Re: Daily build compiler

Jonathan Perkin <[email protected]> writes:

> * On 2026-06-15 at 12:47 BST, Thomas Klausner wrote:
>
>>On Sun, Jun 14, 2026 at 02:05:59PM +0100, Jonathan Perkin wrote:
>>> * On 2026-06-14 at 01:28 BST, Jonathan Perkin wrote:
>>>
>>> > URL: https://reports.pkgci.org/SmartOS/upstream/trunk/20260613T194510Z/report.html
>>> >
>>> > Total:      29084  Platform:   SmartOS 20260122T220528Z/x86_64
>>> > Successful: 20528  Compiler:   gcc-15.2.0
>>> >
>>> > Build Failures                               Breaks  Maintainer
>>> > ----------------------------------------------------------------------------
>>> > netpbm-11.02.09nb8                             1383  [email protected]
>>> > ghc-9.10.1nb7                                   641  [email protected]
>>> > python27-2.7.18nb22                             246  [email protected]
>>> > pari-2.3.5nb33                                  243  [email protected]
>>> > ocaml-dune-3.23.1                               169  [email protected]
>>> > SDL-1.2.15nb47                                  162  [email protected]
>>> > php56-5.6.40nb5                                 151  [email protected]
>>> > cyrus-sasl-2.1.28nb2                            124  [email protected]
>>> > openjdk8-1.8.482nb2                             107  [email protected]
>>>
>>> There doesn't seem to be a huge amount of interest in fixing up the GCC 15.x
>>> issues exposed by these builds.  Should I go back to GCC 14.x so that we
>>> have higher package coverage for now?
>>>
>>> Happy to go back to 15.x once NetBSD is closer to moving to it.
>>
>>I'd prefer going back to GCC 14 for now, yes please.
>
> Ok, let's leave it a few days while nia is taking a look at things,
> and then I'll switch it back so there's still time during the freeze
> to look at anything that is specific to GCC 14.

I view gcc 15 as bleeding edge (especially the unreasonable choice to
default to c23) and don't think it's good to address gcc15 issues during
freeze, at all, because those changes aren't going to improve the branch
for branch users.  So I think you should switch back right away.

Longer term, I think we should separate actual bugs surfaced by gcc15,
and problems from gcc15 defaulting to c23.  I view it as fundamentally
broken for compilers to default to such new standards, and I think it
would make more sense to change the wrappers so that invocations without
a std (themselves buggy I think) end up with a more reasonable default
std, one that we can confidently say "if you don't ask for something
specific, and you can't build with X, you're wrong".  That's true for
c99, and arguably for c11, but I don't think it's true for anything
newer.