[Distutils] PEP 440 and prefix matching

[Nick Coghlan]

On 27 June 2013 19:16, Vinay Sajip <vinay_sajip at yahoo.co.uk> wrote:
> All the examples of explicit prefix matching in PEP 440 show only release
> clause components, but it's not explicitly stated whether that's
> specifically intended. For example, are prefix matching clauses such as
>
> == 1.2.dev0.*
> != 1.2.post1.*
>
> allowed?

Sure. I'm not sure why anyone would actually want to do it, but it's
allowed (disallowing it would actually make the spec more complicated
rather than simpler).

The only thing which explicitly ignores the suffix is the compatible
release clause: "If a pre-release, post-release or developmental
release is named in a compatible release clause as V.N.suffix, then
the suffix is ignored when determining the required prefix match"

The rationale for that restriction is that we're mostly using the same
compatible release semantics as semantic versioning, except that we
assume a reasonable level of backwards compatibility will exist even
for 0.x releases, leaving it to people to use prefix matching if a
dependency exposes a somewhat unstable API. Semver doesn't define
compatibility semantics outside the release number, so we don't
either.

It occurs to me, though, given that we now have an exact prefix
matching notation for more unusual forward compatibility constraints,
we *could* just make all compatible release clauses explicitly
semantic versioning based. Then you could say things like (2.2.2) or
(~= 2.2.2) to mean (>= 2.2.2, == 2.*). Under the current PEP, those
clauses would translate to (>= 2.2.2, == 2.2.*), which differs for no
good reason I can see from the handling of a clause like (~=
2.2.post1), which would translate as (>= 2.2.post1, == 2.*).

I think the semantic versioning based system is actually easier to
explain than the current scheme (mostly because it's more consistent),
so unless someone comes up with a compelling counterargument in the
next few days, I'll switch the specification to always use "== V.*" as
the prefix matching component of a compatible release clause, no
*matter* how many components there are in the version specifier
itself.

Cheers,
Nick.

--
Nick Coghlan   |   ncoghlan at gmail.com   |   Brisbane, Australia