[petsc-dev] more automatic notification of errors in nightly builds

[Barry Smith]

> On Mar 4, 2015, at 10:09 PM, Tobin Isaac <tisaac at ices.utexas.edu> wrote:
> 
> On Wed, Mar 04, 2015 at 06:00:55PM -0600, Barry Smith wrote:
>> 
>>  Toby,
>> 
>>    Wow, this is great! Thanks  (I pushed some trivial changes)
>> 
>>   Satish,
>> 
>>      Can you rig this up to run each morning after 9 am on the next builds automatically sending the email.
>> 
>>     We can do more fine tuning of the script as time goes by but let's get it out there and tested.
> 
> I would also make the `-ignoreNote` option default: gcc-like warnings
> often include `note:`s pointing to other lines that don't have any
> faults in them, and I didn't come up with a good way of attaching
> these to the original warnings.

  I had hoped by leaving them there it would motivate us all to improve the script to handle them properly :-)

> 
> Also IMO sending the emails automatically could be counterproductive.
> If a warning caused by John is misattributed to Jane, only she gets
> the email, and she gets it every day until she (a) brings it to the
> attention of someone who can fix it, or (b) starts ignoring the
> emails, which defeats their intended purpose.

   Could be, but we won't know without trying for a while. And even if it makes Matt fix 50% of the plex problems it will have served its purpose :-)

Barry

> 
>  Toby
> 
>> 
>>   Thanks
>> 
>>     Barry
>> 
>>> On Mar 4, 2015, at 11:17 AM, Tobin Isaac <tisaac at ices.utexas.edu> wrote:
>>> 
>>> On Mon, Mar 02, 2015 at 02:05:49PM -0600, Barry Smith wrote:
>>>> 
>>>>> On Mar 2, 2015, at 1:59 PM, Tobin Isaac <tisaac at ices.utexas.edu> wrote:
>>>>> 
>>>>> 
>>>>> Slightly off topic: if/when the jenkins-ci testing gets off the
>>>>> ground, can we pipe compiler warnings through git blame to automate
>>>>> sending messages to responsible parties?
>>>> 
>>>>  This is independent of the jenkins-ci testing. If you provide the logic this can be done now in nightly builds. I would welcome this logic.
>>> 
>>> I've got a branch 'tisaac/check-build-blame' that puts this logic in
>>> checkBuild.py.  I made local copies of last nights logs and tested
>>> them out with
>>> 
>>> ./bin/maint/checkBuilds.py -blameMail -ignoreNote -ignoreDeprecated
>>> 
>>> It doesn't send the mail, but it generates the messages for review.
>>> There are some obvious flaws with this approach (Karl gets blamed for
>>> running uncrustify, the mswin compilers choke on a lot of things), but
>>> otherwise it works about as expected.
>>> 
>>> Toby
>>> 
>>>> 
>>>> Barry
>>>> 
>>>> 
>>>>> 
>>>>> Toby
>>>>> 
>>>>> On Mon, Mar 02, 2015 at 01:21:39PM -0600, Barry Smith wrote:
>>>>>> 
>>>>>> ---- WARNINGS ----
>>>>>> /sandbox/petsc/petsc.clone-2/src/dm/impls/plex/plexsubmesh.c:1575:17: warning: absolute value function 'abs' given an argument of type 'PetscInt' (aka 'long') but has parameter of type 'int' which may cause truncation of value [-Wabsolute-value]
>>>>>> 1 warning generated.