MicroPython Forum (Archive)

Is it possible to write C/C++ extensions to micropython that can be loaded dynamically at runtime (specifically by the Linux port)? I have found documentation that shows me how to write extensions and compile them directly into the interpreter, but I would like to distribute my (cross compiled) extension package separately from the micropython interpreter baked onto my embedded Linux target. Does upip support this?

Thank you.

--wpd

For linux the answer is yes. You can use FFI. There are examples of calling the C runtime library through FFI in micropython-lib.

I'd recommend that you download a copy of micropython-lib and search through for the string ffi

Having said that, the stuff in the C code will need to be independent of micropython (i.e. it won't be able to make any calls into micropython).

https://github.com/micropython/micropython/pull/1627

Does upip support this?

As such dynamic modules don't exist except as a proof of concept, upip obviously doesn't support them.

https://github.com/stinos/micropython/c ... indows-pyd possibly in combination with https://github.com/stinos/micropython-wrap might do what you want on linux, i.e. compile and link a 'foo.so' then load it dynamically (using dlopen and the likes) upon 'import foo'. Has some disadvantages: completely different implementation from the one pfalcon mentions, not in the offical fork, apart from the Travis build it has not been used extensively on unix (but I do make very heavy use of it on windows without problems), last time I tried I couldn't get FATFS to link properly with -rdynamic.

pfalcon wrote: ↑

Fri Jan 26, 2018 6:39 pm

https://github.com/micropython/micropython/pull/1627

Does upip support this?

As such dynamic modules don't exist except as a proof of concept, upip obviously doesn't support them.

Ist there a roadmap or general idea what has to happen to finish this up?