uucore: leading zeros are ignored in version compare

[shinhs0506]

should fix tests/misc/sort-version.sh

[github-actions]

GNU testsuite comparison:

Congrats! The gnu test tests/misc/sort-version is no longer failing!

[tertsdiepraam]

Have you seen the previous discussion on this? In particular this thread: #4325

[shinhs0506]

@tertsdiepraam Ah, i totally missed that. thanks for pointing it out.

I was using this as a reference for the non-stable sorting. And i guess stable sorting just preserves the relative order in case when the values are equal?

[tertsdiepraam]

Indeed. I'm not sure how intentional that is on GNU's side. We could start a discussion there too, because it is (in my opinion) a bit strange.

[shinhs0506]

they do specify --stable here. Would make sense if they wanted something deterministic.

[tertsdiepraam]

Interesting, I hadn't seen that sentence. That makes sense if it's about the last resort comparison. We should do the same then!

[github-actions]

GNU testsuite comparison:

Congrats! The gnu test tests/misc/sort-version is no longer failing!

[shinhs0506]

@tertsdiepraam should be ready for review

[tertsdiepraam]

This does not fix the issue for both stable and unstable does it? I would imagine that version_cmp needs to stake a stable flag that defines the behaviour. Or do you think it should he fixed in some other way?

[shinhs0506]

In my opinion, the comparison function should return Cmp::Ordering::Equals if they are equal. And the Caller should handle the equal case as it is done here

And i believe this pr fixes both stable and unstable case

[tertsdiepraam]

Returning equal does make sense. How does this solve both cases, though?

Edit: Let me clarify. sort does the right thing now, but ls doesn't do the last resort comparison, so it would be broken now right?

[shinhs0506]

I can't find anything on how ls breaks ties. Nothing will necessarily be 'broken' because it would still sort correctly in a non-deterministic way which makes sense because ls doesn't do last resort comparison as you said.

If we wanted to provide our own rules, and make ls behave similar to sort, that would be a reasonable approach too.

[tertsdiepraam]

Yeah not really broken, but it would be something we might need to fix. That could just mean that we need to open an issue, but we do need to know. Could you do some testing with ls with GNU and uutils before and after this patch?

[shinhs0506]

I did some manual testing and looks like GNU's ls also behaves like sort when breaking ties.

[github-actions]

GNU testsuite comparison:

Congrats! The gnu test tests/misc/sort-version is no longer failing!

[sylvestre]

Thanks :)
Still an improvement over the previous state :)