Commits on Feb 23, 2026

1. fix(runfiles): assume main repository on Windows (#3578)

   With any Python version on Windows, `RUNFILES_*` environment variables
   still point to an intermediate stage path instead of the module path,
   causing `CurrentRepository()` to raise a `ValueError` since
   `rules_python` v1.8.0:
   ```
   ValueError: C:\Users\bot\AppData\Local\Temp\Bazel.runfiles_gh34ij5_\runfiles\+_repo_rules+cpython\my_script.py does not lie under the runfiles root C:/cache/ab1cdef2/execroot/_main/bazel-out/x64_windows-fastbuild/bin/external/+_repo_rules+cpython/install.exe.runfiles
   ```
   
   This was not the case with `rules_python` v1.7.0. The issue stems from
   behavior.
   
   Since #3086 came up with a corresponding workaround in
   `//tests/runtime_env_toolchain:toolchain_runs_test`, the proposed fix
   simply consists in moving it to `CurrentRepository()`, thus adding
   another case to the workaround introduced by #1634 for Python < 3.11. It
   therefore leads to assuming the main module path on Windows as well.
   
   Removing the workaround from `CurrentRepository()` would make the test
   fail as follows:
   ```
   ==================== Test output for //tests/runtime_env_toolchain:toolchain_runs_test:
   E
   ======================================================================
   ERROR: test_ran (__main__.RunTest.test_ran)
   ----------------------------------------------------------------------
   Traceback (most recent call last):
   [...]
   ValueError: C:\Users\user\AppData\Local\Temp\Bazel.runfiles_1f08smy6\runfiles\_main\tests\runtime_env_toolchain\toolchain_runs_test.py does not lie under the runfiles root c:\users\user\_bazel_user\cxxeswjo\execroot\_main\bazel-out\x64_windows-fastbuild-st-c530e4918e48\bin\tests\runtime_env_toolchain\toolchain_runs_test.exe.runfiles
   ```
   
   Fixes #3579.
   
   (cherry picked from commit f78add7)

   

   authored and rickeylev committed

   Feb 23, 2026
   Configuration menu

   Browse the repository at this point in the history

2. fix(pip): preserve PEP 508 URL-based requirements when extract_url_sr…

   …cs=False (#3582)
   
   pip_parse (via pip_repository) passes extract_url_srcs=False to
   parse_requirements. The _package_srcs() function silently dropped PEP
   508 URL-based requirements (pkg @ https://...) in this mode because
   _add_dists() returns can_fallback=False for URL requirements, causing
   them to fall through to the `continue` statement.
   
   Fix the elif condition to also accept the case where extract_url_srcs is
   False but a valid URL dist exists, falling back to pip to handle the URL
   requirement directly.
   
   Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
   (cherry picked from commit 9dcbabb)

   2 people authored and rickeylev committed

   Feb 23, 2026
   Configuration menu

   Browse the repository at this point in the history

3. fix(pypi): normalize extras in requirement strings per PEP 685 (#3588)

   ## Summary
   
   Extras parsed from requirement strings (e.g., from `requirements.txt`)
   were not being normalized, causing mismatches when evaluating PEP 508
   marker expressions. For example, `sqlalchemy[postgresql-psycopg2binary]`
   would fail to resolve `psycopg2-binary` as a transitive dependency
   because the wheel METADATA marker expression `extra ==
   "postgresql_psycopg2binary"` uses the underscore-normalized form (per
   PEP 685), while the extras set retained the original hyphenated form
   from the requirement string.
   
   ## Before
   
   ```
   # requirements.txt
   sqlalchemy[postgresql-psycopg2binary]==2.0.36
   
   # Parsed extras: ["postgresql-psycopg2binary"]
   # Marker evaluation: "postgresql-psycopg2binary" != "postgresql_psycopg2binary" -> MISS
   # Result: psycopg2-binary NOT included as a dependency
   ```
   
   ## After
   
   ```
   # requirements.txt
   sqlalchemy[postgresql-psycopg2binary]==2.0.36
   
   # Parsed extras: ["postgresql_psycopg2binary"]  (normalized)
   # Marker evaluation: "postgresql_psycopg2binary" == "postgresql_psycopg2binary" -> MATCH
   # Result: psycopg2-binary correctly included as a dependency
   ```
   
   ## Changes
   
   - **`python/private/pypi/pep508_requirement.bzl`**: Apply
   `normalize_name()` to each extra during requirement parsing, consistent
   with how the package name is already normalized.
   - **`tests/pypi/pep508/requirement_tests.bzl`**: Updated existing test
   expectation for case normalization and added test case for hyphenated
   extras
   (`sqlalchemy[asyncio,postgresql-psycopg2binary,postgresql-asyncpg]`).
   - **`tests/pypi/pep508/deps_tests.bzl`**: Added
   `test_extras_with_hyphens_are_normalized` integration test confirming
   that dependencies gated behind hyphenated extras are correctly resolved.
   - **`CHANGELOG.md`**: Added entry under Unreleased > Fixed.
   
   Fixes #3587
   
   ---------
   
   Co-authored-by: Ignas Anikevicius <240938+aignas@users.noreply.github.com>
   (cherry picked from commit 9fe42b1)

   2 people authored and rickeylev committed

   Feb 23, 2026
   Configuration menu

   Browse the repository at this point in the history