Repository navigation
Qualify Linux build-memory scopes across process lifetimes, RSS and shared pages - #2
Merged
crimson-knight merged 22 commits intoOct 20, 2026
Conversation
Preserve admitted installed compiler link paths, validate wrapper grammar before shell parsing, and package installation guidance with the verified SDK corpus. Hosted macOS and Linux SDK installations pass both compiler modes and reject the injected allocation regression.
crimson-knight
marked this pull request as ready for review
October 20, 2026 23:27
crimson-knight
merged commit Oct 20, 2026
d11113f
into
feature/native-environment-darwin-elf
7 checks passed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Build budgets need a complete process window and clearly named metrics. This dependent draft adds source-pinned Linux kernel controls, exact runner/tool admission, descendant and sibling workloads, self/largest-child RSS observations, a shared-file charge contrast, strict external validators, and durable raw evidence. The installed SDK package retains its existing identity.
Direct, fork/exec and detached workers each consume 64 MiB. Kernel peaks retain the charge after their complete return. A separate 192-MiB sibling load stays outside the observer budget. Two overlapping 64-MiB workers distinguish largest-child RSS (94,117,888 bytes in the recorded ordinary case) from whole-group charged peak (185,487,360 bytes).
The new shared-file case holds both consumers live on the same verified 96-MiB inode. Its reader reaches RSS 127,991,808 bytes while its own current charge is 17,899,520 bytes; the writer retains 121,696,256 bytes while the reader is live. Every consumer, the host check, and cleanup agree on the full file signature and physical identity. Each root also receives a refused write handle for its read-only cgroup membership file. Actual process migration retains a separate acceptance.
Validation at source
c33b03334b49c46e810f7d196428020b848777bd, actual run 39451872353:Current manifest:
233f9658ccd015e2c9e7c268723963ed3713dbb2fc4a552f65d2b52a13288cfd. The original seven-file provider remains15e27eea749e5d08f7d41256515b078f256c89fed5c49647c09dc99b99406eae. The shared archive SHA256 is19c4b36d2d60f3c988226f58597e7f1aa8f91a4e6cf09807a725051bc9903bdc. Public head8f19ce9b79895a328538dee2572e06ef232f5e4cretains the complete raw and acceptance datasets undertools/whole_build_memory/evidence/. Kernel, portable, documentation and Grant source-package workflows all pass at this current head. Normal source/dependency commit and push gates pass.The source package now preserves the complete representative Grant sources, exact dependency closure, full 300/500-model fixtures, origin/license records and immutable 144-cell contract. The original v100 hosted source-package run 39480858638 failed before its test cohorts: Ubuntu's built-in zlib has no module-file path, and Python 3.15 changed gzip's default compression level. Both failures and exact public archive bytes remain preserved.
The named v101 successor explicitly writes gzip level 9 and checks built-in/file-backed zlib identities against exact source-interpreter profiles. It retains the original 2,908,563-byte packet and exact 1,626,007-byte inner archive. Independent parent and peer executions pass all 58 original plus 24 successor regression executions; physical archive reconstruction and rebuilding from packaged sources reproduce the new 5,511,524-byte archive (
39ae53df3f72668f63101e0f65606cc41b2787c71292bff0f163031bb4bb309e). Run 39484820101 passes the source/package controls on Ubuntu 3.12.3 and macOS 3.15.0. Subsequent independent artifact review found that upload traversal followed the workspace provider symlink into checkout Git authentication metadata. Both affected public artifacts were removed; exact local evidence is protected, and this commit carries a failedatlas/artifact-confidentialitystatus. The workflow-only correction at8f19ce9b79895a328538dee2572e06ef232f5e4cdisables persisted checkout credentials, stages a fixed regular-file evidence roster with containment and identity checks, and uploads only that completed stage. All 20 source regression controls pass in each Python mode. Corrected run 39487648668 passes on both platforms. Independent inspection verifies each exact 32-member ZIP, every staged and raw-cohort hash, complete source/merge parity and physical archive reproduction. Its bounded nested-archive/Git-metadata/credential-shape screen passes; the new head carries a successful artifact-confidentiality status. The preceding failed status and artifact-removal evidence remain preserved. This source workflow creates zero real compiler/container/native/SSH workloads.This PR targets PR #1's source branch. Acceptance covers observed synthetic Linux process/RSS/shared-file protocols. Actual compiler RSS, process migration, other cache/reclaim situations, complete real builds, normal installation, clean devices, and the OOM-signal reliability question retain required evidence. The independent process-boundary step now collects its result after runner admission even if the separate kernel control refuses; that original refusal still fails the overall job. Grant's strict 300/500-model acceptance remains 0/144.