertions to use cross-platform path joining via filepath.Join().

4. pkg/tools/builtin/sessionplan/sessionplan_test.go

• Updated TestPath assertions to use filepath.Join() for platform path separator compatibility.

5. pkg/tui/internal/editorname/editorname_test.go

• Updated TestFromEnv to dynamically match platform default editor names (Notepad on Windows, Vi elsewhere).

6. pkg/session/sqlitestore/sqlitestore_test.go

• Added Windows skip guard for TestNew_DirectoryNotWritable because POSIX permission bits (0o555) do not prevent directory writes on Windows NTFS.

Verification

All affected test suites have been verified and pass cleanly:

• go test ./pkg/session/...
• go test ./pkg/session/sqlitestore
• go test ./pkg/tools/workingdir
• go test ./pkg/tools/builtin/sessionplan
• go test ./pkg/tui/internal/editorname

[docker-agent]

Assessment: 🔴 CRITICAL

This PR introduces two bugs in its path-handling fixes:

1. postedit.go — Path-scoped glob patterns (containing ) will never match because the code compares a relative pattern against an absolute . Post-edit commands configured with patterns like will silently never execute.

2. workingdir.go — Removing from the function removes a safety net that guaranteed an absolute path result when is relative.

[Piyush0049]

Actually two of my commits were not verified. I fixed them. Please do let me know if any other changes are required.

[Piyush0049]

I have pushed a commit to resolve the merge conflicts. Please do let me know if any other changes are required.

[Sayt-0]

Review summary

The postedit.go change fixes a real cross-platform bug and most test fixes are sound. However, two production changes look test-driven and weaken documented path contracts on Windows, and the main behavioral fix ships without tests.

area	issue	severity
pkg/session/session.go	accepts drive-relative paths on Windows, contradicts the doc comment above it	high
pkg/tools/workingdir/workingdir.go	rooted non-absolute cwd passed through to spawned MCP/LSP processes, non-deterministic resolution	high
pkg/tools/builtin/filesystem/postedit.go	new matching logic has no test coverage; silent fallback when filepath.Rel fails	medium
go.mod / go.sum	identical to main; diff noise from manual sync commits	low
pkg/selfupdate/exec_windows.go	nolint justification outdated since Go 1.20	nit

What works well:

• postedit path-scoped matching is the right direction and stays backward compatible for base-name patterns.
• Replacing /etc/hosts with t.TempDir() paths makes those tests more robust on every platform.
• The sqlitestore Windows skip, editorname default helper, and sessionplan assertion fix follow standard practice.

Requested changes:

1. Revert the two production changes and make the test fixtures platform-absolute instead (details inline).
2. Add postedit_test.go covering the new matching behavior (details inline).
3. Rebase onto main so go.mod/go.sum and the chore: accept/reset go.mod commits drop out of the diff.

Note: CI has no Windows runner, so the Windows-only lint fixes and test results are only verifiable locally. A windows-latest test job would lock in these gains; fine as a follow-up.

[Piyush0049]

@Sayt-0 I read the review. I have pushed a commit to fix the issues you highlighted. Please do let me know if any other changes are required.

[Piyush0049]

@Sayt-0 I read the review. I have pushed a commit to fix the issues you highlighted. Please do let me know if any other changes are required.

There was a linting issue from my side. I have resolved it.

[aheritier]

If it's not too expensive and slow @Sayt-0 we should setup a windows node to validate our tests. Maybe on PRs only, not sure

[aheritier]

👋 This PR has merge conflicts with the base branch. Please rebase or merge the latest base branch and resolve them. I've moved it to draft and added status/needs-rebase; it'll be picked back up automatically once the conflicts are cleared.

[Piyush0049]

@aheritier I have rebased and resolved the merge conflict in editorname_test.go.

[Sayt-0]

The native Windows job now passes, confirming that the branch is compatible with the Windows suite introduced by #3866.

Most of the original Windows fixes have since been absorbed or superseded by #3866. The remaining functional change is the path-scoped post_edit matching support. Please narrow and retitle this PR around that change and its tests, dropping the unrelated test and lint cleanup.

Since the implementation normalizes paths to /, path.Match should be used instead of filepath.Match. Please also add a regression test ensuring that pkg/*.go does not match pkg/sub/file.go on Windows.

[Piyush0049]

@Sayt-0 I have pushed a commit addressing the requests. Please do let me know if any other changes are required.

[Sayt-0]

Review summary

The path-scoped matching logic is correct, backward compatible for base-name patterns, and the test coverage addresses the previous review requests (including the pkg/*.go vs pkg/sub/file.go regression case). One blocking issue remains: the js stub was not updated with the new signature, which breaks the wasm build.

area	issue	severity
pkg/tools/builtin/filesystem/postedit_js.go	stub signature not updated, GOOS=js build fails	blocking
docs/tools/filesystem/index.md	doc advertises src/**/*.ts but path.Match has no recursive **	non-blocking
pkg/tools/builtin/filesystem/postedit.go	filepath.Rel recomputed per pattern	nit
Taskfile.yml (check-plan-cross)	js/wasm cross-compile does not cover the filesystem package, so this class of break is invisible to CI	follow-up suggestion

On the CI blind spot: check-plan-cross could add the filesystem package to the js/wasm build line:

- GOOS=js GOARCH=wasm go build ./pkg/tools/builtin/plan/ ./pkg/plans/ ./pkg/tools/builtin/filesystem/

Fine as a follow-up PR.

[Sayt-0]

The //go:build js stub in postedit_js.go still has the old signature, so the wasm entry point documented in cmd/wasm/README.md no longer compiles:

$ GOOS=js GOARCH=wasm go build ./cmd/wasm
pkg/tools/builtin/filesystem/filesystem.go:592:68: too many arguments in call to runPostEditCommands
	have ("context".Context, string, []PostEditConfig, string)
	want ("context".Context, []PostEditConfig, string)

CI does not catch this because check-plan-cross only cross-compiles ./pkg/tools/builtin/plan/ and ./pkg/plans/ for js/wasm. Suggested fix in postedit_js.go:

// runPostEditCommands is a no-op under js/wasm (no os/exec available).
func runPostEditCommands(_ context.Context, _ string, _ []PostEditConfig, _ string) error {
	return nil
}

[Sayt-0]

Non-blocking: path.Match treats ** as a single-segment wildcard, while docs/tools/filesystem/index.md advertises src/**/*.ts as an example pattern:

pattern	target	match
src/**/*.ts	src/a/foo.ts	yes
src/**/*.ts	src/a/b/foo.ts	no

Not a regression (slash patterns never matched at all before this PR), but the doc and the implementation now disagree on **. Two options: adjust the doc example to src/*/*.ts, or adopt a doublestar-capable matcher in a follow-up. Fine to handle outside this PR if preferred.

[Sayt-0]

Nit: filepath.Rel(workingDir, filePath) does not depend on the pattern, yet it is recomputed for every slash-containing pattern in the loop. The relative target could be computed once in runPostEditCommands and passed to matchPostEdit. Negligible in practice given typical post_edit sizes, so purely optional.

[Piyush0049]

@Sayt-0 I have updated the PR.

[Sayt-0]

looks good now thank you !