Re: lode's "add a mode to consider autogen.sh failure 'ok'"

On 30/11/2018 10:02, Markus Mohrhard wrote:
> On Fri, Nov 30, 2018 at 9:18 AM Stephan Bergmann <sbergman@redhat.com 
> <mailto:sbergman@redhat.com>> wrote:
>
>     Anybody remember what the lode repo's
>     
> <https://gerrit.libreoffice.org/plugins/gitiles/lode/+/1d0e18c1ce4bf703efe1266db6ce692ae0e4bafd%5E!/>
>
>     "add a mode to consider autogen.sh failure 'ok'" is supposed to
>     achieve?
>        AFAICS, what it does now is consider "Random Config Linux builder"
>     builds like
>     <https://ci.libreoffice.org/job/lo_tb_random_config_linux/1633/> as
>     successful, even thought they fail in the configure step:
>
>     [...]
>      > checking for gperf... /usr/bin/gperf
>      > checking whether gperf is new enough... configure: error: "GNU
>     gperf 3.0.4" is too old or unrecognized, must be at least gperf 3.1
>      > Error running configure at ./autogen.sh line 296.
>      > No emails were triggered.
>      > Finished: SUCCESS
>
>
>
> The Random Config Job takes random configure values and we consider a 
> combination that configure rejects as a success. The plan behind the job 
> is that either configure rejects a combination or the build needs to 

Ah, I'd thought that the fuzzing logic in configure.ac would "filter 
out" all fuzzed configuration requests that would be rejected (i.e., 
that the "Random Config Linux builder" autogen.sh step should never fail).

> succeed. It is unfortunate that during the update of the underlying VM 
> we see potential flase positives right now but we can just disable the 
> job until the machine has been updated.

I don't really mind about the temporary false positives.  Rather than 
temporarily disabling this job (or any other Jenkins jobs on gandalf, 
which currently all fail), I'd rather like to see that someone updated 
that machine accordingly (as I'd requested in a mail to what I 
understand are all the maintainers of that machine).