mirror of
https://github.com/DeusData/codebase-memory-mcp.git
synced 2026-11-04 10:48:20 +00:00
Martin VogelandFelix Rath
f9f44b5869
With mimalloc 3.3.2's release build, free(NULL) before the allocator's constructor dereferenced a page map that does not exist yet. glibc 2.44's newlocale() does exactly that during libstdc++ start-up, so every Linux binary died with SIGSEGV before main (reported with a reproduction in #2562). Upstream fixed it in b2226c19 (mimalloc #1341), first released in v3.4.4; v3.5.3, upstream's latest published release, contains it. Vendored from microsoft/mimalloc tag v3.5.3 (tag object deac3dcf, commit d4881d338125e1cb7c47ba4cfb398d6f7c0c8d45): LICENSE, include/ and src/, written blob by blob from a bare clone. Before the local patches were reapplied, the staged LICENSE blob and include/ and src/ trees had the tag's exact object ids (50 files; src/arena-meta.c is gone upstream). Afterwards the only differences are two documented hunks: - src/prim/unix/prim.c: macOS always recommits on decommit, carried from63fb7b3d(the phys_footprint fix). - src/prim/osx/alloc-override-zone.c: one comment reworded, because scripts/security-vendored.sh rejects the upstream "fork()" spelling even inside a comment. No code change. The reproducible-banner patch (no __DATE__/__TIME__) is dropped: upstream removed the build clock in v3.4.5 (mimalloc #1355). The two Makefile comments that pointed at that patch now say so. The production object is built with -DMI_PROFILE=0: mimalloc 3.5.2+ compiles its allocation-sampling profiler in by default, with a sample countdown on the allocation path, and the product never uses it. vendored/mimalloc/PATCHES.md records the pin, a re-verification recipe, the check that pins each patch and the refresh procedure; THIRD_PARTY.md points at it, the SBOM version is 3.5.3 and scripts/vendored-checksums.txt is regenerated (mimalloc lines only; security-vendored.sh passes). Measured on Apple M5 Pro (macOS 26.6), full indexes, 2 interleaved reps, mainfceb25d8(3.3.2) -> this change: TypeScript peak phys_footprint 1.07/1.01 -> 1.04/1.04 GB, peak RSS 1.87/1.84 -> 1.85/1.86 GB, wall 30.6/29.1 -> 29.3/30.0 s; django footprint 0.41/0.41 -> 0.38/0.39 GB, RSS 0.59/0.59 -> 0.58/0.60 GB, wall 6.5/6.2 -> 6.5/6.5 s. The graphs are identical: full node and edge hashes, node and edge census, no dangling edges (django 68022 nodes / 322464 edges, TypeScript 306239 / 835984). The same tree without the recommit patch peaks at 2.09/2.04 GB footprint on TypeScript, so63fb7b3dstill does its job. Linux arm64 (glibc 2.39, gcc 13.3): both trees build without warnings; django peak RSS 0.67/0.68 GB on main, 0.69/0.68 GB with this change. Co-authored-by: Felix Rath <felixm.rath@gmail.com> Signed-off-by: Martin Vogel <martin.vogel.tech@gmail.com>
scripts/ci/ — venue plumbing (single implementations)
Support scripts that keep the venues in the SAME shape. Each exists because
the logic used to live inline in workflow YAML or was hand-duplicated between
CI and the local infrastructure — both of which the venue-parity contract
(tests/test_venue_parity_contract.sh) now forbids. Everything here answers
--help (PowerShell: comment-based help, Get-Help <script>).
| script | job | called by |
|---|---|---|
new-protected-temp-root.ps1 |
Create the owner-stamped, inheritance-protected per-user TEMP root the daemon/install suites require (shared /tmp and default runner TEMP grant Authenticated-Users mutation rights, which the trust policy correctly refuses — running there produces security refusals, not signal). -ProtectDir stamps build dirs the same way. |
_test.yml, _soak.yml, vm-run-tests.sh |
clean-test-residue.ps1 |
Sweep cbm-* residue from the Windows VM and assert runner-like free disk (default 14 GB — the GitHub runner's SSD). Long-path \\?\ fallback for the guard suites' adversarial trees; every removal VERIFIED (an earlier version counted attempts and reported 86 swept while 11 GB remained). BLOCKS below the floor: a disk that fills mid-run reads as a product bug. |
win.sh before every build/run |
preflight-docker.sh |
Same idea for Colima/docker: prune runner-unlike residue, assert free space on the filesystem backing the docker data root (not the VM's /). Build cache + named volumes KEPT (the local analogue of actions/cache); --deep drops them. |
test-infrastructure/run.sh |
check-glibc-compat.sh |
Run a linux binary in debian:bullseye (glibc 2.31) — the portable binary must start on old glibc. | _smoke.yml portable legs |
generate-sbom.py |
The release SPDX SBOM (vendored versions reviewable here, diffable by vendoring PRs — was inline YAML). | release.yml |
require-all-green.sh |
The aggregate gate: fail unless every needed job succeeded or legitimately skipped (was inline YAML) — and, under lane selection, every SELECTED lane ran and succeeded (skipped/cancelled/missing = red; unselected lanes may skip). | pr.yml ci-ok |
select-lanes.sh |
Decide which PR lanes a change set runs (tier, lanes, full, reasons) from its file list — the R1 rule set, fail-safe to every lane. Pinned by tests/test_select_lanes.sh and the September 2026 history replay (scripts/test-impact/replay-selector.sh). |
pr.yml changes |
test-impact-shadow.sh |
The report-only test-impact shadow: fetch the latest released binary (checksum-verified) or take a given one, index this checkout into a throwaway cache, run detect_changes against the base and map it to suites (scripts/test-impact/predict.py). Never gates. |
pr.yml test-impact-shadow |
verify-shard-union.sh |
Prove sharded test legs lost nothing: shard count agreement, indices 1..n, identical suite lists, union of slices == full list (was inline YAML). No manifests is green only when the lane selection holds no test leg. | _test.yml shard-completeness |
prepare-release-candidates.sh |
Copy one linker output into stripped/unstripped candidates, finalize signatures, composition-check them without execution, and record their hashes. | _build.yml, local artifact smoke |
stage-release-candidates.py |
Admit exactly eight candidate artifacts / sixteen byte-distinct binaries into the content-addressed VirusTotal scan set. | _build.yml |
select-release-candidates.py |
Apply the reviewed tuple-local VT truth table, or the explicit dry-run stripped default, and atomically copy one content-bound binary per target. | _build.yml |
verify-release-selection.py |
Recompute the selection policy and prove every executable member in all 14 public containers equals its selected SHA-256. | _build.yml, release.yml final draft verification |
check-virustotal.sh |
Poll and validate the exact candidate scan set, enforce engine coverage and the narrow documented Microsoft !ml policy, and emit content-bound results evidence. |
_build.yml |