Repository navigation
test: embed generated source in the archived transmute guard - #548
thanosapollo wants to merge 1 commit into
Conversation
Archived nextest shards do not retain the producer OUT_DIR. Embed x11_colors.rs at compile time and scan that source instead of reopening its build path. Keep the ceiling at six and reject unsupported includes. The pure upstream base passes four focused tests but its archived run fails on missing x11_colors.rs. The regression-first generated-source check fails before the walker correction; all six candidate controls pass directly and from the archive with producer build directories hidden. This changes test infrastructure only, not Lisp/GNU semantics.
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info
📝 Walkthrough
Merge Risk: ⚪ Minimal · up to No actionable merge risk is identified for this test-only change; it is ready for normal checks. Pre-merge checks |
|
Found while investigating the archived core-test failure in main CI run 38017735406:
production_transmute_count_can_only_decreasetries to reopenx11_colors.rsin the producer'sOUT_DIR, which is absent on the shard runner.Embed that generated source at compile time and parse it through the existing walker. The ceiling remains six; generated code is still counted, ordinary includes are still read, and unsupported include forms still fail closed. No production or Lisp semantics change.
Tests against upstream base
baba54d9ae9b:Used nextest archive extraction and workspace remapping as in the shard workflow, with an unprivileged mount namespace hiding producer build directories. No generated assets borrowed from another worktree. This verifies the focused guard, not the full suite.
This PR is agent-assisted.