simplifying build scripts

[krober10nd]

203

• reduces unreadable boilerplate code to build cpp codes on the backend.
• now requiring pybind11 >= 2.6

[krober10nd]

@nschloe hey, hope all is well. I'm trying to migrate over to the new simplified pybind11 syntax for building the cpp codes. Locally when I build on my laptop, things seem to work but as you can see on the CI system none of the builds are able to find the compiled source codes.

Any help would be appreciated.

P.S., I did not understand what the point of this first command after pybind11 is (highlighted); I just made it _SeismicMesh but I suspect that may be a problem too. In pygalmesh it was _pygalmesh so I assumed it was the package name.

[nschloe]

Your CPP module is called _SeismicMesh, not .cpp...

[krober10nd]

I see, thanks. In the previous build system, it put the *.so files in each directory so I could do relative module imports. Now I see that this produces one shared library with all them in it so will have to change the imports all to absolute imports I guess.

[krober10nd]

ah I think I can just create several pybind11 extensions in setup.py so that the *.so end up in the correct folders.

[krober10nd]

Now I get a bunch of

E   ImportError: dynamic module does not define module export function (PyInit__SeismicMesh)

[krober10nd]

It's a naming issue. Working on the tedious fix.

[krober10nd]

cool tests all pass...

so now the user if they're using the recent PEP518 standard, then they will not have to manually install pybind11 as this is done via the pytoml file upon building?

[nschloe]

so now the user if they're using the recent PEP518 standard, then they will not have to manually install pybind11 as this is done via the pytoml file upon building?

It's installed automatically, indeed.

[nschloe]

Biggest gain for you is probably this though:

[krober10nd]

Yea I agree. The more lines, the more problems.