[macOS Apple Silicon] Installation fails due to compiler errors

[zachriggle]

Summary

• OS: macOS
• Architecture: Apple M1
• Psutil version: Not applicable
• Python version: Python 3.8.2
• Type: wheels

Description

When attempting to install psutil on an Apple Silicon Mac mini running Big Sur, I encounter the following compiler error when building psutil/_psutil_posix.c

    psutil/_psutil_posix.c:514:21: error: duplicate case value '16'
                    case(IFM_1000_T):
                        ^
    psutil/_psutil_posix.c:508:21: note: previous case defined here
                    case(IFM_1000_TX):
                        ^
    1 error generated.
    error: command 'xcrun' failed with exit status 1

[giampaolo]

Mmm I don't understand. I can't see anything wrong with the code:

psutil/psutil/_psutil_posix.c

Lines 508 to 514 in 6bdde37

	case(IFM_1000_TX):
	#endif
	#ifdef IFM_1000_FX
	case(IFM_1000_FX):
	#endif
	#ifdef IFM_1000_T
	case(IFM_1000_T):

Since I cannot reproduce it, if you're proficient with C, could you try downloading latest master version from GIT, try to build from sources and see if you can fix it?

git clone [email protected]:giampaolo/psutil.git
cd psutil
make build

[zachriggle]

It seems that the issue is specific to the build of Python installed by my company. Using python3.9 from Homebrew works.

[gsnedders]

See if_media.h in the XNU release: https://opensource.apple.com/source/xnu/xnu-7195.81.3/bsd/net/if_media.h.auto.html

Specifically, this contains:

#define IFM_1000_T      16              /* 1000baseT - 4 pair cat 5 */
#ifdef PRIVATE
#define IFM_1000_TX     IFM_1000_T      /* For compatibility */
#endif /* PRIVATE */

The challenge here is if Python is compiled against the case where PRIVATE is defined, then clang ends up outputting the above error.

Note there's nothing Apple Silicon specific here, it's simple a case of post-processed source ending up with:

                case(11):
                case(14):
                case(15):


                case(16):





                case(16):

                    return 1000;

Checking whether IFM_1000_T and IFM_1000_TX are equal would make it easy to guard against this.

[giampaolo]

Re-opening. Whoever can reproduce it, please provide a patch.