Is an empty string a valid LegacyVersion?

[uranusjr]

From pypa/pip#9405 (comment)

Current pip allows a user to specify requirements like packaging==. packaging parses the part after == as a LegacyVersion containing an empty string. Whether this is a bug or not, depending on how far the “LegacyVersion can contain anything” idea is taken, but the behaviour is quite confusing to both users.

Since distutils and setuptools both use 0.0.0 if a package is built without specifying a version (or an empty one), it is reasonable to disallow LegacyVersion("") IMO.

[pradyunsg]

Let's do this. :)

[pfmoore]

Note that even if LegacyVersion("") is considered valid, I'd still consider there to be a question about whether a specifier of bare == should be allowed. PEP 440 states that "A version matching clause includes the version matching operator == and a version identifier", and it feels to me that == is missing the "version identifier" part. The PEP doesn't cover legacy versions (so in terms of the PEP an empty version is invalid), so it's not a conformance issue, but I feel that allowing a bare comparison operator goes against the spirit of the PEP.

Edit: Sorry, didn't refresh my browser, so I missed @pradyunsg's comment.

I agree, let's do it.

[pfmoore]

Actually, what's the timescale for the next major release of the project, because it appears that LegacyVersion is due to be removed (see here)...

[pradyunsg]

Anytime this year. ;)

packaging also follows a CalVer cadence.

[pfmoore]

More to the point, then, I guess there's no reason to disallow "" as a LegacyVersion, because the same release that does that will also remove LegacyVersion (or am I reading too much into the statement "creating a LegacyVersion has been deprecated"?)

[pradyunsg]

#321 (comment)

No, you're right. This project will drop Legacy Version very soon, and we probably don't need to fix this code before removing it.

[brettcannon]

More reason to drop LegacyVersion then! 😁

[hrnciar]

#321 (comment)

No, you're right. This project will drop Legacy Version very soon, and we probably don't need to fix this code before removing it.

Hello, I see packaging 21.0 was released in July, but Legacy Version is still present. May I ask why it wasn't removed yet?

[pradyunsg]

See #407 for the discussion. Basically, I didn't have time for the PR removing it, and that resulted in not removing legacy in the first release.

[hrnciar]

I see, thank you. I am asking because I am trying to write a parser for PEP 508 as discussed here. So I would like to know whether it should support Legacy Version or I can ignore these tests.

[pradyunsg]

Definitely ignore them. Legacy versions are non-PEP 508 versions.

[pradyunsg]

Could you file a new issue for the hand written parser for PEP 508 against this repo? It'll let us drop the dependency on pyparsing, which would be nice.

[hrnciar]

Could you file a new issue for the hand written parser for PEP 508 against this repo?

Sure thing, see #468