[U-Boot] Travis CI

[york sun]

On 11/30/2016 04:39 PM, Tom Rini wrote:
> On Wed, Nov 30, 2016 at 09:55:52PM +0000, york sun wrote:
>> On 11/30/2016 01:43 PM, york.sun at nxp.com wrote:
>>> On 11/30/2016 01:35 PM, Stephen Warren wrote:
>>>> On 11/30/2016 12:47 PM, york sun wrote:
>>>>> Guys,
>>>>>
>>>>> I noticed you have been improving .travis.yml. Can you share some
>>>>> performance data when using it? I am still using Jenkins to verify my
>>>>> branches for every commit. Due to the limitation of computing power, I
>>>>> don't cover all targets. This has caused several "oops". Please share
>>>>> your experience. Thanks.
>>>>
>>>> The Travis results are all public. For example, you can look at the URL
>>>> below to see the build for Tom's github clone of the main u-boot.git:
>>>>
>>>> https://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Ftravis-ci.org%2Fu-boot%2Fu-boot&data=01%7C01%7Cyork.sun%40nxp.com%7C09609202d44942fb21e908d41968d98b%7C686ea1d3bc2b4c6fa92cd99c5c301635%7C0&sdata=gD64O8j98pt1nZZW39YCPWd7NDM4OcctfJG8pjBGsNc%3D&reserved=0
>>>>
>>>>
>>>> The last test run appears to have taken about 49 minutes total. IIRC Tom
>>>> mentioned 1-2 hours total in the past.
>
> Note that the "elapsed time" can become quite wrong if you go to a
> failed build, see that one of the sub-jobs failed for bad reasons (for
> example, sometimes we get a failure with the ppa we use for qemu), and
> restart just that part and it completes.  Most recently I see one took
> around 2.5 hours as I've added in more build targets since I gave the
> initial time.
>
> That said, those 2.5 hours build just about every buildable target in
> U-Boot, and run test.py via qemu for a few.  And it's totally free.  To
> me it's a great "last test" before sending out a PR or a big series.  I
> use it as part of my pre-pushing tests.
>
>>>> Travis uses github hooks to detect pushes, so I typically see it start
>>>> builds almost immediately after a push to an enabled branch.
>>>>
>>>
>>> Stephen,
>>>
>>> Thanks for the pointer. I am getting myself familiar with Travis CI. I
>>> have been using Jenkins. Compiling 100 commits for powerpc (418 targets)
>>> takes many hours to finish.
>>
>> I took a test run and found without the "-b", buildman only builds
>> current source tree, not for each commit. It is a good tool but maybe
>> this is not for my purpose.
>
> Yes, this is not a tool for building every part of a series.  I still
> use a different machine to build before/after for just about everything
> and look at size changes for example.
>

Thanks, Tom. I noticed Travis CI runs on low powered servers. It takes 
quite long to finish everything. But as you said, it is good to compile 
everything before pushing.

York