structured_sources: Allow build_tgt to be added as source

[sp1ritCS]

In some cases, there is a need to use a built library as a "source" for another target. Specifically Java JAR might want to contain their own native JNI libraries in order to provide a single file "executable".

I'm kind of surprised this works out of the box without any further changes 🙃

[tristan957]

Can you please extend a test?

[sp1ritCS]

@tristan957 added

seems like this needs at least java 9. Is it enough to bump the dependency('jni') or do I need to add another mechanism the skip the builds for java 1.8?

[tristan957]

For skipping on < Java 9, do this:

jni_dep = dependency('jni')
if jni_dep.version().version_compare('< 1.9')
  error('MESON_SKIP_TEST: this test requires Java 9')
endif

You may also want to take a look at the CI images to make sure they aren't set to install Java 8. You'll also want to skip the test if the dependency isn't found.

[tristan957]

Java doesn't have a stem function for this?

[sp1ritCS]

It seemingly doesn't. And even if it did, I'd consider a stem function is just as hacky. Ideally you could just access lib prefix and suffix from java, but you can't, so you have to gain access to the suffix for mkstemp somehow.

No idea if indexOf or lastIndexOf is better, the first breaks if the library prefix or libname contains a ., the second breaks if the extension were to contain a . (dunno, imagine a system capable of loading .so.gz or so)

[tristan957]

I don't think you need a whole new test for this. I would extend this one: https://github.com/mesonbuild/meson/tree/master/test%20cases/java/9%20jni. What do you think?

[tristan957]

@dcbaker is there any other place that would be a good test in addition to this? Not sure if there is a more generic structured_sources test.

[sp1ritCS]

I don't think you need a whole new test for this. I would extend this one: https://github.com/mesonbuild/meson/tree/master/test%20cases/java/9%20jni. What do you think?

I've thought about it, but it doesn't really work well with how the test is structured in src/ and lib/. As first we need to generate the native headers in the src/ tree, then compile the library (lib/) that is then again needed in src/ for the structured_sources call.

One could generate the native headers in lib/, but then your throwing long paths around and I'd want to avoid that.

[tristan957]

If you see a good way to refactor the existing test, that would be fine.

[tristan957]

We need to get the image jobs to pass. Not your fault. I'm working on fixing Fedora.

[tristan957]

If you rebase on master, you should pick up the Fedora CI fixes. I think msys2 will also start to pass.

[sp1ritCS]

Status report:

• Arch is failing due to mirror issues
• openSUSE & Ubuntu have issues with "dub" dependencies (no idea what that is or how to fix it)
• Fedora has issues with zlib

[eli-schwartz]

Disable debug support in AUR packages, instead. We are not e.g. testing that valgrind can perform valgrind on scalapack / wxwidgets-gtk2 so there's no reason to spend longer compiling and produce bigger docker containers.

[eli-schwartz]

Since you're not trying to update the opensuse image today, only the Arch one, I would suggest letting opensuse fix this bug and having meson pick that fix up next week on the weekly refresh job.

You're not going to fix the opensuse image anyway unless you get dub to work, but that's tracked by #12143

[eli-schwartz]

Thanks for debugging the upstream opensuse issue and submitting a fix to them, though. :)

[sp1ritCS]

coincidentally im probably the one at fault, given that I submitted lcov 2.0 in the first place 🙃

[tristan957]

I have added a PR to really fix the fedora test at #13436