test(windows): set_home helper + sweep ~133 HOME monkeypatch sites (refs #644) by memtomem · Pull Request #714 · memtomem/memtomem

memtomem · 2026-05-02T06:07:40Z

Summary

Path.home() on Windows reads USERPROFILE first (then HOMEDRIVE+HOMEPATH), so monkeypatch.setenv("HOME", str(tmp_path)) is silently ignored on the runner and tests end up touching the real C:\Users\runneradmin\ instead of tmp_path. Per #644 this was the largest single class of Windows CI failures.

This PR introduces a small set_home(monkeypatch, path) helper in tests/helpers.py that sets both HOME and USERPROFILE (no-op on POSIX, correct on Windows), then sweeps all ~133 HOME monkeypatch sites in packages/memtomem/tests/ to use it. Manual HOME + USERPROFILE pairs that already existed in 5 files (test_context_settings.py, test_web_routes_extended.py, test_web_routes.py, test_context_agents.py, test_server_tools_context_settings_gate.py) are collapsed into single set_home(...) calls — exactly the redundancy that motivated centralizing the helper.

Mechanical, test-only change. No production code touched. ruff check + ruff format --check clean.

What's covered

88 sites in test_init_cmd.py
25 sites in test_web_routes.py (24 HOME-only + 1 manual pair)
4 sites across test_web_routes_extended.py (2 manual pairs collapsed)
4 sites across test_context_settings.py (2 manual pairs collapsed)
Manual pairs in test_context_agents.py and test_server_tools_context_settings_gate.py collapsed
11 other test files with single sites (test_cli_index_noop_e2e.py, test_fastembed_cache.py, test_indexing_engine.py, test_lazy_init_acceptance.py, test_runtime_paths.py, test_uninstall_cmd.py, test_web_exclude_guard.py, test_web_hot_reload.py, test_web_routes_context_locks.py, test_web_routes_context_mutators.py, test_web_routes_context_projects.py)

18 files modified, +149 / -149 (1:1 mechanical substitutions). The full sweep makes the helper the only HOME-redirection pattern in the test suite — no two-pattern drift to manage.

Cross-check (BLOCKING per plan)

Source: gh run view 25245215735 --log-failed --job=74028225027 (latest main run after PR #711 / #713 merged, 2026-05-02).

Originally-targeted "auto-discovery" tests:

Test	Was failing on Windows?	Status after this PR
`test_config_overrides.py::test_detect_provider_dirs_excludes_gemini`	❌ (false-positive pass)	✅ now exercises invariant correctly
`test_config_overrides.py::test_detect_provider_dirs_filters_empty_claude_memory`	✅	⚠️ still failing — set_home now redirects discovery, but `_bucket()` calls `categorize_memory_dir()` whose forward-slash regex (#316) classifies the backslash path as `"user"` so the claude-memory bucket stays empty
`test_config_overrides.py::test_detect_provider_dirs_finds_claude_plans_and_codex`	✅	⚠️ still failing — same #316 root cause
`test_init_cmd.py::TestProviderDirsStep::test_write_merges_provider_dirs_into_memory_dirs`	❌ (inert HOME patch)	✅ migrated for correctness
`test_init_cmd.py::TestIncludeProviderFlag::test_flag_resolves_categories_to_dirs`	✅	✅ fixed
`test_init_cmd.py::TestIncludeProviderFlag::test_flag_silently_skips_unavailable_categories`	✅	✅ fixed
`test_init_cmd.py::TestIncludeProviderFlag::test_no_flag_writes_no_provider_dirs`	✅	✅ fixed

The 2 still-failing tests need both this PR (USERPROFILE redirect so _detect_provider_dirs looks at tmp_path instead of the runner home) and the #316 production fix (so categorize_memory_dir accepts backslash paths). After #316 lands, these 2 tests should immediately go green on Windows with no further test changes — the migration this PR ships is necessary, just not sufficient on its own.

The full sweep brings ~30 additional HOME-related Windows failures into scope (notably the test_init_cmd.py and test_web_routes.py clusters from #644's symptom report).

Windows CI impact

Metric	Before (run 25245215735)	After sweep (run 25245680115)
Total Windows failures	105	44
Reduction		−61
Auto-discovery failures still pending #316		2

macOS, Ubuntu, lint, typecheck, golden-path, notebooks all green on the sweep run — negative pin held, no POSIX regression. Helper is a no-op on POSIX (stdlib Path.home() ignores USERPROFILE).

Out of scope (explicit follow-ups)

PR 6: production fix for categorize_memory_dir's forward-slash regex (Windows: wizard fails to auto-detect Claude/Codex memory dirs (backslash path regex) #316). Production change, separate concern. Closes the 2 still-failing test_detect_provider_dirs_* cases.
PR 7+: re-triage Windows fail residual after Windows: wizard fails to auto-detect Claude/Codex memory dirs (backslash path regex) #316 lands. The original splashing-dragon roadmap's PR 4 (symlink/git status) and PR 5+ (multi-factor) follow.
Final PR: flip .github/workflows/ci.yml Windows job from continue-on-error: true → false once Windows fails are deterministic-skip-only.

#644 can close on this PR's merge — the helper sweep delivers its full scope (HOME redirect across the whole suite). The 2 lingering _detect_provider_dirs_* tests stay open under #316.

Test plan

uv run pytest packages/memtomem/tests/ -m "not ollama" -q — full suite green locally on macOS (3951 passed)
uv run ruff check packages/memtomem/tests + ruff format --check packages/memtomem/tests — clean across all 158 files
Mypy advisory: no new errors (the 2 pre-existing errors in cli/agent_cmd.py and storage/mixins/schedules.py are unrelated)
CI Windows job (informational): 105 → 44 failures (−61); macOS / Ubuntu green with no pass-count delta

🤖 Generated with Claude Code

`Path.home()` on Windows reads `USERPROFILE` first (then `HOMEDRIVE`+`HOMEPATH`), so `monkeypatch.setenv("HOME", str(tmp_path))` is silently ignored and tests read/write the real runner home. This manifests as `~/.claude/projects` auto-discovery returning the runner's real home contents (see #644) and `~/.memtomem/config.json` writes landing outside `tmp_path`. Introduce a small `set_home(monkeypatch, path)` helper in `tests/helpers.py` that sets both `HOME` and `USERPROFILE` (harmless no-op on POSIX, correct on Windows). Apply to the 7 auto-discovery callsites that exercise `Path.home()` indirectly: - `test_config_overrides.py::test_detect_provider_dirs_excludes_gemini` - `test_config_overrides.py::test_detect_provider_dirs_filters_empty_claude_memory` - `test_config_overrides.py::test_detect_provider_dirs_finds_claude_plans_and_codex` - `test_init_cmd.py::TestProviderDirsStep::test_write_merges_provider_dirs_into_memory_dirs` - `test_init_cmd.py::TestIncludeProviderFlag::test_flag_resolves_categories_to_dirs` - `test_init_cmd.py::TestIncludeProviderFlag::test_flag_silently_skips_unavailable_categories` - `test_init_cmd.py::TestIncludeProviderFlag::test_no_flag_writes_no_provider_dirs` Of the 7, 4 currently fail on the informational Windows job (filters_empty_claude_memory, finds_claude_plans_and_codex, and the 3 `TestIncludeProviderFlag` tests). The other 3 are migrated for correctness — `excludes_gemini` was a false-positive pass (assertion held trivially when `_detect_provider_dirs()` returned empty due to USERPROFILE redirect), and `write_merges_provider_dirs` had a HOME monkeypatch that was inert but uses the same auto-discovery shape. Test-only change. The 130+ other `monkeypatch.setenv("HOME", ...)` sites across 18 test files (notably 91 in `test_init_cmd.py` and 25 in `test_web_routes.py`) will be migrated in follow-up PRs to keep this change axis single (one PR, one change). Refs #644. Part of the Windows CI green journey (PR 1 #711, PR 2 #713). Co-Authored-By: Claude <[email protected]>

…me docstring Per PR #714 review, surface the 5 manual "set both" call sites that predate this helper so the next contributor sweeping the residual ~130 ``monkeypatch.setenv("HOME", ...)`` sites doesn't re-introduce the manual pattern. Refs #644. Co-Authored-By: Claude <[email protected]>

…t_home helper Mechanical migration of all remaining HOME-only monkeypatch sites in packages/memtomem/tests/ to the set_home helper introduced in 9825210. On Windows, Path.home() reads USERPROFILE first; bare HOME setenv was silently ignored, leaking real user-home reads in fresh runners. Sweep covers: - 88 sites in test_init_cmd.py (HOME-only) - 25 sites in test_web_routes.py (24 HOME-only + 1 manual HOME+USERPROFILE pair) - 4 sites across test_web_routes_extended.py (2 manual pairs) - 4 sites across test_context_settings.py (2 manual pairs) - Manual HOME+USERPROFILE pairs in test_context_agents.py and test_server_tools_context_settings_gate.py collapsed into one set_home call - Single-site replacements in 11 other test files Manual pairs (HOME setenv followed by USERPROFILE setenv with the same value) are collapsed into a single set_home(monkeypatch, path) call, removing the redundancy that motivated the helper. Helper docstring is trimmed: the manual call-site list is no longer needed now that the sweep is complete. No production code changes. ruff check + format clean. Full impacted test set (856 tests across the 17 files) passes locally on macOS. Co-Authored-By: Claude <[email protected]>

pandas-studio and others added 3 commits May 2, 2026 15:06

memtomem changed the title ~~test(windows): set_home helper for HOME→USERPROFILE Windows fix (refs #644, PR 3/N)~~ test(windows): set_home helper + sweep ~133 HOME monkeypatch sites (refs #644) May 2, 2026

memtomem merged commit 5ca3cd3 into main May 2, 2026
8 of 9 checks passed

github-actions Bot locked and limited conversation to collaborators May 2, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

test(windows): set_home helper + sweep ~133 HOME monkeypatch sites (refs #644)#714

test(windows): set_home helper + sweep ~133 HOME monkeypatch sites (refs #644)#714
memtomem merged 3 commits intomainfrom
fix/windows-ci-home-monkeypatch-helper

memtomem commented May 2, 2026 •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

memtomem commented May 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

What's covered

Cross-check (BLOCKING per plan)

Windows CI impact

Out of scope (explicit follow-ups)

Test plan

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

memtomem commented May 2, 2026 •

edited

Loading