Release policy

[kyamagu]

Currently, skia-python follows the release policy that publishes beta until the majority of the tests passes, thanks to @HinTak 's tremendous effort. I start to think about the drawback of this release policy given the recent CI issue on maintaining the old m87 release.
#314 (comment)

Perhaps we can relax the release policy to allow some failing tests, so that we don't need to keep the old skia package. This can be justified by the release cycle of the upstream skia, which often brings breaking changes and has no stable release.

[HinTak]

I think this might be a necessity now, if/when people ask for arm windows builds. That's definitely very far from m87. Arm-based windows probably did not exist then!

[HinTak]

Arm windows builds are more complicated than arm linux builds, as Microsoft don't ship any version of gn/ninja. That means we'll either have to build them (like we do aarch64 linux until the later part of #314 where we switch to use vendor-shipped ninja/gn), or patch m87 to download and use more recent ninja/gn binaries from Google...

[HinTak]

@kyamagu at the moment, I think my decision would be to declare the m1xx release that comes with arm windows support (if numpy etc releases arm windows support soon so we can give that a go) as the next stable, and m87 release with some arm related updates and with or without also possibly arm windows support as the last of m87. Despite the known small incompatibility / non-upgradability.

I.e. The next m13x and m87 will both switch to native aarch64 linux builds, to keep them in sync. When arm windows support get added to m13x (we are waiting for numpy to do that first), we'll make some attempt at putting that in m87, too, and make another m87 release, whether we can add arm windows support or not, and declare that the last. Adding arm windows support to m7 (whether successful or not) will be some effort, so that will be the end of m87.

Since arm windows did not exist in m87's time, whether successful or not, using that as a stop seems to be a good clean cutting line.

Cc @pavpanchekha

[HinTak]

Don't know how soon numpy can add arm windows support, but we are still a month or two from m138 yet :-). I'll put some announcement/warning out on the usual open-source places to get some feedback?

[HinTak]

In short, the first m13x with arm windows will be the next stable; and there is one more m87 releases to try to keep the build machinery in sync (with or without adding arm windows support), and that will be the last of m87.

[pavpanchekha]

Seems super reasonable to me.

[HinTak]

@kyamagu @pavpanchekha in case you are watching - numpy merged my arm windows build matrix addition earlier, and have just released 2.3.0rc1 with it. (2.3.0 proper is due late June/early July). So I am trying to get skia-python to build on arm windows now. It seems appropriate to call skia-python v138 - possibly in a week or so, an "rc" too, with v140, which should be late July, or maybe v139 which might be after numpy 2.3 0, as next "stable".

v138 is turning out to be a lot of changes / additions.

[HinTak]

Come to think of it, doing a v138.1 as stable might be better, as a continuation of the "even number" release trend lately.

[pavpanchekha]

Any kind of stable release works for me. I'll switch the WBE website over as soon as both skia-python and Python itself switch over. (This way, y'all don't need to bother with obsolete releases any more once a new stable is out and I drag out the switch-over date to at least October and possibly later.)

[HinTak]

Latest m138: 2199 passed, 104 skipped, 18 xfailed, 26 warnings
M87: 2173 passed, 27 skipped, 8 xfailed

M87 went +2 from two new EGL/GLX tests from 2171.

So about 100 new tests have been added and ~70 old one s skipped. The warnings are mostly about Font("").

[HinTak]

My private repo has 2229 / 98 / 20 / 37 . I think the extras are with extra test files.