[squeak-dev] Re: What's up on build.squeak.org

Frank Shearar frank.shearar at gmail.com
Wed May 28 20:26:34 UTC 2014


On 28 May 2014 21:06, Frank Shearar <frank.shearar at gmail.com> wrote:
> On 25 May 2014 21:52, Tobias Pape <Das.Linux at gmx.de> wrote:
>>
>> On 25.05.2014, at 22:43, Nicolas Cellier <nicolas.cellier.aka.nice at gmail.com> wrote:
>>
>>>
>>> 2014-05-25 18:34 GMT+02:00 Nicolas Cellier <nicolas.cellier.aka.nice at gmail.com>:
>>>
>>> 2014-05-25 2:39 GMT+02:00 Chris Muller <asqueaker at gmail.com>:
>>>
>>> Hi, thanks for noticing and investigating this!  Seeing this now, I
>>> think we should signal an Error rather than a Warning to be more
>>> TestCase friendly but also because persisting empty packages is so
>>> painful, it should be an error, hands down.  Better to force a
>>> resolution to the issue than silently persist modules of
>>> future-pain...
>>>
>>>
>>> In any case, I think you accidently uncovered bugs in Tests-Monticello.
>>> I'm looking at a fix for a few hours, and it's really messy.
>>>
>>>
>>> Hooray! After publishing Tests-nice.297 there is now a trunk CI job that finished
>>> http://build.squeak.org/job/SqueakTrunk/857/
>>>
>>> Now we can see we have some regressions...
>>> How could we live without CI before?
>>>
>>> Of course, for dissecting which change introduced which regression after a 2 month interrupt, that's going to be more pain than necessary...
>>>
>>
>> To not make that happen again, can we make the jenkins mail squeak-dev on
>> important (read me, not all) information?
>> Like:
>>         more tests fail
>>         errors
>>         (possibly one nightly information)
>
> We possibly could, but I'd suggest _not_ doing so because even without
> Nicolas's work (thanks very much, Nicolas!) we _normally_ have failing
> tests, which means that the normal status of the SqueakTrunk build
> would not be green.
>
> Really, the ideal place to be is for the job to usually be green,
> because then CI saying anything at all is cause for alarm, not yet
> another false positive. And we're just not there, on a number of
> fronts.

I should add that it doesn't help that build agents that run for a
long time - Java processes - eventually run out of PermGen space and
stop working. I've pinged the owner of two of the slaves - thanks,
Tony! - and we're tentatively trying out a hack so disgusting I'm not
going to talk about it.

frank

>> best
>>         -tobias


More information about the Squeak-dev mailing list