On 7/11/25 4:55 AM, Thomas Huth wrote:
On 10/07/2025 17.51, Pierrick Bouvier wrote:
On 7/10/25 2:42 AM, Thomas Huth wrote:
This patch tackles two issues. First, the tests/functional folder has
become quite crowded already, some restructuring would be helpful here.
Second, we currently encode the target architecture twice in the test
names since a lot of the test file names contain the target, too.
This contributes to the very long output lines when running "make
check-functional".
I agree with the "crowded" aspect of functional tests, and subfolders are
definitely welcome.
Concerning the long line, it seems that setup ("thorough" in this example)
is repeated 3 times on it, which is a bigger issue than having arch name
being repeated.
A simpler solution for the "long line" issue would be to change
mtest2make.py to generate shorter test target names. This would not benefit
for using meson test directly, but from what I understood, this is not
recommended nor supported anyway.
I agree that this is a bigger annoyance here, but as far as I understood
that that repeated "thorough" term is just a listing of all the test suites
that the test is part of. I think this would rather need to be fixed on the
meson test runner side - why does meson list all test suites here? The one
that applies to the current should be enough, shouldn't it?
Meson is usually explicit (thus sometimes verbose) and unambiguous,
which are two very nice virtues, especially for build systems and tests,
at least for my personal taste. You don't want to spend time reverse
engineering the test infrastructure itself when you try to understand
why a fail happened or when writing a new test.
Another additional benefit of listing setup + suite + test in an
exhaustive way is that your command line never change to run any
specific test, whatever your current setup is.
If I had to choose between a wider terminal output and spending time
with grep and find to understand what is running, I would pick the
former any time.
Maybe the complexity comes from our usage of test setups here, to
differentiate categories that will download, vs the one that will not.
We already talked about that in another thread, and I still think that
having a single setup (thorough), which could be the default, would make
things easier, even if people should accept to connect to Internet from
time to time to update their tests assets. They do it to download new
versions of QEMU or interact with the mailing list anyway. That said,
I'm not trying to open the can of worms again, just pointing that it may
be related to this topic.
Thomas
Regards,
Pierrick
