On Fri, 02 Aug 2024 at 21:03:05 +0200, Niels Thykier wrote:
> I think it is more of a question whether [artifacts in /tmp] will be supported
> initially (might require a per-source TMPDIR too for buildd support, so
> artifacts does not get tainted because the buildd was running two builds at
> the same time, where once had ICE errors and the other had not).
If it's really $TMPDIR, then that's easy to implement. The $TMPDIR
could even be inside the build directory, like you would often want to
do in Gitlab-CI to avoid any weird effects of /tmp being on an overlayfs
(the build directory is typically mounted as a volume rather than being
part of the container image, so commands like rm -r --one-file-system
in the build directory work as you'd expect, which as I've found out
from experience is not necessarily true for /tmp). It's a pity the name
debian/tmp is already in use for something different!
If /tmp is hard-coded, then I think there's going to be no substitute
for the sbuild-equivalent mounting a unique /tmp for each build, which
shouldn't be difficult to achieve either.
smcv
