get_release_info() function assumes that custom python interpreter URL has release date in it

[ybaturina]

🐞 bug report

Affected Rule

The issue is caused by the function:

get_release_info

Is this a regression?

Yes, the previous version in which this bug was not present was: 0.39.0

Description

The issue is caused by this line:

release_id = int(u.split("/")[-2])

If the URL doesn't follow a specific format, this code will fail.

🔥 Exception or Error

Example:

For the URL file:///__w/jax/jax/python-tsan.tgz the error below is thrown:

ERROR: Traceback (most recent call last):
    File "/__w/jax/jax/jax/WORKSPACE", line 76, column 23, in <toplevel>
        python_init_toolchains()
    File "/__w/.cache/bazel/bazel/_bazel_root/a40625930fdef4f0a3483cd60aa1bb86/external/xla/third_party/py/python_init_toolchains.bzl", line 52, column 35, in python_init_toolchains
        python_register_toolchains(
    File "/__w/.cache/bazel/bazel/_bazel_root/a40625930fdef4f0a3483cd60aa1bb86/external/rules_python/python/private/python_register_toolchains.bzl", line 127, column 88, in python_register_toolchains
        (release_filename, urls, strip_prefix, patches, patch_strip) = get_release_info(platform, python_version, base_url, tool_versions)
    File "/__w/.cache/bazel/bazel/_bazel_root/a40625930fdef4f0a3483cd60aa1bb86/external/rules_python/python/versions.bzl", line 1048, column 25, in get_release_info
        release_id = int(u.split("/")[-2])
Error in int: invalid base-10 literal: "jax"

🌍 Your Environment

Operating System:

  
Linux x86
  

Output of bazel version:

  
7.4.1
  

Rules_python version:

  
1.6.0
  

Anything else relevant?

[rickeylev]

To add some context: In some cases, JAX uses python_register_toolchains(tool_version=...) and uses a url like file://whatever/python.tar.gz. Using a custom URL seems to trip up some later URL parsing logic.

IIRC, the url is allowed to have e.g. {version} placeholders, so it gets parsed as if it was a python-build-standalone URL. But, that should be optional -- the APIs let you specify an explicit url for this sort of thing, so it shouldn't require the PBS url format. As far as I can tell, it only wants to parse the URL to that it can determine some other parts of the URL it should be setting.