Set correct compile flags for Apple M1 chips as well as 32bit machines.

[tmikus]

This PR also fixes a Segmentation Fault when calling the detect function on Macs running M1 chip.
The array returned from the PyArg_ParseTupleAndKeywords was being set to an incorrect address, which caused a segmentation fault. Surprisingly, changing the call so that it requests a null-terminated string fixes the problem.

I tried using other forms of that function (s*, y*, y#) but to no avail.
I suspect this might be a problem with Python (3.9.5 for Apple Silicon).

One thing to remember is that the detect function will no longer accept byte-like objects. It will still work fine with utf-8 strings.

[NickHilton]

This works well for me on Mac M1. Is there a way this can be merged / released as a separate version to pypi?

[gabrielcabola]

Hey, where this will be merged?
This is still relevant as an issue for M1 machines.
Thanks

[akgerber]

Thanks for posting this @tmikus — made my day much easier

here's a script to apply this fix in a virtualenv located at path $MYVIRTUALENV:

mkdir $MYVIRTUALENV/src
cd $MYVIRTUALENV/src
wget https://files.pythonhosted.org/packages/21/d2/8b0def84a53c88d0eb27c67b05269fbd16ad68df8c78849e7b5d65e6aec3/pycld2-0.41.tar.gz
tar -xzvf pycld2-0.41.tar.gz
cd pycld2-0.41
wget https://patch-diff.githubusercontent.com/raw/aboSamoor/pycld2/pull/44.patch
patch -p1 < 44.patch
CFLAGS="-w -O2 -fPIC -march=armv8-a" pip install --no-cache-dir --no-deps --no-build-isolation  .

[cotterpl]

@aboSamoor why this is still not solved?

[leegunwoo98]

still getting an error

[jurifm2406]

got an error

error: unknown target CPU 'armv8-a'

on apple m1 arm64

[gregoriopellegrino]

You can install it using:

pip install git+https://github.com/tmikus/pycld2.git

[puja93]

Thanks @tmikus , your PR works fine on my M1 for utf-8 but not byte object, as you mentioned.
It has been sometime, but i do wonder if you've found a solution to include byte object as well ?

I'm currently using polyglot that depends on this. Currently i just switch this line in order to make it work.
cld2.detect(t) -> cld2.detect(t.decode('utf-8')

So i wonder if you've made new findings on byte objects....