Kay Hayen

Nuitka uses the running Python version to determine things at compile time. That does not tie in with cross compilation at all. What would be needed is to run a...

So am I reproducing this indeed with Anaconda on Windows. I am a bit surprised it uses pandas, as that is making the compilation a lot larger than I would...

At one time I was using CPython under Linux using "wine". If the MinGW64 compiler worked with wine, then you would have have to invoke it that way. I would...

So, I monkey patched the `find_spec` like this in `jpype._core`: ``` o_find_spec = _util.find_spec def my_find_spec(*args, **kwargs): print("FIND_SPEC", args, kwargs) r = o_find_spec(*args, **kwargs) print("RES", r) return r _util.find_spec =...

I was then however using gcc cross compiler, but that won't work anymore, I gave up on that. But what I wrote above, ought to work, maybe with a little...

And adding a traceback, this is indeed, `_jpype` asking for the spec of *itself*, and it's not doing that by looking up the `__spec__` value, which arguably could be considered...

Nah, confusing myself there, this is the code: ``` module = sys.modules[fullname] if module is None: return None try: spec = module.__spec__ except AttributeError: raise ValueError('{}.__spec__ is not set'.format(name)) from...

Interestingly we create the module in 2 phase load through a spec value with `PyModule_FromDefAndSpec` which one could expect to set the value of `__spec__`, but so far I didn't...

So, I added setting the `__spec__` for the two phase loading path, where clearly it was missing, but I didn't get an improvement out of that. I am currently verifying...

@lyssdod This is exactly where Nuitka is supposed to be different than Cython, to Nuitka the C code is machine code replacement only, could also be LLVM bytecode, but it...