[PATCH] python-cffi: read version from notmuch version file

[Floris Bruynooghe]

On Tue 23 Jun 2020 at 13:43 +0300, Frank LENORMAND wrote:

> On Tue Jun 23 12:33:36 2020, David Bremner wrote:
>> Frank LENORMAND <lenormf.ml at gmail.com> writes:
>> > For example, 0.30.1, with the first two numbers coming from the main
>> > repository, and the last one acting as major for the bindings.
>> >
>> > 0.29.3 → 0.29.1
>> > 0.30-rc2 → 0.30.1-rc2
>> > etc.
>> >
>> 
>> I'm mainly interested in supporting two use cases for notmuch: building
>> everything from source, and binary packages of released versions. We've
>> already gone to some trouble to tell Emacs users that try to mix and
>> match versions that they are on their own, and this seems to apply even
>> more strongly to bindings users.
>>    
>> With that said, if Floris thinks some hierarchical version is useful,
>> and is willing to maintain it, I can live with it. I would ask that:
>> 
>> 1) You keep the whole "upstream" version number. So the first example
>> would be 0.29.3.1.  0.29.1 is a previous version of notmuch, and that
>> ambiguity can only cause trouble.
>
> The idea was that the bindings will work with the X.Y version they were
> released for, since the last component in X.Y.Z is for minor changes that
> shouldn't affect the API.

Minor nitpicking, but API is not strong enough here, you'd need to
ensure ABI compatibility.

> So we can keep X.Y from NotMuch itself, and append some information that
> hint at the state of the bindings.
[...]
> Or the exact same version number, but then what should happen to it when
> the bindings are modified, but not NotMuch?

If it was bad enough to need a new release then I guess everyone gets
the same version bump as the entire project gets a bugfix release?

I honestly like the simplicity of just having the same version number
and not having to think about maintaining it separately.  It also means
we mostly don't have to worry about how setuptools/pip is going to view
the version number.

The only way I think this could break is if we want to break backwards
compatibility in the bindings, but we're not supposed to do that
(realistically an impossible task in Python if you ask me, but we can
aim for at least avoiding doing this knowingly).

The most likely version number sin is that the python bindings get a new
feature while libnotmuch only gets bugfix.  I also don't think this is
terrible, but that's perhaps unusual and frowned upon.  Maybe this
warrants a README in the bindings to warn the version number just tracks
libnotmuch and as far as python goes can only be used to order the
releases.