DrudgeCAS / gristmill

91%

master: 92%

LAST BUILD ON BRANCH fix/deterministic-optimization

branch: fix/deterministic-optimization

CHANGE BRANCH

x

Reset

Sync Branches

• fix/deterministic-optimization
• ci/release-0.9.0
• ci/release-please
• ci/release-please-setup
• ci/ruff
• copilot/add-cprinter-unit-tests
• copilot/sub-pr-16
• copilot/sub-pr-9
• copilot/update-git-submodules
• copilot/update-git-submodules-again
• dependabot/github_actions/actions/checkout-6
• dependabot/github_actions/actions/setup-python-6
• dependabot/github_actions/astral-sh/setup-uv-7
• dependabot/submodules/deps/fbitset-3a54534
• dependabot/submodules/deps/fbitset-bf87654
• dependabot/submodules/deps/libparenth-b1e240e
• dependabot/uv/dependencies-1ab08fcbb0
• dependabot/uv/dependencies-1d49c30b6a
• dependabot/uv/dependencies-890d39c709
• dependabot/uv/dependencies-b6b76a6067
• dependabot/uv/filelock-3.20.3
• dependabot/uv/idna-3.15
• dependabot/uv/pygments-2.20.0
• dependabot/uv/pytest-9.0.3
• dependabot/uv/requests-2.33.0
• dependabot/uv/tornado-6.5.7
• dependabot/uv/urllib3-2.7.0
• deps/bump-cpp-submodules
• deps/refresh-drudge-lock
• fix/deterministic-dimension-chunks
• fix/index-free-product-crash
• fix/order-independent-constriction
• fix/scikit-build-core-config-keys
• fix/verify-external-symbols
• master
• scikit
• test/restore-ccsd-energy-assertion
• tmp/check-effective-t-on-linux

Build # 31896956605

Build Type

Pull #39

github

Committed by chenpeizhi

Commit Message

refactor: Take the dimensions of a factor without scanning them all

Addressing review feedback.

Determinism was reached by looping over every dimension of the product and
keeping the ones the factor involves.  That is O(all dimensions) per
factor, where the original was O(symbols in the factor).  Collect the
dimensions the factor does involve and sort those instead.

The order is the same either way.  `dumm2dim` is filled in category then
index order, so walking it yields exactly the ascending `(cat, index)`
order that sorting produces.  The evaluation sequences come out
byte-identical, and the result is still the same across eight hash seeds.

No measurable speed difference on a twelve-matrix chain, 0.194 against
0.195 seconds, since this loop is a small part of the work.  The change is
worth having for the complexity rather than for a number.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

Pull Request Pull Request #39: fix: Stop unordered iteration deciding the optimization result

Coverage Stats

3 of 3 new or added lines in 1 file covered. (100.0%)

1782 of 1953 relevant lines covered (91.24%)

0.91 hits per line