Repository navigation
Conversation
Retain build edges for cached top-level versions so later cache misses rebuild in dependency order (RHAI-7566, RHAI-22532, RHAI-22533). Co-Authored-By: Codex GPT-6 <noreply@openai.com> Signed-off-by: Vikash Shaw <vshaw@redhat.com>
|
No actionable comments were generated in the recent review. Walkthrough
Merge Risk:
|
|
Tick the box to add this pull request to the merge queue (same as
|
In multiple-version bootstrap, cached top-level versions were left in graph.json without build edges and omitted from build-order.json. A later cache miss made build-parallel rebuild these nodes in round 1, before their build backends were available in the local binary-only index.
Retain cached top-level versions for dependency discovery. The existing cached-wheel preparation path reads recorded requirements from the wheel, and the build phase reuses that wheel. Cache filtering of transitive dependency versions is preserved. This trades additional metadata/dependency processing for a complete build graph; it does not serialize builds or force recompilation of cache hits.
The regression uses two cached wheel archives with different backend requirements. It checks metadata reuse without source downloads or wheel compilation and verifies that both backends are scheduled before their applications when the remote cache is subsequently unavailable. A parameterized test also covers mixed cached and uncached top-level versions while preserving transitive cache filtering.
Validation: 234 bootstrap, graph and requirement resolver tests passed; Ruff check, Ruff format check and git diff --check passed. The new regression fails on unmodified upstream main. Production bootstrap timing and the affected wheel builds still need CI validation.
Related to RHAI-7566, RHAI-22532 and RHAI-22533. RHAI-44490 is a possible additional consumer, but its historical graph artifacts are expired and this PR does not claim to resolve it. Fondue must adopt a released Fromager version containing this change before its pipeline tickets can be validated and closed.
Co-Authored-By: Codex GPT-6 noreply@openai.com
Signed-off-by: Vikash Shaw vshaw@redhat.com