[Bug]: wsl-attachment-paths.test.ts fails on native Windows since #296 (test portability) + remaining Windows capability baseline

[Ralle1976]

Product or interface

Source build or repository tooling

Version

Source @ ae65651 (0.5.1). Node.js v24.18.0, pnpm 9.12.0.

Platform

Windows

OS version and architecture

Windows 11 x64 (build 26200). No Developer Mode (no unprivileged symlink privilege). Windows PowerShell 5.1 default; Git Bash present.

Issue area

Tests / verification gates

Steps to reproduce

1. Build from source on Windows: npx pnpm@9.12.0 install --frozen-lockfile && pnpm build.
2. pnpm test:capabilities

Expected and actual behavior

Since ae65651 (#296), packages/tui/test/unit/wsl-attachment-paths.test.ts fails 12 of 25 cases on native Windows, every one via:

Caused by: Error: wslpath did not return an absolute Linux path.
  ❯ resolveWslPath packages/tui/src/host/wsl-path.ts:37:13

Root cause: the suite's beforeEach feeds the host temp path back as mocked wslpath output (host.executeFile.mockResolvedValue({ stdout: ${imagePath}\n }) with imagePath = join(tmpdir(), …)). On POSIX CI that path is /tmp/…, which satisfies the production posix.isAbsolute(mapped) check. On native Windows it is C:\Users\…, which the production validation correctly rejects — the fixture path cannot double as a Linux absolute path on Windows. There is no product defect; the tests are not portable.

Prepared fix (fork branch, per the collaborators-only PR policy)

test/windows-wsl-path-portability (single commit) in https://github.com/Ralle1976/minimax-code/tree/test/windows-wsl-path-portability

describe.skipIf(process.platform === "win32") with a comment explaining the fixture constraint — the same platform-gating precedent already used by config-file-permissions.test.ts, session-diff-fallback.test.ts, update-application.test.ts and others.

Verified: the suite is skipped on Windows (12 failures gone, file still registered in test/vitest-suites.json); full pnpm test:capabilities improves 26 → 13 failures with no new failures. POSIX runs are unchanged by construction (skipIf is inert there) but were not executed locally.

Remaining Windows capability-gate baseline (context for #238)

After both fixes, pnpm test:capabilities on this machine: 4213 pass / 13 fail / 80 skip. The remaining 13:

• 11 × symlink EPERM (canonical-agent-config 6, agent.repository 3, mcp/project-config 1, headless-invocation 1): fs.symlink on Windows needs Developer Mode/admin. As a CI-reliability option, detecting the missing privilege and reporting an explicit environment SKIP would separate "cannot run here" from "regression".
• 2 × observability.test.ts: logging assertion mismatches; not yet root-caused.

Related gates: test:policy fails ~15 cases in local-permission-facade.test.ts because the POSIX rm → mavis-trash rewrite expectations do not hold on Windows (deny instead of allow+rewrite); test:sandbox fails 1 case in srt-macos.test.ts because that darwin suite runs unfiltered on Windows and asserts /tmp/... while the host renders \tmp\.... typecheck, check:source, check:tsconfig, build, test:smoke, test:byok, test:status-contract all pass.

Earlier community baselines (#249, #237) reported "11 locale + 12 symlink EPERM"; the zh-CN locale failures no longer reproduce on this machine.

Before submitting

• I have searched existing issues.
• I have included my version and removed sensitive information.

[Ralle1976]

Follow-up on the remaining baseline from this issue, after e7d809d (#303) — everything below is prepared on test/windows-observability-followup (two single-purpose commits on top of current main):

1. The two observability.test.ts failures are test portability, not product. resolveTuiObservabilityDirectory joins with the host separator, so the POSIX-literal assertions (expected '\data\v2\observability\mcode' to be '/data/v2/observability/mcode') fail on Windows. Fix: normalize the received values with the existing replaceAll('\', '/') idiom and accept either separator in the flush-path matcher; expected values stay readable literals and POSIX behavior is unchanged.

2. One WSL case from #296 was left ungated by the skipIf port in #303. keeps native attachment resolution working in WSL without a subprocess builds its "native" path from the host temp dir; on Windows that resolves to a drive-letter path and enters the conversion path (wslpath did not return an absolute Linux path). Gate it with the file's existing skipWslConversion constant, consistent with the other gated groups.

Updated Windows capability baseline (main @ e7d809d + the two fixes above, Windows 11 x64, Node 24.18.0, pnpm 9.12.0): 4249 pass / 11 fail / 60 skip. All 11 remaining failures are the symlink-privilege EPERM cases in the four known files (canonical-agent-config 6, agent.repository 3, mcp/project-config 1, headless-invocation 1) — environmental on a host without Developer Mode, not code defects. With the new windows-latest profile from #303 plus these two ports, the non-environmental Windows capability failures are down to zero on this machine.

Checks run: focused vitest for both files (all green/skipped), full pnpm test:capabilities (26 → 12 → 11 across the three rounds). Not run: POSIX hosts, full pnpm verify.

[Ralle1976]

Follow-up status on the prepared branch test/windows-observability-followup (rebased onto 44b13d3 today, 2026-09-22; force-pushed, same two commits, new SHAs):

Re-verification after rebase (Windows 11 x64, Node 24.18.0):

• wsl-attachment-paths.test.ts + observability.test.ts: all green / platform-gated (25 passed, 12 skipped) — the observability fix and the one WSL case fix(windows): clarify source checkout requirements #303's skipIf port left ungated (keeps native attachment resolution working in WSL without a subprocess) both still apply cleanly on current main.
• Capability-gate failures attributable to this follow-up: 3 → 0 (wsl 1, observability 2). The remaining 11 on my host are the documented symlink-EPERM environment cases — now covered with root causes and prepared fixes in [Bug]: Windows local verification: 30 test failures + 1 gate failure in 4 root causes — fixes on 4 prepared branches (+3 findings) #322 ([Bug]: Windows local verification: 30 test failures + 1 gate failure in 4 root causes — fixes on 4 prepared branches (+3 findings) #322), which also folds in three further Windows test failures and two findings (among them: the fix(windows): clarify source checkout requirements #303 NTFS preflight false-negatives on non-elevated shells, which is what test:windows hits on my machine).

No urgency intended — noting the counts and the new base per review hygiene so the numbers above are reproducible against a named SHA.