Stryker goes: 2,400 lines, two packages, a patch and an override fewer

.github/dependabot.yml

@@ -57,13 +57,6 @@ updates:
       shipped-minor-and-patch:
         patterns: ['fuse.js', 'mediabunny', 'sortablejs']
         update-types: ['minor', 'patch']
-      # Stryker's packages come out together, and each requires the others at its own
-      # version exactly (@stryker-mutator/vitest-runner's peerDependencies): one pull
-      # request for all of them, majors included. Before tooling-minor-and-patch, which
-      # would take them otherwise. While the vitest runner carries a patch, both are
-      # ignored below and move together through the Patched libraries workflow.
-      stryker:
-        patterns: ['@stryker-mutator/*']
       # WebdriverIO's packages come out together and only work at the same version: one pull
       # request for all of them, majors included (10.0.0 came as four, each red alone,
       # 2026-10-10). Before tooling-minor-and-patch, which would take them otherwise.
@@ -79,14 +72,10 @@ updates:
     # from the Patched libraries workflow instead, which re-cuts the patch with
     # tools/recut-patch.mjs and opens the pull request itself when that is clean
     # (docs/updating-patched-libraries.md). Keep this list in step with
-    # patchedDependencies, and the packages that move with a patched one (LOCKSTEP in
-    # tools/check-patched-updates.mjs: @stryker-mutator/core with the vitest runner, which
-    # requires its exact version). When every change of a patch has landed upstream,
-    # tools/recut-patch.mjs drops the patch and takes the library and those packages off
-    # this list in the same pull request.
+    # patchedDependencies. When every change of a patch has landed upstream,
+    # tools/recut-patch.mjs drops the patch and takes the library off this list in the same
+    # pull request.
     ignore:
-      - dependency-name: '@stryker-mutator/core'
-      - dependency-name: '@stryker-mutator/vitest-runner'
       - dependency-name: 'Coloris'
       - dependency-name: 'dashjs'
       - dependency-name: 'gif.js'

.github/workflows/firefox-beta.yml

@@ -131,8 +131,8 @@ jobs:
             fi
             exit 0
           fi
-          # A suite that runs out of its time ends cancelled, not failed (as in
-          # mutation-tests.yml): a Firefox that hangs the suites said nothing.
+          # A suite that runs out of its time ends cancelled, not failed: a Firefox that
+          # hangs the suites said nothing.
           if [ "$RESULT" != failure ] && [ "$RESULT" != cancelled ]; then
             echo "e2e result: $RESULT - nothing to report."
             exit 0

.github/workflows/firefox-stable.yml

@@ -192,8 +192,8 @@ jobs:
             done
             exit 0
           fi
-          # A suite that runs out of its time ends cancelled, not failed (as in
-          # mutation-tests.yml): a Firefox that hangs the suites said nothing.
+          # A suite that runs out of its time ends cancelled, not failed: a Firefox that
+          # hangs the suites said nothing.
           if [ "$RESULT" != failure ] && [ "$RESULT" != cancelled ]; then
             echo "e2e result: $RESULT - nothing to report."
             exit 0

.github/workflows/runner-images.yml

@@ -226,8 +226,8 @@ jobs:
           }
           report() { # result, images, what the image is to -latest
             local result=$1 images=$2 role=$3 title number
-            # A job that runs out of its time ends cancelled, not failed (as in
-            # mutation-tests.yml): an image that hangs CI said nothing.
+            # A job that runs out of its time ends cancelled, not failed: an image that
+            # hangs CI said nothing.
             if [ "$result" != success ] && [ "$result" != failure ] && [ "$result" != cancelled ]; then
               echo "$images: ${result:-not run} - nothing to report."
               return

.gitignore

@@ -40,9 +40,5 @@ lint-*.json
 # vitest --coverage output.
 coverage/
 
-# Stryker (pnpm run test:mutation): its reports and its sandboxes
-reports/mutation/
-.stryker-tmp/
-
 # The helper's launch lock (faststream-mpv-host.mjs launchLockFile), next to it while mpv starts.
 native-host/faststream-mpv-launch-*.lock

docs/maintenance.md

@@ -207,12 +207,6 @@ third attempt was cancelled too or GitHub refused the re-run: re-run it from the
 tab, and the issue closes at the next green CI run on `main`. "Auto release failed" means
 such a run was red or its release failed; the issue links the run.
 
-Once a week, `mutation-tests.yml` changes the code the unit tests cover one thing at a
-time and checks the tests notice. When they miss some, "Mutation testing: week to <date>"
-lists them: each is a place a fix could ship without a test that fails without it. Nothing
-waits on it; the next week's issue replaces it, and a week with every change caught closes
-it.
-
 ## What update-prs.yml checks before it says "ready to merge"
 
 Nothing merges itself (the owner's choice, 2026-10-02; from 2026-10-01 every kind of update

docs/notes/mpv.md

@@ -34,8 +34,7 @@ the new hashes). The installer counts because `update-local.ps1` tells an instal
 from the repository's by the host file alone; a raised `HostVersion` changes that file. The
 extension changes with it, so such a push releases, Firefox updates the extension, and
 the next hand-off on the owner's PC shows the "!" until `update-local.cmd` (or
-`install.ps1`) has run. The hash check is skipped in Stryker's sandbox, whose copy of
-the host holds every mutant.
+`install.ps1`) has run.
 
 **The thin helper (host 15, 2026-10-13).** The host changed 14 times in 10 days, each one a
 reinstall by hand, because policy lived in it. The owner chose to move the policy into the

docs/notes/workflows.md

@@ -327,8 +327,7 @@
   start: gh can give a log with every step "UNKNOWN STEP"). Without
   E2E_SHARD a config runs all its spec files (local `pnpm run verify`). The private
   browsing suite (one spec file) runs in the first extension group only. A run is 18 jobs
-  at once of the account's 20, so more groups would only queue; the weekly mutation run
-  (Mondays, 7 jobs for hours) leaves 13, and a push during it runs in two waves.
+  at once of the account's 20, so more groups would only queue.
   `fail-fast: false`, so one failing group does not cancel the others and their lists of
   retried specs. Per-job artifacts: `e2e-retried-<os>-<suite>-<group>`,
   `e2e-logs-<os>-<suite>-<group>`; `e2e-moz-logs-windows-<suite>-<group>` only from the Windows
@@ -647,37 +646,8 @@
 - All of it was dry-run with a stub `gh` against the real upstream history (sync: adopted
   -> closed, push -> no rebuild, new release named, comment only on upstream movement;
   patched libraries: open once, no duplicates, close when caught up or unpatched).
-- **`mutation-tests.yml`** (Mondays 06:40 UTC), 2026-10-01, T4: Stryker
-  (`stryker.config.mjs`, `pnpm run test:mutation`) over 25 pure-logic modules with unit
-  tests (10 small ones added 2026-10-03, #253; the config says which wait, and why). A report, not a gate: the run summary gets a table per file
-  (`tools/mutation-report.mjs`), the HTML report is the `mutation-reports` artifact, and when
-  the unit tests missed mutants (survived, or no test reaches them), one issue "Mutation
-  testing: week to <date>" lists them, assigned + @mention; next week's replaces it, a week
-  with all caught closes it. **The command runner**, not the vitest one: every mutant runs the
-  whole unit suite (`__STRYKER_ACTIVE_MUTANT__`), about 1.2 s each with 4 workers locally - on
-  CI a mutant the tests miss runs the whole ~18 s suite, and one job per area ran out of its
-  240 minutes at 51-74% with no report (#349, 2026-10-05): since then the three areas are
-  seven shards (`PARTS` in the config, split by file size), each in a job of 350 minutes, and
-  a shard that runs out of time (it ends cancelled) is reported as a failure. **Since
-  2026-10-06 the vitest runner** with per-test coverage: a mutant runs only the tests that
-  reached its code, in a Vitest that stays up (143 mutants in 31 s locally, ~12.5 tests a
-  mutant). Its 10.0.0 release (August 2026) predates vitest 5, which joins a test's suite
-  chain with ' > ', so its test filter matched nothing and every mutant "survived"
-  (stryker-js #6210); FastStream carries the fix, stryker-js PR #6214, as a pnpm patch
-  (`patches/@stryker-mutator__vitest-runner@10.0.0.patch`, the compiled files). When a
-  release has it, the Patched libraries workflow opens the pull request that drops the patch,
-  moves @stryker-mutator/core along (LOCKSTEP in `tools/check-patched-updates.mjs`: the runner
-  requires its core's exact version) and hands both back to Dependabot, whose `stryker` group
-  then keeps all of Stryker's packages in one pull request. A test that runs a module as a
-  process of its own (mpvHostVersion's host; the Windows-only installer and update-local
-  tests) does not see a mutant there. Compared mutant by mutant (command-runner run
-  37384050152, 4.7 hours; vitest-runner run 37393654214, 10 minutes): 11,313 of 11,370 alike.
-  The rest: each tool's entry line (`if (process.argv[1] ...`) a crashed run where the command
-  runner caught, two swapped on one line of SubtitleUtils - and the mpv host's message loop
-  (`main()`, `sendMessage()`), 24 mutants with no test reaching them. So the host is an area of
-  its own on the command runner (`COMMAND_AREAS`, `host-1`, about an hour and a half), and core,
-  network and tools are one vitest-runner shard each: four jobs. The sandbox leaves out `tsconfig.json`: Stryker rewrites it through
-  TypeScript's JS API, which TypeScript 7 does not have. `tests/workflows/mutation-tests.test.sh`.
-  Baseline, 2026-10-01 (73 min locally): 71.8% of 3,614 mutants caught; the weakest are the
-  mpv host (47.7%), MpvBackend (58.4%), SubtitleUtils (63.2%) and TabTracker (68.4%), the
-  best SubtitleSyncUtils (96%), MultiRegexMatcher, UrlMatchList and StreamPick (92%).
+- **`mutation-tests.yml`** (2026-10-01 to 2026-10-22): Stryker over the unit-tested modules,
+  weekly. Removed with Stryker: its reports (about 4,700 mutants not caught, 64.6% to 68%)
+  were closed by hand each week and no fix came from them, while it carried a 131-line patch
+  of its vitest runner, an override and 34 MB of packages. The fixes kept their own check: a
+  hand-made mutation of each one, which its test must catch.