emrg: merge stop-git.ps1 into stop-emrg.cmd as single file (rant 2026-08-12T14:00:05)

[argszero]

Merge stop-git.ps1 into stop-emrg.cmd as a single file (rant 2026-08-12T14:00:05)

Background

Host decision 2026-08-12 (B plan: merge first, release v0.2.28 later). PR #701
delivered CRLF enforcement + truthful stop-git failure but kept the separate
bin/stop-git.ps1 (make-installer still references it, stop-emrg.cmd step 4
still calls it). The host explicitly asked for a single file.

Changes

1. bin/stop-git.ps1 deleted — the guilt-by-association kill logic now lives
   inline in bin/stop-emrg.cmd step 4 as a single powershell -Command
   invocation:
   • prefix-scoped: $env:USERPROFILE\.emrg\install\git\* only (system Git in
     Program Files never matched)
   • pass 1: force-kill the git tree (git/ssh/plink/bash)
   • pass 2: guilt-by-association — force-kill every survivor under the prefix
     (sh/vim included, host 2026-08-11T19:47:44 decision)
   • each pass re-queries Get-CimInstance (no stale snapshot); the survivor
     check uses the latest snapshot; residual → exit 1 (truthful failure, emrg: enforce CRLF for Windows .cmd/.ps1 + truthful stop-git failure (rant 2026-08-12T12:30:41) #701)
   • quoting: uses the same \" escaping already proven on real Windows in the
     TUI-fallback and :verify inline snippets (shipped since v0.2.25-v0.2.27),
     avoiding both the 0.2.26 quoting-escape bug and temp-file generation
2. packaging/make-installer.sh: removed the [Files] dontcopy entry for
   stop-git.ps1 and the GitStopScript/ExtractTemporaryFile('stop-git.ps1')
   plumbing in PrepareToInstall — now only extracts and runs stop-emrg.cmd
3. packaging/build-runtime.sh: removed the stop-git.ps1 copy line
4. tests/test_installer_stop.py: rewritten to pin the single-file wiring
   (inline -Command + \" escapes, no -File/GITSTOP remnants, 3 fresh
   Get-CimInstance enumerations, truthful exit 1); added a guard that
   bin/stop-git.ps1 does not exist and packaging has no functional references

Acceptance (from rant)

• bin/ has no stop-git.ps1; grep stop-git matches history comments only ✅
• make-installer.sh / build-runtime.sh have no stop-git functional references ✅
• stop-emrg.cmd single file: step 1-3 (GUI/TUI/daemon) + step 4 (inline git kill)
  • verify, full flow ✅ (text-wired by tests; real Windows run by host)

Verification

• pytest: 730 passed (count unchanged — one test replaced, one added)
• Import + CLI sanity OK
• tests/test_cmd_crlf.py green (CRLF preserved: 105/105)
• No version bump — release v0.2.28 to follow after host confirms (per rant)

[argszero]

✅ LGTM — cycle

Reviewed the full diff on feature/stop-git-merge-single-file:

1. Single-file merge is faithful: bin/stop-git.ps1 deleted and its logic inlined into bin/stop-emrg.cmd step 4 as one powershell -NoProfile -ExecutionPolicy Bypass -Command invocation. Semantics identical to the proven standalone script:
   • prefix-scoped to $env:USERPROFILE\.emrg\install\git\* only (system Git never matched)
   • pass 1 kills the git tree (git/ssh/plink/bash), pass 2 guilt-by-association kills every survivor under the prefix
   • each pass re-queries Get-CimInstance (3 fresh enumerations, no stale snapshot); the survivor check uses the latest query
   • exit 1 names survivors (truthful failure per #701), exit 0 only when nothing survives
2. Escaping: the \" pattern matches the :verify inline snippet already proven on real Windows since v0.2.25–v0.2.27 — avoids the 0.2.26 quoting-escape bug; single-quoted PS literals and @(...) array-wrap are correct.
3. Packaging: [Files] dontcopy entry + ExtractTemporaryFile('stop-git.ps1') + GitStopScript/%~1 plumbing all removed from make-installer.sh; build-runtime.sh copy line removed. No functional stop-git references remain (grep matches history comments only).
4. Tests: rewritten to pin the single-file wiring (no -File/GITSTOP remnants, 3 enumerations, truthful exit 1) + guards that bin/stop-git.ps1 does not exist and packaging has no functional references. tests/test_installer_stop.py + tests/test_cmd_crlf.py: 6/6 passed.
5. Full suite: uv run pytest tests/ → 730 passed in 22.9s (matches PR claim; CRLF 105/105 preserved).
6. No CI/workflow changes; no version bump (correct — v0.2.28 release to follow per the rant).

[argszero]

✅ LGTM — cycle

Reviewed the full diff of #702 (single-file merge of stop-git.ps1 into stop-emrg.cmd):

1. Faithful merge: stop-git.ps1 deleted; its guilt-by-association logic inlined as one powershell -ExecutionPolicy Bypass -Command in step 4 — prefix-scoped to %USERPROFILE%\.emrg\install\git\* (system Git never matched), pass 1 kills git tree (git/ssh/plink/bash), pass 2 kills every survivor (sh/vim included per host 2026-08-11T19:47:44).
2. Fresh enumeration: each pass + the survivor check re-query Get-CimInstance (3 times) — no stale snapshot; residual → Write-Host names + exit 1 (truthful failure per #701).
3. Escaping: \" matches the proven TUI/verify inline snippets shipped since v0.2.25 — avoids the 0.2.26 quoting bug and temp-file generation.
4. Packaging: no dontcopy/GitStopScript remnants in make-installer.sh or build-runtime.sh; tests pin the wiring (incl. bin/stop-git.ps1 absent + CRLF 105/105).
5. CI: test run 31568952199 PASS. Local pytest 730 green (net-zero test delta). No version bump — release v0.2.28 deferred until host confirms per rant.

[argszero]

✅ LGTM — cycle

3rd consecutive ✅ (R1057 + parallel + this cycle, no ❌ in between). Head 9721194 unchanged since my earlier audit — full diff re-verified: single-file merge faithful to the deleted stop-git.ps1 (prefix-scoped guilt-by-association, 3 fresh Get-CimInstance queries, truthful exit 1), \" escaping matches the proven :verify pattern, packaging references removed, CRLF 105/105, full suite 730 passed. Merge condition satisfied.