Use more words to describe the graphs

[mattip]

Thanks for putting this together.

I am not sure what the terms mean. For instance, in the "Lowest (required) manylinux policy" graph, if I hover over Dec 25, 2024, I see "manylinux1 11.8% 99.9% overall". What do those numbers mean? Perhaps a textual description of the percentages could help.

[mayeut]

Thanks for your feedback, I'm not exactly sure how to word this/where to put this. If you have some proposals, I'd be happy to consider them. For this specific graph, which is probably of low interest for the vast majority of packagers:

For any given project producing manylinux wheels, each wheel will have a "required" policy and an "available" policy, e.g. manylinux2014_x86_64.manylinux_2_28_x86_64 has a manylinux2014 "required" policy and a manylinux_2_28 "available" policy. If the project is producing multiple wheels for x86_64, the "required" policy is the minimum "required" policy across those wheels and the "available" is the maximum "available" policy across those wheels.

As for the numbers, the first one is the percentage of projects, the second one is a cumulated sum. e.g. "manylinux1 11.8% 99.9% overall", 11.8% of projects have a manylinux1 "required" policy and 99.9% of projects have a "required" policy that's manylinux1 or higher.

[mattip]

11.8% of projects have a manylinux1 "required" policy and 99.9% of projects have a "required" policy that's manylinux1 or higher

So 11.8% are running exactly manylinux1 (i.e. glibc 2.5 - glibc 2.11 inclusive), and 99.9% are running something higher?

[mayeut]

So 11.8% are running exactly manylinux1 (i.e. glibc 2.5 - glibc 2.11 inclusive), and 99.9% are running something higher?

This is the package statistics section (i.e. producer side of things) so I wouldn't say "running" but "targeting" (and thus glibc 2.5 maximum in this specific exemple). 99.9% - rounding errors or some unaccounted for glibc version - are targeting glibc 2.5 or a higher glibc version.

[mattip]

Right. OK, so I understand the first line. The next line in the "Lowest (required) manylinux policy" is manylinux_2_5, which is the same as manylinux1, no? So why does it have "0%, 88.6% overall"? Likewise, the aliases for manylinux2010 and manylinux2014 are confusing.

[mayeut]

manylinux_2_5, which is the same as manylinux1, no?

While they are both targeting the same version of glibc, aliases are kept explicitly as the support for PEP600 might be lacking in some tools (or tools are not being updated by consumers), easily seen in consumer graphs for Python 3.6 & 3.7.

[mayeut]

Information about the graphs added in 33691ff4a053df7999db2e669cc4af884f9bf5ca

[mattip]

Thanks