Hi. It would be great to get as many mesa fixes into buster as
possible, so I decided to review all your git commits. I have no idea
if the release team will find this useful or not, but there's no way I
could construct a useful argument in favor of the inclusion without the
review.
I am *not* a member of the release team; I do not speak for anyone but
myself.
I got a chance to review all the 18.3.6 commits and started on 18.3.5.
Release team, if this is useful, I'm happy to finish off reviewing the
remainder of the commits. Please explicitly let me know if that would
be valuable to you.
Based on the review so far, I find that the maintainer is probably
right. Namely, the upstream is very conservative in what they accept
with one exception I'll call out below. Unfortunately, mesa has a lot
of interdependencies, and the changes are inter-related. If upstream
actually does have a good CI infrastructure across a lot of hardware,
then cherry-picking changes is likely to be significantly less stable
than pulling in a full upstream release.
We can do nothing. We know we'll at least be suffering from a number of
crashes and graphics corruption issues if we do that.
We can take targeted fixes.
I think that's better than doing nothing.
However given upstream's conservative appropach, I actually think the
risk is lower if we take advantage of their test coverage and take all
or most of their release.
There's one area where upstream is not as conservative as we might like.
They have introduced a number of cases where additional linker or
compiler errors are added to shaders.
In general, they have turned cases where programs would previously build
but crash or produce bad graphics when run into build time errors.
>From our standpoint they have turned illegal input that would produce
graphical corruption or worse into ftbfs bugs or runtime compilation
failures depending on when the shader code is compiled.
Personally I think that's acceptable, but it's the thing I most want to
flag here. I've flagged such commits as linker (or compiler) change. I
think these commits could be reverted with less risk of interdependency
problem than other commits.
Changes I know we don't want:
97de8f48894844361d6d5e69137154741e114b74
Does not meet the freeze policy. I've seen the release team reject
packages for making similarly large changes to packaging at this
stage in the process.
Comments on commits:
600f314d63ad3d916f1eaa46953f86846668685e
Fixes use-after-free
d29dca6d
Must include in buster if 7536af is included in buster to fix
segfault
eac9b87
linker change
76719ecb
This is a crash so would probably be useful to get
96a01a5
I suspect that the bug this fixes will result in graphical
corruption
0f976b
linker change
208fd66
linker change
db517e
linker change
2c163bfe
fix for bug definitely causing graphics corruption
2a1e743e
Fixes segfault
