Repository navigation
01 modernize - #99
Merged
Merged
01 modernize#99
Conversation
The project files date from NetworkX 3.2, and pre-commit no longer runs to completion on a current interpreter: pyupgrade 3.15 fails under Python 3.14. Hook pins now track python-graphblas, and the formatting the new hooks ask for is applied in the same commit. - Require Python 3.11 (classifiers, black, ruff and pyupgrade targets). - Declare the license as a PEP 639 expression (setuptools >= 77). - Drop the networkx.plugins entry point and the graphs' __networkx_plugin__ attribute, which NetworkX stopped reading in 3.3; networkx.backends, networkx.backend_info and __networkx_backend__ remain. - The test extra requires networkx >= 3.6, adds pytest-xdist and drops tomli: test_core reads pyproject.toml with tomllib. - pytest runs with --strict-config, --strict-markers and -ra. - Move ruff's lint settings under [tool.ruff.lint], and ignore the rules its new version adds that this code does not follow. - New hooks: shellcheck, prettier (README included; the generator's markers take prettier's form), actionlint, zizmor, yamllint, schema checks for the workflows, and a check for plain ASCII punctuation. - Current GitHub actions, pinned by commit as zizmor asks; checkouts no longer persist credentials, every workflow's token is read-only, and Dependabot waits a week before proposing an action's new release. The upload to PyPI still uses an API token, so zizmor's trusted-publishing audit is waived on that line. - Add dev-requirements.txt, the pip mirror of environment.yml.
The function list NetworkX reads from _nx_graphblas was kept by hand, beside a dispatch table that already had a generator. A new _nx_graphblas/core.py writes it from the Dispatcher and the wrappers' signatures, and documents each parameter NetworkX's signature lacks with its live default. The generated get_info() equals the hand-written one, with the functions in sorted order. scripts/maketree.py finds the implemented names by scanning nxapi against NetworkX's registry instead of reading the dispatch table it writes, so a new wrapper no longer has to be listed by hand first. `make all` runs both generators, the dispatch table first, since the backend info is read from it. run_nx_tests.sh drops NETWORKX_GRAPH_CONVERT, which NetworkX no longer reads.
The README said the package registers a "networkx.plugin" entry point and that graphs carry `__networkx_plugin__`. NetworkX has dispatched through the "networkx.backends" group and `__networkx_backend__` since 3.2, and the plugins entry point is gone. Drop the sentence calling dispatch a new feature of NetworkX 3.0.
The structure checks in test_match_nx.py ran only with --check-structure, which CI passes, and test_dispatched_funcs_in_nxapi fails there: the aliases fast_graph_could_be_isomorphic, faster_graph_could_be_isomorphic and tournament.is_path displace the real definitions in its map of where each dispatched name lives. NetworkX 3.6 dispatches none of the three, so they go. The option, its marker and the root conftest.py that defined it go too, and the structure checks always run.
Three files are written by generators: the function list and parameter table in _nx_graphblas/__init__.py, the algorithm tree in README.md and the dispatch table in interface.py. Nothing noticed when one of them fell behind the code. test_autogenerated_code_is_fresh runs both generators against copies in tmp_path and compares, so a stale checkout fails instead of being repaired; it skips where scripts/ is not shipped. A second test compares the committed backend info with the Dispatcher, which also works on an installed package. rich, which maketree imports, joins the test extra and the CI install list. maketree splices its two files itself instead of calling python-graphblas's private _autogenerate_code, which wrote a file back unchanged when a marker was missing, so a damaged file compared as fresh, and took a duplicate marker for a second block. Both generators now require each marker exactly once. The splice reads and writes UTF-8, since the README tree is box-drawing characters that Windows' default code page cannot encode. maketree no longer runs black; its output is already formatted. maketree's get_names stops on a public nxapi name that NetworkX does not dispatch, instead of leaving it out of the generated code while it stays importable. A module's public names are its __all__, or else the functions it defines. On the older supported NetworkX this also stops on a wrapper for a name only the newer one dispatches. The README test matches whole words: as a substring, degree_centrality counts as documented by in_degree_centrality. The fixture of implemented names scans nxapi as the generator does, not the generated table, and the branch for NetworkX 3.0 and 3.1 goes. Two tests pin where dispatched names come from: every Dispatcher entry is an nxapi wrapper, and nxapi.linalg's submodules are its own. Without __all__ in nxapi/linalg/graphmatrix.py, the star imports replace nxapi.linalg with the kernel package, and adjacency_matrix dispatches to the kernel. dispatcher_functions lives in tests/_dispatched.py, a helper module the test-naming hook now skips.
Each cell of the Tests matrix ran our tests and then NetworkX's suite through the backend in one step. The suite does not pass yet, so the step failed, and fail-fast let the first failing cell cancel the rest: whether the package installs and passes its own tests was not visible. The suite moves to a job of its own, one cell on ubuntu with Python 3.13, which still uploads its coverage. Its step may fail and the job still passes; the next step writes the summary line to the step summary, with a warning annotation when the suite failed. The step sets pipefail itself, since the runner adds none to a custom shell and tee would otherwise decide its status. In the pytest job: - fail-fast is off, so one failing cell no longer cancels the others. - The cells are named "pytest (os, version)", which renames the checks. - macos-15-intel and ubuntu-24.04-arm join, at Python 3.13. - setuptools, which test_packages needs, is installed. pytest-mpl, which brings matplotlib for NetworkX's drawing tests, moves to the NetworkX job. - The pypy conditionals on the channels go; no cell is pypy.
Codecov Report❌ Patch coverage is Additional details and impacted files@@ Coverage Diff @@
## main #99 +/- ##
==========================================
- Coverage 76.82% 74.44% -2.38%
==========================================
Files 118 120 +2
Lines 4271 4403 +132
Branches 818 727 -91
==========================================
- Hits 3281 3278 -3
- Misses 751 877 +126
- Partials 239 248 +9 🚀 New features to boost your workflow:
|
On Windows, setup-miniconda writes a bash wrapper for mamba that runs `cmd.exe /C CALL mamba.bat`. Git Bash rewrites `/C` to the path `C:/`, so cmd.exe opens an interactive prompt, reads end of input and exits 0. Nothing was installed, and the Windows cells failed later on `import graphblas`. conda's default solver is libmamba, so calling conda directly costs nothing.
The backend info and the generator that writes it live in _nx_graphblas, outside the measured source, so the tests that run the generator counted for nothing.
Member
Author
|
Passing CI! This is a good first step. Merging. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
First step to modernize and begin bringing graphblas-algorithms up to date