how to set custom path to libpython3.so?

[properchopsticks]

hi,

i'm working at a project right now and i would like to embed the python interpreter within the rust main program but i would also like to have the python3.so embedded into the binary or at least have pyo3 use the so i provide, not the one installed on the os.

i tested this on windows and i noticed that the program maps the dll installed on the system into the address space of the program and calls apis from there but i would like to use my own version of the python3 library or even better have it completely embedded into the binary and have it mapped into memory at runtime and executed from there.

do you know how can someone embed the so into the binary? or at least make the binary only look for the so provided?

thanks

[adamreichold]

Did you have a look at the PyOxidizer project? Maybe this fulfils your needs already?

[properchopsticks]

hey, thanks for the reply. i looked at pyoxidizer but it still wasn't quite what i was looking for. in the end i managed to statically link the python library by passing arguments to the linker in config.toml:

[build]
rustflags = [
    "-Ctarget-feature=+crt-static",
    "-Crelocation-model=static",
    "-L/usr/lib/python3.10/config-3.10-x86_64-linux-gnu",
    "-L/usr/lib/x86_64-linux-gnu",
    "-lstatic=python3.10",
    "-lstatic=expat",
    "-lstatic=m",
    "-lstatic=z",
    "-lstatic=c",
    "-lstatic=util",
    "-lstatic=dl",
    "-lstatic=pthread"
    ]

oddly enough it doesn't work when trying to use the pyo3 module but in an empty project with no dependencies it works fine.
the problem then is that you have to implement your own python interpreter using the c apis.

[properchopsticks]

hey, i wanted to add a few more details if anyone reads this in the future, so it is possible to link the libpython3.10.a library statically to your binary but then you have a problem with dynamically loaded modules.
not all python modules are written as .py files, some are written as .so libraries and they only have dynamic linked versions and you can only load the by using the "import" keyword in python, so even if you managed to compile and link them statically to your binary they would still have to be imported at runtime and they would need to be done as dynamically linked libraries otherwise the import would fail.
i think a better approach is to link the binary dynamically to the python library and use patchelf to set runpath for the binary and all the libraries required to the $ORIGIN folder, that way it only loads the libraries from the specified folder and nowhere else.