[jonyoder]

Bumps go-pubgrub to v0.6.0 and sets Solver.Interrupt = ctx.Err where resolveOnce builds each solver, so a solve stops when its context is cancelled or times out (rstudio/package-manager#22613). The error comes back unwrapped by explain, so errors.Is(err, context.DeadlineExceeded), context.Canceled and solver.ErrInterrupted all match from resolver.Resolve.

---

Where the solver is built

resolver.resolveOnce is the only solver.New call outside tests. It covers the first solve, every YankExemptTransitivePins re-solve in its loop, and every UpgradeRoots re-solve (upgradeRoots calls resolveOnce). A cancel during the UpgradeRoots pass still returns the first answer with no error, as before; the re-solve now stops right away.

Tests

New resolver/interrupt_test.go, with an index that ignores the context so only the solver's check can see a cancel:

• expired deadline: Resolve returns at once, errors.Is matches DeadlineExceeded and ErrInterrupted, not a ResolutionError
• cancel mid-solve: errors.Is(err, context.Canceled)
• cancel as an UpgradeRoots re-solve starts: that re-solve makes no decision past the root; base answer returned

All three fail with the Interrupt line removed. go test ./... passes; vet and gofmt clean.

Independence audit

PASS: no retrieval of a known reference implementation in 222 session(s). (after agentsview sync, --paths on this worktree and this session's scratchpad). It also lists the same 2 cross-repo risk entries as go-pubgrub#26 (agent-a63f03d8042244b81, downloading a uv release binary on a dogfood host); not from this work.

Clean-room: written from go-pyresolver and go-pubgrub code only; no uv or pubgrub-rs source read.

RoboRev job 5245 on 3435110: no findings.

🤖 Generated with Claude Code