[Qgis-psc] Handling the Travis CI situation

[Nyall Dawson]

On Fri, 6 Nov 2020 at 20:57, Matthias Kuhn <matthias at opengis.ch> wrote:
>
> This would be desirable.
>
> It's unfortunately a bit of complexity included. As Alessandro said, the rendering tests are a good example of tests which are hard to maintain and - unfortunately - not well defined functions like we happen to have on e.g. database engine unit tests.
>
> To recall, when we started with CI some 5 years ago we were in the situation that we had a set of tests, which randomly passed on some dev machines and some on others. So effectively you had no chance to know if a test doesn't pass because of your machine or your patch.
>
> At least now we have a reference platform and know if a patch actually does cause changes. Which is the most important information.
>
> Currently many of the tests are already cross platform. And nobody is actively writing CI centric tests. It's just a matter of fact that the tests have a requirement to at least pass on the CI env because it's the only thing that can be enforced.
>
> When tests are updated to a new platform (ci or not), many of them will pass on a wider variety of platforms. Because bugs are fixed or they are made more generic. And some will just have reference images for one more platform added.
>
> So whatever we do we will do a step into the right direction. But there will be no guarantee that every test passes on your machine except if you make them all pass on your machine - which would be very welcome!!
>
> What you could also do is listing tests that are generic / platform independent (e.g. the expression tests would be a very good example) and then always just run these instead of the whole set of tests.


Just to add a few extra thoughts to the already comprehensive replies
given by Matthias and Denis:

- we had to do a very similar effort with updating existing tests when
we moved from Qt 4 -> Qt 5. We'll probably have to do another similar
effort in another 18-24 months or so. It's going to be a regular,
recurring task to go through and update all the tests which have been
introduced since the previous effort and ensure that they are
sufficiently tolerant to pass under different environments/software
versions. (Maybe we should consider adding "test maintenance" as a
regular yearly expense of the nature of 1.5 weeks?)
- we really should do a similar effort to get the existing tests
passing under mac os and windows too. I suspect there's some valid
bugs that the test suites would reveal if we could reliably run them
under windows/mac, but the real issues are drowned in the noise of
tests which haven't been designed to be cross platform compatible.
(This could be a good grant proposal idea for future funding rounds!)

And then my personal 2c:

We have a great test suite, and fantastic tools for making and
managing tests. Sure, there's a learning curve involved with them.
Sure, they ARE different to the test suites used by gdal, or geos, or
<insert other project here>. But that's just business as usual in
software development, and not at all reflective of inferiority in the
test suites. Can we please move on from this recurring point once and
for all and focus on the current relevant parts of this discussion
instead?

Nyall







>
> Matthias
>
>
> On Fri, Nov 6, 2020 at 11:29 AM Sandro Santilli <strk at kbt.io> wrote:
>>
>> On Fri, Nov 06, 2020 at 05:07:40PM +1000, Nyall Dawson wrote:
>>
>> > - different Qt versions on Github vs Travis
>> > - different underlying library versions (e.g. updated proj, gdal, and
>> > others vs the older Travis build environment)
>> > -  other differences in the build setup, e.g. different limits on the
>> > environment under which the tests run
>> >
>> > Before we can make the switch and move away from Travis, we'll need to
>> > update these tests and get as many of them passing as possible on
>> > Github, and then handle the tricky job of backporting the test fixes
>> > and Github action setup to the 3.16/3.10 branches.
>>
>> I'd recommend changing the tests to return success results with BOTH
>> versions of Qt, proj, gdal etc. The goal would be making the testsuite
>> pass on _developers_ machines, whatever libraries they are using.
>>
>> The CI-centric nature of QGIS testsuite has always been a problem for
>> me as it makes it impossible to check the quality of changes locally.
>>
>> There are likely existing tickets addressing these issues.
>> A 'testsuite' label exists. I'm not sure all the relevant tickets
>> have the label applied, but it could be a good starting point:
>>
>>   https://github.com/qgis/QGIS/issues?q=is%3Aopen+is%3Aissue+label%3Atestsuite
>>
>> --strk;
>> _______________________________________________
>> Qgis-psc mailing list
>> Qgis-psc at lists.osgeo.org
>> https://lists.osgeo.org/mailman/listinfo/qgis-psc