You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
P-007: remaining POOL004/POOL005 slices — view stored into another field; deeper escape shapes #198
Agent brief — tier: mid (above-average). One PR (Closes #<this>); branch from main. House rules: AGENTS.md.
Goal
ROADMAP milestone 4 lists the remaining Pool/Span slices: POOL005 for a full-length view stored INTO another field (today: local and pooled-byte[]-field reads are covered) and deeper POOL004 escape shapes. POOL001–POOL003 are end-to-end; this extends working machinery, not new architecture.
Sources of truth
docs/proposals/P-007-arraypool-span.md — the pattern family, non-goals included.
Existing POOL001–POOL005 fixtures and tests — the style and pinning conventions to follow.
docs/ROADMAP.md milestone 4 — the exact remaining list.
Deliverable
Core (+ extractor where the shape needs C# facts) support for the two slices, each with bad/ok fixtures pinned in tests.
If a slice turns out to require interprocedural reasoning beyond the current machinery, stop, and record the boundary in the PR description (a deliberate deferral beats a soft false positive).
Guardrails
Precision-first: ambiguous flows stay silent or warning-tier per the house severity tiering — never a hard error on an unproven shape.
No FP regressions on the existing corpus: python tests/run_tests.py must stay green, and existing ok-fixtures must stay clean.
C# extractor changes are CI-validated only (no local dotnet).
Done when
New fixtures pinned (bad flags with the expected code, ok stays silent), full test suite + ruff check . + mypy green.
Activity