Travis Vector_Copy_Elision_Level

[Navneet Potti <na...@gmail.com>]

I noticed that we only run Travis with copy elision levels set to the two non-default cases. The default case, which is also the only one that’s turned out to be useful in the benchmarks, is not being tested at all. Does anyone know why? Any objections to changing that? 

Re: Travis Vector_Copy_Elision_Level

[Navneet Potti <na...@gmail.com>]

If we only have resources for testing two, I believe one should be the default level. That’s the one that matters most. 

The most time consuming build would be good to add if we become more resource efficient or have more resources in the future. But since it’s never useful in benchmarks, why bother with it now?

> On Jun 8, 2016, at 21:26, Zuyu Zhang <zz...@pivotal.io> wrote:
> 
> We could add the other two options if all CI tests finish in a reasonable
> time frame, say 2 hours.
> 
> This is a trade-off between the development and the tests.
> 
> Historically, the CI takes too long to finish due to the limited concurrent
> jobs, so we choose to build two copy elision level options: the least and
> the most time consuming one.​

Re: Travis Vector_Copy_Elision_Level

[Zuyu Zhang <zz...@pivotal.io>]

We could add the other two options if all CI tests finish in a reasonable
time frame, say 2 hours.

This is a trade-off between the development and the tests.

Historically, the CI takes too long to finish due to the limited concurrent
jobs, so we choose to build two copy elision level options: the least and
the most time consuming one.​

Re: Travis Vector_Copy_Elision_Level

[Zuyu Zhang <zz...@pivotal.io>]

Hi Nav,

Please go ahead with all the four options as a trial. We've solved all the
CI issues.

Cheers,
Zuyu

Re: Travis Vector_Copy_Elision_Level

[Navneet Potti <na...@gmail.com>]

OK, I’ll put in a PR once the clang/gcc failures are resolved. 

> On Jun 8, 2016, at 21:43, Jignesh Patel <ji...@pivotal.io> wrote:
> 
> Great idea Nav. I have no objections. 
> 
> Cheers,
> Jignesh 
> 
>> On Jun 8, 2016, at 8:45 PM, Navneet Potti <na...@gmail.com> wrote:
>> 
>> I noticed that we only run Travis with copy elision levels set to the two non-default cases. The default case, which is also the only one that’s turned out to be useful in the benchmarks, is not being tested at all. Does anyone know why? Any objections to changing that?
> 

Re: Travis Vector_Copy_Elision_Level

[Jignesh Patel <ji...@pivotal.io>]

Great idea Nav. I have no objections. 

Cheers,
Jignesh 

> On Jun 8, 2016, at 8:45 PM, Navneet Potti <na...@gmail.com> wrote:
> 
> I noticed that we only run Travis with copy elision levels set to the two non-default cases. The default case, which is also the only one that’s turned out to be useful in the benchmarks, is not being tested at all. Does anyone know why? Any objections to changing that?