Validate the build tag passed to wheel pack --build-number

[dylanpulver]

wheel tags --build runs its value through parse_build_tag; wheel pack --build-number does not, and splices it straight into the file name (pack.py:109, name_version += "-" + build_number). The sibling --local-version option, handled a few lines further down the same function, is fully validated.

Measured on 60cfdda, packing a minimal test-1.0 tree:

--build-number foo  ->  test-1.0-foo-py3-none-any.whl
    wheel's own WHEEL_INFO_RE : build=None, pyver='foo', abi='py3', plat='none-any'
    packaging 26.3            : InvalidWheelFilename: Invalid build number: foo

--build-number 1-2  ->  test-1.0-1-2-py3-none-any.whl
    wheel's own WHEEL_INFO_RE : build='1', pyver='2', abi='py3', plat='none-any'
    packaging 26.3            : InvalidWheelFilename: wrong number of parts

So pack writes Build: 1-2 into WHEEL and wheel info on the file it just produced reports Build: 1; with foo it reports no build tag at all and reads none-any as the platform tag. The filename grammar is {distribution}-{version}(-{build tag})?-{python tag}-{abi tag}-{platform tag}.whl with the build tag required to start with a digit, so a - cannot occur inside it.

The fix is type=parse_build_tag on the pack argument — the validator tags already uses.

Ran pytest tests/: 74 passed. Reverting only src/wheel/_commands/__init__.py to origin/main and keeping the new test gives 72 passed / 2 failed. A narrower variant that checks only the leading digit (the rule _bdist_wheel.py:309 applies) still fails the 1-2 case, so both halves are load-bearing. ruff check on the two files I touched is clean; the 22 errors ruff 0.16.5 reports repo-wide are identical on an unmodified checkout, so I did not reformat anything.

Not fixed here, but measured: bdist_wheel has the leading-digit half of the rule and not the - half, and setup.py bdist_wheel --build-number 1-2 produces demo-1.0-1-2-py3-none-any.whl the same way. I left it out to keep this small, since it is a behaviour change on a much more used path — happy to add it here or in a separate PR, whichever you prefer.

AI assistance: this change was drafted with Claude Opus 5 (claude-opus-5). Every number above was produced by running the commands locally against this branch.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 63.45%. Comparing base (60cfdda) to head (f3624f2).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #698   +/-   ##
=======================================
  Coverage   63.45%   63.45%           
=======================================
  Files          15       15           
  Lines        1267     1267           
=======================================
  Hits          804      804           
  Misses        463      463           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[agronholm]

Alright, looks sensible. Thanks!