A specifier of "==*" is not handled properly

[pfmoore]

This is somewhat useless, and I'm not entirely sure it's PEP 400 conformant, but the following fails:

>>> from packaging.specifiers import SpecifierSet
>>> from packaging.version import Version
>>> s = SpecifierSet("==*")
>>> s.contains(Version("1.0.0"))
False

This form of specifier is used in pip's test suite here. I'll fix the test by using a different specifier, but I think that packaging should either reject the specifier as invalid, or treat it as "always matching".

[di]

A specifier of ==* is not a valid Specifier:

>>> from packaging.specifiers import Specifier
>>> Specifier("==*")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/Users/dustiningram/git/pypa/packaging/packaging/specifiers.py", line 112, in __init__
    raise InvalidSpecifier("Invalid specifier: '{0}'".format(spec))
packaging.specifiers.InvalidSpecifier: Invalid specifier: '==*'

Instead when used in a SpecifierSet it becomes a LegacySpecifier:

>>> from packaging.specifiers import SpecifierSet
>>> s = SpecifierSet("==*")
>>> s._specs
frozenset({<LegacySpecifier('==*')>})

Which means that a Version will never be contained in it, but a LegacyVersion will, but only if the version is an exact string match:

>>> from packaging.version import LegacyVersion
>>> s = SpecifierSet("==*")
>>> s.contains(LegacyVersion("*"))
True

PEP 440 says:

The specified version identifier must be in the standard format described in Version scheme, but a trailing .* is permitted on public version identifiers as described below.

where the minimum valid version is a single-digit release segment, so I would expect that particular test to use ==7.* (or ==7.1.*) instead.

[pfmoore]

Bah. Legacy stuff. I didn't even realise there was a LegacySpecifier class (I did know about LegacyVersion).

Thanks for the clarification. I had indeed changed that test as described.

I wonder whether pip's code is being inconsistent in where it allows "legacy" versions of stuff like this. Certainly we have at least one code path that was accepting a version string of "7" as matching a specifier string of "==*". Never mind, my immediate issue is addressed, and there's no bug in packaging.

[pradyunsg]

I wonder whether pip's code is being inconsistent in where it allows "legacy" versions of stuff like this.

It's not -- we just allow it everywhere. :)

[pfmoore]

Nope, the triggering case would have worked then. I think the new resolver code uses Version (i.e., non-legacy) in some places. But this is both off-topic here, and not really that important anyway 🙂