On 11/09/2016 10:11 AM, Lars Tangvald wrote:
Forgot to add in the bug.
On 11/09/2016 10:01 AM, Lars Tangvald wrote:
On 11/09/2016 08:59 AM, Lucas Nussbaum wrote:
On 09/11/16 at 08:17 +0100, Lars Tangvald wrote:
On 11/07/2016 03:43 PM, Lucas Nussbaum wrote:
Hi,
I don't think it's random: the rebuild is automatically retried
when it
fails. However, maybe a change in another package fixed it. If you
have
a full build log, maybe you could diff it with me to see if a
build-dep
changed, which could explain the result?
I plan to do another archive rebuild soon (probably over next
week-end).
I'll double-check at this point.
Lucas
I was able to reproduce the failure:
It happens when the test is run as root. Since we haven't had this
failure
before in the normal system, have you changed anything else about
how the
package is built?
Nothing I know about, no.
Lucas
5.7.16 was uploaded and seemed mostly fine, though I'm not too clear
on how to read the results. Should be able to diff the log, though?
https://buildd.debian.org/status/logs.php?pkg=mysql-5.7&ver=5.7.16-1&suite=sid
I'm fairly sure the issue is with the environment for the test run.
It tries to log in as an anonymous user, but if the login command is
run as root it will instead try to log in as 'root'@'localhost',
which will fail. So I don't think updating to 5.7.16 will help.
I'll see if I can reproduce with a full sbuild run locally, but we
might consider disabling the test pending it being reviewed upstream.
Another potential cause for this starting to happen now is that we
recently fixed an issue in the build script where test failures
wouldn't cause build failures, so this test might have failed in
similar runs before (at build time) without it being noticed.
I've made a patch for this test so it can run as root, and will upload
it to the git once I've built and tested so you can verify that it
resolves the issue.
I don't think this is a critical issue, since it doesn't affect the
package build system currently in unstable, and there is something in
the build environment causing it, so not sure if we need to upload a fix
right away or if it can wait?
Could it be related to sbuild version? I have 0.67, your failed build
uses 0.7, while the Debian 5.7.16 build uses 0.65
Bit of a stretch, only somewhat related thing I can see in the changelog
is (0.69):
- Set APT::Sandbox::User=root when running apt-get source in the
build-procenv autopkgtest, to avoid stderr noise from current apt.
--
Lars
