On 22/11/2019 10:42, Paul Moore wrote:
> Paul Moore <p.f.moore@gmail.com> added the comment:
>
> PyPA member here - if this PR is defining new compatibility tags,
replacement platform_tag, not compatibility tag.
>  I would have expected it to need discussion as a revision to PEP 425, much like the manylinux efforts did. (It may actually be a closer parallel with MacOS, which didn't have a tagging PEP - I'm not sure how that would relate here, but as MacOS is a much more mainstream platform I'd be inclined to err on the side of caution and look for AIX to be explicitly covered in the tag specs).

As I understand the 'manylinux*' compatibility tags - which are used as
a replacement for platform tags - these compatibility tags are added on
by the packager who works in a carefully configured environment in order
to support binary compatibility over multiple Linux distributions and
releases.

Before AIX 6.1 binary compatibility in AIX was not guaranteed upfront.
Starting with AIX 6.1 IBM has a policy of guaranteeing binary
compatibility for applications.

From https://www.recarta.co.uk/our-partners/software-partners/ibm-aix/

This page, from IBM, speaks of binary compatibility even going back to
AIX 3.2 -
https://www.ibm.com/support/knowledgecenter/en/ssw_aix_72/install/binary_compatability.html

So, again - the PR is to provide an adequate platform tag. As is, an
adequate tag works on equality. Once the platform tag is established -
then - "we" (I guess that will mean me writing and "the community"
providing feedback) can look at how to use the tags over different
`releases` of AIX. Like MacOS, unlike Linux - there is only one distributor.

I am willing to work on an informational PEP - as long as waiting for
that process to complete does not mean an adequate platform_tag is left
in limbo for months to years. This new PEP would have two pieces - a)
describe the AIX platform tag; b) describe how tags of different
releases are compared and accepted or rejected as binary compatible.

Please note: the PR does not address any aspect of point b). Without
point a) there is no basis for point b).

Sincerely,

Michael

>
> I thought that was something that had been discussed on the Pip tracker, but maybe the implications weren't clear (I don't really understand the AIX scheme, so I'm relying on someone else, probably Michael, to propose something that could be added to PEP 425 and to lead/guide the discussion).
>
> ----------
> nosy: +paul.moore
>
> _______________________________________
> Python tracker <report@bugs.python.org>
> <https://bugs.python.org/issue38021>
> _______________________________________
>