Try to remove install_dir usage from all meson.build files

[rgommers]

For context see mesonbuild/meson#9478 (comment). If it works with 0.60.0 then we can safely get rid of it.

[eli-schwartz]

For the record, it's mesonbuild/meson#9134 that fixed it.

[rgommers]

It's not fixed. I tried to remove two separate instances of install_dir, and both gave the same exception:

...
Dependency openblas found: YES 0.3.12 (cached)
Build targets in project: 199

SciPy 1.7.0.dev0+ababababab

  User defined options
    backend: ninja
    prefix : /home/rgommers/code/bldscipy/installdir

Found ninja-1.10.2 at /home/rgommers/anaconda3/envs/scipy-meson/bin/ninja
Traceback (most recent call last):                                                                  
  File "/home/rgommers/anaconda3/envs/scipy-meson/lib/python3.9/site-packages/mesonbuild/mesonmain.py", line 138, in run
    return options.run_func(options)
  File "/home/rgommers/anaconda3/envs/scipy-meson/lib/python3.9/site-packages/mesonbuild/msetup.py", line 294, in run
    app.generate()
  File "/home/rgommers/anaconda3/envs/scipy-meson/lib/python3.9/site-packages/mesonbuild/msetup.py", line 185, in generate
    self._generate(env)
  File "/home/rgommers/anaconda3/envs/scipy-meson/lib/python3.9/site-packages/mesonbuild/msetup.py", line 247, in _generate
    intr.backend.generate()
  File "/home/rgommers/anaconda3/envs/scipy-meson/lib/python3.9/site-packages/mesonbuild/backend/ninjabackend.py", line 557, in generate
    self.generate_install()
  File "/home/rgommers/anaconda3/envs/scipy-meson/lib/python3.9/site-packages/mesonbuild/backend/ninjabackend.py", line 1120, in generate_install
    self.create_install_data_files()
  File "/home/rgommers/anaconda3/envs/scipy-meson/lib/python3.9/site-packages/mesonbuild/backend/backends.py", line 1490, in create_install_data_files
    pickle.dump(self.create_install_data(), ofile)
  File "/home/rgommers/anaconda3/envs/scipy-meson/lib/python3.9/site-packages/mesonbuild/backend/backends.py", line 1478, in create_install_data
    self.generate_target_install(d)
  File "/home/rgommers/anaconda3/envs/scipy-meson/lib/python3.9/site-packages/mesonbuild/backend/backends.py", line 1622, in generate_target_install
    dir_name = os.path.join('{prefix}', outdir)
  File "/home/rgommers/anaconda3/envs/scipy-meson/lib/python3.9/posixpath.py", line 90, in join
    genericpath._check_arg_types('join', a, *p)
  File "/home/rgommers/anaconda3/envs/scipy-meson/lib/python3.9/genericpath.py", line 152, in _check_arg_types
    raise TypeError(f'{funcname}() argument must be str, bytes, or '
TypeError: join() argument must be str, bytes, or os.PathLike object, not 'NoneType'
FAILED: build.ninja 
/home/rgommers/anaconda3/envs/scipy-meson/bin/meson --internal regenerate /home/rgommers/code/bldscipy /home/rgommers/code/bldscipy/build --backend ninja
ninja: error: rebuilding 'build.ninja': subcommand failed

One of my code comments points to mesonbuild/meson#3206, which indeed looks related. Installing files generated by a custom_target call seems broken if install_dir isn't used.

[rgommers]

Ack one sec, I should replace install_dir by subdir.

[rgommers]

Which results in a saner error at least:

ERROR: custom_target got unknown keyword arguments "subdir"

EDIT: that is with meson 0.60.1

[eli-schwartz]

subdir is a kwarg for py_installation.install_sources()

install_dir is a kwarg for custom_target() and without specifying a location to install to, a custom_target cannot install anywhere, since it does not know what to do.

However, it is supposed to give a useful error message, not a traceback, as reported in mesonbuild/meson#9522 (and fixed by me on meson master).

ERROR: "install_dir" must be specified when installing a target

[eli-schwartz]

The place where you can remove install_dir from is python_installation.extension_module() rather than custom_targets.

[rgommers]

Great, thanks for confirming. My only uses of install_dir are within custom_target - so we're all good here already.

For the record, usages:

scipy/meson.build
111:  install_dir: py3.get_install_dir() + '/scipy'
124:  install_dir: py3.get_install_dir() + '/scipy'

scipy/special/meson.build
480:    install_dir: py3.get_install_dir() + '/scipy/special/tests/data'

scipy/linalg/meson.build
25:  install_dir: py3.get_install_dir() + 'scipy/linalg'

[eli-schwartz]

Ah okay, seems like I was confused about what functions you were using it for, in the linked ticket from the OP.

[rgommers]

No worries, I'm confused all the time here - nontrivial issues & design questions.

[eli-schwartz]

To be clear, actually passing the platlibdir to extension_module() would always be done internally by meson itself and joined to subdir and not exposed to users unless users really, really, really want to use install_dir there for some reason.

I'm pretty sure you only used install_dir because there was a bug (which I fixed) in the subdir kwarg?

This assumed you were using the python module functions (which may in fact be salvageable if install_sources is modified to learn how to handle generated sources. There is resistance to directly supporting install_sources(built_outputs), but perhaps adding a command: kwarg to install_sources() would be more palatable to those people?)

[rgommers]

which may in fact be salvageable if install_sources is modified to learn how to handle generated sources.

Yes, that'd be a great solution. I tried that, and then found mesonbuild/meson#7372.