A list for the developers of CellML tools

Text archives Help


[cellml-dev] buildbot update: from now on, please pay attention to buildbot status changes


Chronological Thread 
  • From: ak.miller at auckland.ac.nz (Andrew Miller)
  • Subject: [cellml-dev] buildbot update: from now on, please pay attention to buildbot status changes
  • Date: Thu, 23 Jul 2009 11:21:51 +1200

Alan Garny wrote:
> Hi Andrew,
>
> First thanks a lot for this. I have made a few comments below.
>
>
>> The CellML buildbot is now succeeding on all builds except the Win32
>> Java build (meaning that the nightly overall build also shows up as red).
>>
>> Our build system automatically keeps track of any changes pushed to the
>> CellML API or OpenCell, and does an incremental build of the trunk after
>> a period of two minutes with no further changes has passed. Because we
>> mirror the repository locally, it can take a few minutes for changes to
>> propagate through and the no-change countdown to elapse.
>>
>
> 1) Shouldn't we, for political reasons, separate the CellML API and OpenCell
> build systems? As it stands, it gives the impression that OpenCell and the
> CellML API projects are tightly coupled and, most importantly, that OpenCell
> is *the* CellML tool by excellence. Whether any of these is true or not is
> irrelevant here, but it nonetheless remains that we shouldn't give that kind
> of impression.
>
I think that all it indicates is that both share development resources,
and for the most part, people, at the Auckland Bioengineering Institute.
Only people involved in development of the CellML API and OpenCell are
likely to use the buildbot, and they will understand the relationship
between OpenCell and the CellML API, so I don't think this is a huge issue.

> 2) I am a bit unclear about when incremental builds are done. Are you saying
> that they are done 2 minutes after a push has been detected and assuming
> that no additional push has been made in those 2 minutes?
>
Yes.
> 3) Why do an incremental build and not a complete build? I remember
> (personal) cases where an incremental build would succeed, but then the
> generated code would fail/crash, while a complete build would be fine. I
> agree that a complete build takes longer, but it would be safer. If
> anything, maybe it's something that could be done in addition to incremental
> builds and at less frequent intervals?
>
This is in fact what happens - nightly builds are complete builds, while
incremental builds happen as things are checked in. We do need to
improve our dependency system - it is a bug in the build system if
incremental builds don't do the right thing, although not necessarily a
large enough one to warrant spending much time on, as it is so easily
worked around.

Releases are based on clean builds, the purpose of the incremental build
is to check for problems.

>
>> The buildbot will send an e-mail if a previously working platform stops
>> working. Please pay attention to these e-mails and fix any problems you
>> introduce promptly (or backout your changes and recommit them when you
>> get it working). This will ensure that we don't get complex regressions
>> in platform support involving multiple different changes.
>>
>
> 4) Do such emails mention who is responsible for the failure (i.e. the
> person who did the last push)? I think that would save people a bit of time
> (e.g. if I am not responsible then I could 'ignore' the email... :)).
>
It does list people who have pushed since the last build in the e-mail.
This won't always be reliable (e.g. with build system problems, which
might take longer to come up). Obviously, some failures will result from
things other than code commits.

>
>> Please monitor the status of the builds at:
>> http://autotest.bioeng.auckland.ac.nz/cellml-build/waterfall
>> If you have access to the old Subversion repository, the same access
>> credentials may be used to manually trigger builds. Log in at:
>> https://autotest.bioeng.auckland.ac.nz/cellml-admin/waterfall
>> to access that functionality.
>>
>
> 5) I wish the builds were on the 'Y axis' and the timings on the 'X axis'. I
> believe this would allow to see all the different builds at once (at least
> for those of us who have a big enough screen resolution) and one would just
> have to scroll horizontally to see their results through time. At the
> moment, one has to scroll both horizontally and vertically.
>
There are different views, but that is the only one with both timings
and builders. This one shows the latest status of all builders in one
screen:
https://autotest.bioeng.auckland.ac.nz/cellml-admin/one_box_per_builder

Best wishes,
Andrew

> Alan
>
> _______________________________________________
> cellml-tools-developers mailing list
> cellml-tools-developers at cellml.org
> http://www.cellml.org/mailman/listinfo/cellml-tools-developers
>





Archive powered by MHonArc 2.6.18.

Top of page