Index | Thread | Search

From:
George Koehler <kernigh@gmail.com>
Subject:
Re: Question RE: Failed bulk build standard practice
To:
"izder456 via ports" <ports@openbsd.org>
Date:
Tue, 19 Mar 2024 01:55:57 -0400

Download raw body.

Thread
On Wed, 13 Mar 2024 11:56:31 -0500
"izder456 via ports" <ports@openbsd.org> wrote:

> I take MAINTAINER on several ports with no ONLY_FOR_ARCH set.
> When looking through the bulk build reports in the ports@ lists this
> morning, I saw that some ports I worked on have failed on hardware I
> don't have access too.
> 
> Is there a special practice we do when this occurrs, or is the port
> just never built into binary at release for that arch?

We have a few ports that fail in every bulk (on some arch) and just
never become packages.

We have both ONLY_FOR_ARCHS and BROKEN-<arch>.  For example, lang/ldc
has ONLY_FOR_ARCHS = amd64, and lang/gambit has BROKEN-riscv64 (so
everything but riscv64 tries to build gambit).  I see that gambit
failed on powerpc, so I might want to add BROKEN-powerpc.

Some failures should not get BROKEN-<arch>.  I once saw lang/sbcl
fail in a powerpc bulk, but succeed on my powerpc at home.  I wanted
the next powerpc bulk to retry sbcl, so I didn't add BROKEN-powerpc.

Some failures are easy to fix (like adding -lm to fix undefined
"sqrt").  This might need someone with the right hardware to check
whether the diff is a fix.  If you mail a diff to ports@ then someone
might check it.

Some ports never get fixed on some archs, because too few people use
the port, and none of them run the failed arch.
--gkoehler