As a stop-gap measure, can we all run the debug build on our local machine
before opening a PR? Harshad makes a good point about getting into a mode where
the DEBUG ASSERTS are never checked.
I’d also suggest that we make a note of this in each PR description so the
reviewer does not have to ask.
Comments?
On 11/17/16, 10:24 AM, "Harshad Deshmukh" <hars...@cs.wisc.edu> wrote:
One option to lower the compilation time is to reduce the template code
explosion. Based on the discussion on the PR:
https://github.com/apache/incubator-quickstep/pull/129 it seems that
removal of a storage format is on the cards. I hope it helps in avoiding
the Travis timeout.
On 11/16/2016 10:08 PM, Harshad Deshmukh wrote:
> Hi Zuyu,
>
> Completely disabling the debug builds is not ideal, in my opinion. A
> lot of DEBUG ASSERT statements, which are potentially quite useful in
> debugging issues, only fire up in debug builds. I see from some of the
> previous pull requests that not all debug configuration run for 50+
> minutes. A reasonable trade-off could be to disable only those debug
> builds that run for 50+ minutes.
>
> In any case, it is worthwhile to bring down the build time. It is also
> better from developer productivity point of view. Any
> efforts/suggestion in this regard are highly welcome.
>
>
> On 11/16/2016 09:56 PM, Tony wrote:
>> Hi,
>>
>> I would like to discuss how to avoid the Travis CI timeout issue in both
>> the master branch and the PRs.
>>
>> The problem is that now Travis CI precisely kills all jobs running
>> for 50
>> minutes, but unfortunately two GCC debug builds run about 1 hour.
>>
>> I would suggest that the Travis CI only tests for the release builds so
>> that we won't have such issue, as shown in the following PR.
>>
>> https://github.com/apache/incubator-quickstep/pull/135
>>
>> Any comments?
>>
>> Cheers,
>> Zuyu
>>
>
--
Thanks,
Harshad
