Avatar for the alkahest-cas user
alkahest-cas
alkahest
BlogDocsChangelog

Performance History

Latest Results

fix(ci): ruff lints, and a budget test that two branches raced `ruff check` (CI pins 0.15.14 via uv.lock) reported 9 errors this branch introduced: RUF059 unused unpacked bindings and an E741 ambiguous `I` in test_parametric_groebner.py, and three SIM117 nested `with` blocks in test_resource_budgets.py. The `I` is the SIR model's infected compartment, so it is renamed `infected` in code while the symbol's *name* stays "I". More interesting: `test_prove_nonneg_honours_a_wall_budget` was a race created by merging two branches that were each correct alone. `fix/resource-budgets` wrote it against a build where `prove_nonneg` on the homogeneous ternary Motzkin form ran 418 s, so asserting a 1000 ms budget is exceeded was unmissable. `fix/sos-motzkin` then added the half-Newton-polytope reduction, which makes that same call *succeed* in ~0.95 s — measured 0.999 / 0.951 / 0.955 s here. 1000 ms against 950 ms is a coin flip, which is why it passed alone and failed in-file. The budget is now 100 ms. What the test is for is that a budget stop surfaces as BudgetExceededError / E-BUDGET-001 rather than E-SOS-002 — any budget well under the real cost demonstrates that, and 1, 10, 50 and 100 ms each tripped 4/4. The rationale is in the test so the next person to speed up SOS does not have to rediscover it. Also formats tests/test_ball_enclosure_soundness.py, the only file this branch leaves unformatted under CI's pinned ruff. (Under 0.16.3 two `tests/**/README.md` files also reformat; those reproduce on unmodified main and are a version artefact, so they are left alone.) Verified with CI's own pins — ruff 0.15.14: `ruff check python/ tests/` all checks passed, `ruff format --check python/ tests/` 191 files already formatted. pytest tests/: 3547 passed, 61 skipped, 0 failed. test_resource_budgets.py 11 passed on three consecutive runs. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
integrate/autoresearch-2026-08-19
12 minutes ago
fix(ci): unblock the four failing checks on this branch Three distinct causes, only one of them introduced here. **Ours.** The new private `validated::interval` module, the ball-soundness and resource-budget internals, and the SOS ceiling accessors documented themselves with rustdoc links to private items (`[`super::interval`]`, `[`IntervalEval::eval_node`]`, `[`sub_live`]`, `[`MAX_GCD_BITS`]`, `[`MAX_GCD_WORK`]`, `[`MAX_FREE_PARAMETERS`]`, `[`super::multiplier_search`]`). `cargo doc` runs with `-D warnings` in the docs job, so eight of these failed the build. Demoted to plain code spans, which is what a private item can be named as. **Pre-existing, clippy 1.98.** `needless_late_init` now fires on `matrix/smith_poly.rs:266` and `matrix/smith.rs:292`. Both files predate this branch (last touched 2026-05-03), and CI's `dtolnay/rust-toolchain@stable` has moved to 1.98 while local was 1.97 — which is why this passed locally and failed in CI. Rewritten as `let result = if ... { res } else { res };`. Verified by installing 1.98.0 and running CI's whole clippy matrix (default, egraph, parallel, cranelift, groebner-cuda) at `-D warnings`: zero errors. **Pre-existing, macOS.** `ci-cross.yml` exported `DYLD_LIBRARY_PATH` at Homebrew LLVM 15. That variable takes precedence over a dylib's own install path, so it redirected *rustc itself*: `librustc_driver` links its own bundled LLVM, and `rustc -vV` died with `Symbol not found: __ZN4llvm15SmallVectorBaseIjE13mallocForGrowEPvmmRm` (SIGABRT) before a single crate compiled. Switched to `DYLD_FALLBACK_LIBRARY_PATH`, which is consulted only when a library is not found by its normal path — what the jit build needs, and nothing else. CodSpeed is not addressed here: it is a performance check with no branch protection, and it reports against the branch rather than failing a build step. Verified: cargo fmt --check clean; cargo doc --workspace --no-deps with RUSTDOCFLAGS="-D warnings" clean; cargo +1.98.0 clippy across all five CI feature combinations clean; matrix:: tests 61 passed. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
integrate/autoresearch-2026-08-19
49 minutes ago

Latest Branches

CodSpeed Performance Gauge
-13%
fix: the 2026-08-19 autoresearch run — two Critical soundness bugs, a false mathematical claim, and an uncatchable process abort#310
20 minutes ago
a0cf86f
integrate/autoresearch-2026-08-19
CodSpeed Performance Gauge
0%
52 minutes ago
e115260
feat/llvm21-upgrade
CodSpeed Performance Gauge
0%
Generalize telescope2d to arbitrary m>=1 bound indices#308
7 days ago
1343304
worktree-agent-a69c57d7de51c2040
© 2026 CodSpeed Technology
Home Terms Privacy Docs