python.extension_module() writes the absolute destination path in `meson-info/intro-install_plan.json`

[dnicolodi]

The documentation says that the destination paths in meson-info/intro-install_plan.json should use placeholders for the base directories. meson-python relies upon this for detecting whether a project installs python extensions modules, checking for any of destinations including any of {bindir}, {libdir_shared}, {libdir_static}, {py_platlib}, {moduledir_shared}.

Entries for python.extension_module() targets however do not contain any placeholder. IIUC,meson-python expects them to contain {py_platlib}.

Is this a bug on how intro-install_plan.json is put together or is meson-python expectation wrong?

I tested this with Meson 0.60 to 0.63.2 and the behavior is the same.

See https://github.com/FFY00/meson-python/issues/124

[eli-schwartz]

Please double check on both:

• the latest release
• meson git master

This is tracked in meson-python as https://github.com/FFY00/meson-python/pull/104

[eli-schwartz]

Duplicate of #10601

[dnicolodi]

Thanks @eli-schwartz However, judging form the discussion in #10601 the solution proposed in https://github.com/FFY00/meson-python/pull/104 is the wrong one and what is currently done in meson-python is correct. What is needed is a fix for #10601.

#10601 says that this is new in 0.63, however, I tested it with 0.60 and later and the behavior on my test case is the same.

I'll see if I can help with #10601. It seems that the python Meson module generates the install paths with the placeholders, but these get lost somewhere.

[eli-schwartz]

The underlying reliability issue is partially fixed in 0.63.0 (placeholders are never wrong, if we cannot get it right, we just use full paths) and fully fixed in git master (placeholders everywhere).

Note that that ticket was recently closed as fixed. Really recently. It missed the opportunity to make it into 0.63.2, but it may make it into a .3 release, we'll see how things go.

[dnicolodi]

I just tested git master and it does not work for me.

Sorry, scratch that, I run the test in the wrong virtual env...

[eli-schwartz]

It's confirmed to work for SciPy, so I'll need more details than that.

Please note: reproducer should use Meson directly, or otherwise ensure that a PEP 517 build backend isn't messing things up by reinstalling unfixed versions of Meson into an empty virtualenv. (You can also tell PEP 517 to reinstall Meson git master into an empty virtualenv.)

[dnicolodi]

Sorry @eli-schwartz, I run the wrong test. I confirm that git master works as intended.