Cannot import pySLSQP on mac-m1

[haotang1995]

Hi, I have a similar issue to this. I managed to install pyOpt on my system (mac-m1, 12.1, py3.8.8, gfortran) without errors and have done everything as suggested. However, I cannot import SLSQP as well.

I am now stuck with the error while "import slsqp" in the pyOpt/pySLSQP directory. The error message said that "mach-o file, but is an incompatible architecture (have 'arm64', need 'x86_64')"
May I ask if this is expected? And is there a solution to use pyOpt on mac-m1? Thanks!

[madebr]

You need to pass -arch arm64 to the compiler/linker.
I think you can do this by using CFLAGS.

CFLAGS="-arch arm64" setup.py build_ext

Perhaps you need to define LDFLAGS as well.

LDFLAGS="-arch arm64" CFLAGS="-arch arm64" setup.py build_ext

If you need other extra compiler arguments, such as -isysroot, I can't really help you further as I don't own anything Apple.

[haotang1995]

Hi madebr, thanks for the hint and the quick reply! The "CFLAGS & LDFLAGS" didn't work, but I do find a related suspicious point:
In my environment, the default --plat-name in "python setup.py build" is macosx-10.9-x86_64, which may well explain the reason for compiling pySLSQP in x86_64.

However, I couldn't find the valid plat-name for mac 12.1 arm64 on Internet... Do you have any further suggestions? I will surely search more. Thanks!

[madebr]

First of all, make sure you're running a python arm executable.
After some googling, I found a user having success by doing export ARCHFLAGS="-arch arm64".

[haotang1995]

Thanks~ I will try it later. I am now reinstalling pure python without conda to make the environment simple and standard, after some googling as well. I will definitely try the suggestion after that. Thanks!

[haotang1995]

Hi @madebr , thanks again for the help! The problem is from conda... After re-installing python3.9 without conda, the default value of plat-name for "python setup.py build_ext" becomes "macosx-10.9-universal2". The installation went smoothly afterward.
(Just for others' convenience, numpy-ext has strict version control and, for now, it only supports numpy-1.21 which is not supported in my environment. This incompatibility resulted in failures while installing numpy-ext using pip. I just removed this constrain in requirement.txt in numpy-ext and installed it manually using "setup.py". All tests of numpy-ext were passed and pyOpt ran smoothly.)