fix(cli): tolerate undecodable alias lookup output

[qdivan]

What does this PR do?

Fixes the current Windows/profile-alias decoding failure in #47939 without changing process-global subprocess behavior.

check_alias_collision() invokes where/which with text=True. If that command emits bytes that the platform-selected text codec cannot decode, subprocess.run() can raise UnicodeDecodeError before Hermes can handle the lookup result.

This patch adds errors="replace" only at that call site. It deliberately preserves Python's platform-default codec for localized Windows command output and avoids the previous import-time subprocess.Popen.__init__ monkeypatch.

Related Issue

Fixes #47939

Type of Change

• 🐛 Bug fix
• ✅ Tests
• ✨ New feature
• 🔒 Security fix
• 📝 Documentation update
• ♻️ Refactor

Changes Made

• hermes_cli/profiles.py: tolerate undecodable bytes from the alias lookup subprocess with a localized errors="replace" policy.
• tests/hermes_cli/test_profiles.py: exercise the real check_alias_collision() path with a subprocess that emits undecodable stderr.

Validation

• Baseline RED on current origin/main: the targeted regression failed in check_alias_collision() with UnicodeDecodeError.
• Updated branch GREEN: tests/hermes_cli/test_profiles.py — 156 passed.
• Changed-file Ruff — passed.
• git diff --check — passed.

Duplicate / overlap check

• fix: add explicit UTF-8 encoding to all subprocess text=True calls (#53428) #55339 touches the production call as part of a 76-file encoding sweep, but has no test files or behavioral regression for check_alias_collision().
• fix(environments): pass encoding='utf-8' errors='replace' to all text-mode Popen calls #47951, fix: add explicit encoding="utf-8" to subprocess.run calls with text=True #52249, and fix(cli): add windows_subprocess_kwargs() to force UTF-8 in subprocess text output #25200 do not modify this call path.
• No recently merged equivalent fix was found on origin/main.

Checklist

• Read the contributing and repository guidance.
• Conventional commit message.
• Searched current main, open PRs, and recently merged PRs for equivalent work.
• Contains only the narrow fix and regression test.
• Added behavior-focused regression coverage.
• Considered Windows/localized subprocess behavior.
• No config, dependency, generated-file, or documentation changes are required.

Screenshots / Logs

Not applicable; this is a CLI subprocess-decoding path with no UI changes.

[tonydwb]

Code Review Summary

Verdict: Approved — hardened subprocess text decoding.

• 2 files changed (+59/-0)
• Hardens text subprocess decoding for edge cases
• No security concerns, no debug artifacts

---

Reviewed by Hermes Agent

[teknium1]

Thanks for targeting a real Windows decoding failure. The current-head reproduction path is still present at hermes_cli/profiles.py:407-410, where where/which is run with text=True and no error policy.

Problems

• The proposed hermes_cli/__init__.py import-time replacement of subprocess.Popen.__init__ changes every later text subprocess in the process, including callers outside the reported profile-alias path. Current CLI code uses explicit policies at individual call sites, for example hermes_cli/tools_config.py:1069-1072.
• tests/hermes_cli/test_ensure_utf8_locale.py only verifies injected kwargs through a mocked initializer. It does not cover check_alias_collision() or a real text-mode subprocess pipe.

Suggested changes

• Apply the narrow decoding policy at hermes_cli/profiles.py:407-410 and add regression coverage for that call path.
• Keep explicit caller overrides intact, but avoid a permanent process-global Popen monkeypatch.

This is an automated hermes-sweeper review.

[qdivan]

Updated in response to the latest hermes-sweeper review.

Changes:

• removed the import-time global subprocess.Popen.__init__ monkeypatch entirely;
• applied a localized errors="replace" policy only to check_alias_collision()'s where/which subprocess call, preserving the platform-default codec;
• replaced the kwargs-only shim test with behavioral coverage of the real check_alias_collision() path using a subprocess that emits undecodable stderr.

Validation:

• baseline RED on current origin/main: targeted regression failed with UnicodeDecodeError;
• updated branch GREEN: 156 passed in tests/hermes_cli/test_profiles.py;
• changed-file Ruff passed;
• git diff --check passed.

I also refreshed duplicate/fix-existence checks. #55339 overlaps this production call as part of a 76-file sweep, but has no test files or check_alias_collision() regression; the other active subprocess PRs checked do not touch this path.

[qdivan]

Closing this because current main now contains the equivalent fix via c89481db5ec2400e936de6ac402c89e7a39cfe19 (fix: add explicit UTF-8 encoding to all subprocess text=True calls (#53428)). The live check_alias_collision() call now uses encoding="utf-8", errors="replace", so this older conflicting branch is superseded. Thanks to the reviewers who helped narrow the original patch.