|
| 1 | +# FEATURE-ci-macos-make-fixture: preserve make frontend identity in fixture aliases |
| 2 | + |
| 3 | +Status: In Progress — implementation ready; caller validation/finalization pending. |
| 4 | +Base: local `v3` at `9b93ebe8eb355b6da6099f8ed7764cda8c23bb90`. |
| 5 | + |
| 6 | +Native macOS package job `120582231719` and archive Darwin job `120582524154` |
| 7 | +both fail the new named `gmake` regression during real configure. Their |
| 8 | +61-test fixture suites report one failure and three skips. Canonical GNU make |
| 9 | +version detection and absolute/default selections succeed; renaming Apple's |
| 10 | +developer-tool frontend through a `gmake` symlink makes its name-sensitive |
| 11 | +dispatch fail to locate `gmake`. |
| 12 | + |
| 13 | +The fixture now creates executable POSIX launchers for its `make` and `gmake` |
| 14 | +PATH entries. Each launcher execs the shell-quoted original tool path with |
| 15 | +`"$@"`, preserving the tool's invocation identity and argument boundaries. |
| 16 | +The broken default still executes the deliberately failing unselected tool. |
| 17 | +Selected MAKE values, ambient conflicts, real configure/dependency generation, |
| 18 | +compile/install/check behavior and assertions remain intact. The production |
| 19 | +driver, including both selected-MAKE environment assignments, is byte-identical |
| 20 | +to the base. No dependency, platform skip, workflow or production change. |
| 21 | + |
| 22 | +Action items: |
| 23 | + |
| 24 | +- [x] Read the plan, literal journal, both actual native failure logs and complete fixture before editing. |
| 25 | +- [x] Create the isolated Groundwork worktree from the exact local v3 base; preserve unrelated work. |
| 26 | +- [x] Record the actual native regression as RED before the minimal launcher repair. |
| 27 | +- [x] Replace renamed frontend symlinks with canonical-tool exec launchers. |
| 28 | +- [x] With the corrected fixture, run all three real Autoconf cases against the original pre-MAKE-fix driver as counterfactual RED. |
| 29 | +- [x] Run the same three cases against the current unchanged driver as GREEN; retain actual dependency and exact MAKE receipts. |
| 30 | +- [x] Verify relevant preservation tests, launcher syntax/argument/status behavior, Python syntax and diff checks. |
| 31 | +- [x] Retain raw evidence, executed sources, source bindings, versions and verification limits. |
| 32 | +- [ ] Caller: complete required Groundwork validation and local finalization into `v3`. |
| 33 | +- [ ] CI / v3 PR: confirm repaired fixture on native macOS package/archive jobs. |
| 34 | + |
| 35 | +Verification: 22 existing installed-shell, ps and matrix contracts pass before |
| 36 | +and after the repair. The corrected fixture with the exact original driver |
| 37 | +from `c2d5c626f456f218676a6d840c8ed1442aa28438` fails all three cases in real |
| 38 | +`config.status` dependency-fragment bootstrapping (expected RED). With the |
| 39 | +current driver all three pass in 7.083s: actual configure, compile and install |
| 40 | +succeed, then the deliberate check failure reaches the required marker. |
| 41 | +`.deps/probe.Po` includes `probe.h`; build, install and check receipts retain |
| 42 | +exactly `gmake`, `/usr/bin/gmake` or `make` for their respective selections. |
| 43 | +The named case still has an unusable PATH default; the absolute/default cases |
| 44 | +still override a conflicting inherited MAKE. |
| 45 | + |
| 46 | +Both real valid launchers also execute a GNU make target containing spaces |
| 47 | +through `-f -`, preserve its output and propagate GNU make's exit2 from the |
| 48 | +deliberate recipe exit17. All captured shell launchers pass `/bin/sh -n`; |
| 49 | +`python3 -m py_compile scripts/test-v3-installed-consumer.py` and |
| 50 | +`git diff --check` pass. No Python linter is configured in repository guidance. |
| 51 | +No full baseline, broad retry, unrelated repair or old runtime diagnosis was |
| 52 | +performed. Native repaired macOS proof remains incomplete and CI/v3-PR owned; |
| 53 | +local Linux proof does not claim Apple frontend execution. |
| 54 | + |
| 55 | +The main repository's ignored `.debug/ci-macos-make-fixture-fdno1_wr/` retains |
| 56 | +the literal journal, native logs, before/after preservation logs, source-bound |
| 57 | +`baseline-actual-configure/` and `fixed-actual-configure/` projects/commands, |
| 58 | +launcher contract output and versions. The evidence harness reuses the prior |
| 59 | +task's actual execution setup, imports the registered test class and executes |
| 60 | +the original driver blob in its normal module with the original file binding. |
| 61 | +It does not model configure or the macOS frontend. |
| 62 | + |
| 63 | +SHA256 bindings: final fixture |
| 64 | +`88ca80f1759e5784534dd458da374814dc76b47652efc851e742877fced393a3`; |
| 65 | +original driver |
| 66 | +`61ef4487227c4196fbb7fb486e90b3bba255452e678ba47e8754ed8fa0383bc8`; |
| 67 | +unchanged current driver |
| 68 | +`6c565c7a7c3af16d46edc126638384fe15d81b8d7c36c098bb2661ec593dafb4`. |
| 69 | + |
| 70 | +Project notes / task-executor memory: v3 uses Autotools/C++20; this test-only |
| 71 | +repair needs no provider/full library setup. Focused checks use |
| 72 | +`python3 scripts/test-v3-installed-consumer.py InstalledMakeTest InstalledShellTest InstalledShellPSTest MatrixTest` |
| 73 | +and Python compilation. The real fixture uses `autoreconf --install --force`, |
| 74 | +Autoconf2.71, Automake1.16.5, GNU Make4.3, GCC13.3.0 and Python3.12.3 locally. |
| 75 | +Its existing script registration/distribution remains sufficient. Groundwork's |
| 76 | +identity helper resolves the worktree/branch; git metadata requires sandbox |
| 77 | +escalation. Preserve canonical frontend invocation names with exec launchers |
| 78 | +rather than renamed symlinks. No merge, push or CI cancellation by executor. |
0 commit comments