9:40 a.m.
On Mon, Jul 08, 2019 at 02:17:44PM +0200, Vít Ondruch wrote:
I am skeptical about this proposal. While this might work for your
package, I am afraid it won't work generally and trying to do something
like this is wasted energy. Let me explain.
well, that may be fair, but when you way it "won't work generally",
I could
paraphrase that to say "it may not be sufficient for all packages". I would
make the argument that the implication is that, for some subset of packages,
this approach works quite well. That is in fact, the case for all the packages
I maintain.
When RubyGems were designed in 2004, TDD, BDD and testing in general
was
becoming good practice. Therefore they decided that it is good idea to
ship code together with its test suite and also execute the tests during
installation. So you were supposed to be able to run something like "gem
install -t foo". But it proven to be problematic. Generally this "-t"
option never worked and could be used just for the most naive test
suites. So the option was removed [1]. Nowadays, more and more gem
packages does not ship their tests suites and even the support for
shipping the test suites is deprecated [2].
Ok, that seems like a fair conclusion to draw from your experience, but it is by
no means the only experience. I currently maintain:
cscope
irqbalance
rng-tools
dropwatch
rstp
and a few others, all of which have very mature test suites embedded into their
soruce code. For these pacakages, being able to run their test code in the CI
environment would be very helpful to me (and I think, by extension), others who
have simmilarly constructed packages.
If you want to understand how complex it might be to run the test
suite
of some packages, I welcome you to explore the %check sections of
rubygem- packages. Some are simple, but most are not that simple,
starting with small differences as load path, ending with test suites
which needs to run various servers or check against cloud services. Not
mentioning test matrices.
I do run make check and part of an rpm %check script on most of these, and if
the consensus is that that approach is sufficient for CI purposes, so be it
(that would in fact be much easier for me). But its my understanding that there
is a desire to move all our testing to a CI pipeline (or perhaps it would be
beter to say that we wish to add CI to all our packages), and running the
included self tests with an upstream source tree within the CI pipeline is
dificult at best (owing to the lack of source availability in the CI
environment), nor is running a %check script during a build an available trigger
to clear gating during CI, hence my desire to find a way to make my included
selftests from the source tree available in CI via -test rpm subpackages.
Also, for all these tests we usually try to simulate the environment
as
similar as possible to upstream repository, because this is how these
test suites are designed. It would be even harder to execute the test
suites against installed packages.
Unless you were able to extract the test suite into a separate package, install
it in the environment, and run it, while pointing to the installed binaries that
you are trying to test :)
Neil
>
> Vít
>
>
> [1]
> https://github.com/rubygems/rubygems/commit/429f883210f8b2b38ea310f7fc663...
>
> [2] https://github.com/rubygems/rubygems/issues/735
>
>
> Dne 05. 07. 19 v 19:49 Neil Horman napsal(a):
> > Hey all-
> > I was starting to setup CI for one of my packages in Fedora (cscope),
> > which requires that I have access to the sources to run my test (cscope uses
its
> > own source tree to search for various symbols to confirm that its working
> > properly). Getting the sources in the CI environment is a bit of a pain, so I
> > started working on trying to do this by creating a test subpackage
(specifically
> > named -citest) to package up the sources solely for the purpose of getting them
> > installed and available during CI runs. It occured to me that this offers
> > several advantages, among them:
> > 1) the ability to codify dependencies within the ame spec file, rather than
> > having to copy them to the test.yml file, and keep them in sync
> >
> > 2) The ability to use a file format (rpm spec files) that I'm more familiar
with
> >
> > 3) Easy access to tests that are embedded in the source tree
> >
> > 4) minimizing the test harness setup in test.yml
> >
> > For anyone interested, I've got a pull request started here:
> > https://src.fedoraproject.org/rpms/cscope/pull-request/2
> >
> > If anyone wants to take a look at the changes I had to make to do this (fair
> > warning, its still very rough).
> >
> > That all said, I was wondering if perhaps there was general interest in making
> > this kind of test model somewhat more formal (i.e. creating an rpm macro
library
> > to make test package generation a bit easier, creating a standard entry point
to
> > run tests, etc).
> >
> > Thoughts welcome
> > _______________________________________________
> > devel mailing list -- devel(a)lists.fedoraproject.org
> > To unsubscribe send an email to devel-leave(a)lists.fedoraproject.org
> > Fedora Code of Conduct:
https://docs.fedoraproject.org/en-US/project/code-of-conduct/
> > List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
> > List Archives:
https://lists.fedoraproject.org/archives/list/devel@lists.fedoraproject.org
> _______________________________________________
> devel mailing list -- devel(a)lists.fedoraproject.org
> To unsubscribe send an email to devel-leave(a)lists.fedoraproject.org
> Fedora Code of Conduct:
https://docs.fedoraproject.org/en-US/project/code-of-conduct/
> List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
> List Archives:
https://lists.fedoraproject.org/archives/list/devel@lists.fedoraproject.org