Fix priority of platform tags for manylinux

[zanieb]

Closes #2477

See also:

• Fix priority of ABI tags #2489

[zanieb]

I fixed the ordering of the abi tags, but I'm not sure where it's documented that linux_x86_64 should take priority over manylinux_<abi>_x86_64. If someone has a link for that part of the specification I'd appreciate it — I can add it to the code.

Looks correct per https://github.com/pypa/packaging/blob/fd4f11139d1c884a637be8aa26bb60a31fbc9411/packaging/tags.py#L434-L444

[charliermarsh]

I think it might be the other way around, I think higher priority tags come first in that code, so linux_x86_64 would be last? That's at least my read of the code, but I didn't test it.

[charliermarsh]

You can test by comparing to:

from packaging import tags

for tag in tags.sys_tags(): print(tag)

[zanieb]

Apologies for all the back and forth apparently this needed a more rigorous look :)

I've addressed the last round of comments. I might look into a better way to test we match Python's behavior.

[zanieb]

It looks like our ordering of ABI tags is wrong, i.e. Python does

cp311-cp311-manylinux_2_28_x86_64
...
cp311-cp311-manylinux_2_17_x86_64
cp311-cp311-manylinux2014_x86_64
cp311-cp311-manylinux_2_16_x86_64
...
cp311-cp311-manylinux_2_12_x86_64
cp311-cp311-manylinux2010_x86_64
cp311-cp311-manylinux_2_11_x86_64
...
cp311-cp311-manylinux_2_5_x86_64
cp311-cp311-manylinux1_x86_64
cp311-cp311-linux_x86_64
cp311-abi3-manylinux_2_28_x86_64
...
cp311-abi3-manylinux_2_17_x86_64
cp311-abi3-manylinux2014_x86_64
cp311-abi3-manylinux_2_16_x86_64
...
cp311-abi3-manylinux_2_12_x86_64
cp311-abi3-manylinux2010_x86_64
cp311-abi3-manylinux_2_11_x86_64
...
cp311-abi3-manylinux_2_5_x86_64
cp311-abi3-manylinux1_x86_64
cp311-abi3-linux_x86_64
cp311-none-manylinux_2_28_x86_64
...
cp311-none-manylinux_2_17_x86_64
cp311-none-manylinux2014_x86_64
cp311-none-manylinux_2_16_x86_64
...
cp311-none-manylinux_2_12_x86_64
cp311-none-manylinux2010_x86_64
cp311-none-manylinux_2_11_x86_64
...
cp311-none-manylinux_2_5_x86_64
cp311-none-manylinux1_x86_64
cp311-none-linux_x86_64
cp310-abi3-manylinux_2_28_x86_64
...

but we do

cp311_cp311_manylinux_2_28_x86_64
cp311_none_manylinux_2_28_x86_64
...
cp311_cp311_manylinux_2_17_x86_64
cp311_none_manylinux_2_17_x86_64
cp311_cp311_manylinux2014_x86_64
cp311_none_manylinux2014_x86_64
cp311_cp311_manylinux_2_16_x86_64
cp311_none_manylinux_2_16_x86_64
...
cp311_cp311_manylinux_2_12_x86_64
cp311_none_manylinux_2_12_x86_64
cp311_cp311_manylinux2010_x86_64
cp311_none_manylinux2010_x86_64
cp311_cp311_manylinux_2_11_x86_64
cp311_none_manylinux_2_11_x86_64
...
cp311_cp311_manylinux_2_5_x86_64
cp311_none_manylinux_2_5_x86_64
cp311_cp311_manylinux1_x86_64
cp311_none_manylinux1_x86_64
cp311_cp311_linux_x86_64
cp311_none_linux_x86_64
cp311_abi3_manylinux_2_28_x86_64
...
cp311_abi3_manylinux_2_17_x86_64
cp311_abi3_manylinux2014_x86_64
cp311_abi3_manylinux_2_16_x86_64
...
cp311_abi3_manylinux_2_12_x86_64
cp311_abi3_manylinux2010_x86_64
cp311_abi3_manylinux_2_11_x86_64
...
cp311_abi3_manylinux_2_5_x86_64
cp311_abi3_manylinux1_x86_64
cp311_abi3_linux_x86_64
cp310_abi3_manylinux_2_28_x86_64
...

This is a different part of the code though, I'll address in a second pull request.

[charliermarsh]

Oh interesting… they might be mutually exclusive such that it hasn’t mattered? Hard to reason about from my phone :D Thank you for digging into this!

[zanieb]

I'm no expert, just mindlessly matching Python's behavior at this point. See #2489 for the ABI changes.

I'll wait for @konstin to review both of these.

[charliermarsh]

Great!

[charliermarsh]

I’m comfortable signing off on this if you want to merge.

[bdice]

Thanks, this looks right to me too.