Regression in fix for #793? MatchSpec parsing appears to fail when channel is passed as url

[cisaacstern]

Checklist

• I added a descriptive title
• I searched open reports and couldn't find a duplicate

What happened?

Using the latest py-rattler==0.6.3 I am unable to parse MatchSpecs when the channel is passed as a url. In #793 (comment) the following example is given (copying for visibility here):

>>> from rattler import MatchSpec
>>> m = MatchSpec("https://conda.anaconda.org/conda-forge::python[version=3.9]")
>>> m
MatchSpec("conda-forge::python ==3.9")
>>> m.channel
Channel(name="conda-forge", base_url="https://conda.anaconda.org/conda-forge/")

However on my machine I cannot reproduce this. I have just done the following in an empty directory (on mac M1):

1. ✨ Create a new pixi project and add py-rattler

   $ pixi init
   $ pixi add --manifest-path pixi.toml py-rattler
   ✔ Added py-rattler >=0.6.3,<0.7

2. 🔧 Add test_matchspec.py to the project directory (adapted from the previously-mentioned example in MatchSpec parses incorrectly when channel is passes as a URL #793 (comment)):

   # test_matchspec.py
   
   from rattler import MatchSpec
   
   if __name__ == "__main__":
       with_url_channel = MatchSpec("https://conda.anaconda.org/conda-forge::python[version=3.9]")
       print(f"{with_url_channel = }")
       print(f"{with_url_channel.channel = }")
       with_named_channel = MatchSpec("conda-forge::python[version=3.9]")
       print(f"{with_named_channel = }")
       print(f"{with_named_channel.channel = }")

3. 🤔 Run the script. Note that the with_url_channel does not parse correctly (contrary to the linked example in MatchSpec parses incorrectly when channel is passes as a URL #793):

   $ pixi run --manifest-path pixi.toml -e default python test_matchspec.py
   with_url_channel = MatchSpec("*")
   with_url_channel.channel = None
   with_named_channel = MatchSpec("conda-forge::python ==3.9")
   with_named_channel.channel = Channel(name="conda-forge", base_url="https://conda.anaconda.org/conda-forge/")

For reference, here is the output of pixi list for this environment:

$ pixi list --manifest-path pixi.toml -e default    
Package          Version    Build               Size       Kind   Source
bzip2            1.0.8      h99b78c6_7          120 KiB    conda  bzip2-1.0.8-h99b78c6_7.conda
ca-certificates  2024.8.30  hf0a4a13_0          154.8 KiB  conda  ca-certificates-2024.8.30-hf0a4a13_0.conda
libexpat         2.6.3      hf9b8971_0          62.4 KiB   conda  libexpat-2.6.3-hf9b8971_0.conda
libffi           3.4.2      h3422bc3_5          38.1 KiB   conda  libffi-3.4.2-h3422bc3_5.tar.bz2
libsqlite        3.46.1     hc14010f_0          810.1 KiB  conda  libsqlite-3.46.1-hc14010f_0.conda
libzlib          1.3.1      hfb2fe0b_1          45.8 KiB   conda  libzlib-1.3.1-hfb2fe0b_1.conda
ncurses          6.5        h7bae524_1          783.5 KiB  conda  ncurses-6.5-h7bae524_1.conda
openssl          3.3.2      h8359307_0          2.7 MiB    conda  openssl-3.3.2-h8359307_0.conda
py-rattler       0.6.3      py312h3ddc590_1     3.9 MiB    conda  py-rattler-0.6.3-py312h3ddc590_1.conda
python           3.12.6     h739c21a_0_cpython  12.3 MiB   conda  python-3.12.6-h739c21a_0_cpython.conda
python_abi       3.12       5_cp312             6.1 KiB    conda  python_abi-3.12-5_cp312.conda
readline         8.2        h92ec313_1          244.5 KiB  conda  readline-8.2-h92ec313_1.conda
tk               8.6.13     h5083fa2_1          3 MiB      conda  tk-8.6.13-h5083fa2_1.conda
tzdata           2024a      h8827d51_1          121.3 KiB  conda  tzdata-2024a-h8827d51_1.conda
xz               5.2.6      h57fd34a_0          230.2 KiB  conda  xz-5.2.6-h57fd34a_0.tar.bz2

Is this user error on my part, or a regression, or something else?

Thank you all so much!

Additional Context

No response

[wolfv]

Argh, that looks like a bug / regression on first glance. Woudl you be able to whip up a PR with a test case?

[cisaacstern]

Thank for the prompt reply, and yes would be happy to. Will get that up shortly.

[cisaacstern]

Done in #870. As noted there, the regression is not present for me when running those tests locally (from a fresh build), so maybe the regression is already fixed in main and only present in 0.6.3 release?

[wolfv]

It's possible that we need to cut a new release of the Python bindings. I kicked off the CI run for your new tests.