bindepend: suppress missing library messages for macOS system libraries

[rokm]

Starting with macOS Big Sur, the system libraries are hidden. This results in a slew of error messages about missing libraries when
a shared library is scanned for imports. However, all these error messages are harmless, because we avoid collecting system libraries on macOS, nor do we need to further analyze them for dependencies (as they can depend only on other system libraries). Therefore, suppress the missing-library error messages for system libraries.

Closes #5107.

[BoboTiG]

A NEWs fragment would be needed IMO.

[BoboTiG]

Suggested change

	logger.error('Can not find path %s (needed by %s)',
	logger.error('Cannot find path %s (needed by %s)',

[rokm]

Yeah, I was thinking about changing the message. But then, it would need to be changed in other places for consistency, and that makes it out-of-scope for this PR.

[BoboTiG]

Addressing such tiny issue when the code is changed seems reasonable. But yes, this is a very small thing to consider, it is OK for me if you change it or not :)

[bwoodsend]

Out of curiosity, how does in_system_path() behave on Big Sur?

[rokm]

Out of curiosity, how does in_system_path() behave on Big Sur?

It should be version-agnostic, as it's just checking the path prefix.

[bwoodsend]

Also, Github actions now allow you to set runs-on: macos-11.0. Our current CI is running macos-latest which ironically is actually 10.15.7. Would it be worth creating a second branch from this one, setting runs-on: macos-11.0 and letting CI run to find any other issues with Big Sur?

[rokm]

Also, Github actions now allow you to set runs-on: macos-11.0. Our current CI is running macos-latest which ironically is actually 10.15.7. Would it be worth creating a second branch from this one, setting runs-on: macos-11.0 and letting CI run to find any other issues with Big Sur?

I've triggered a run on a branch in my fork: https://github.com/rokm/pyinstaller/actions/runs/493923759

[bwoodsend]

Hmm, in hindsight we probably should have loosened the max of 3 failures rule.

FAILED tests/unit/test_depend_utils.py::test_ctypes_util_find_library - asser...
FAILED tests/functional/test_apple_events.py::test_osx_custom_protocol_handler
FAILED tests/functional/test_apple_events.py::test_osx_event_forwarding - sub...
!!!!!!!!!!!!!!!!!!!!!!!!!! stopping after 3 failures !!!!!!!!!!!!!!!!!!!!!!!!!!!
!!!!!!!!!!!! xdist.dsession.Interrupted: stopping after 3 failures !!!!!!!!!!!!!
= 3 failed, 326 passed, 28 skipped, 22 xfailed, 4 warnings in 292.95s (0:04:52) =

This is still a lot better than I'd hoped. I was expecting every second test to pop and die.

I'd guess that the 1st error is to be expected. libc will now be hidden.

def test_ctypes_util_find_library(monkeypatch):
    # for lind_library() we need a lib actually existing on the system
    if is_win:
        libname = "KERNEL32"
    else:
        libname = "c"
    code = "ctypes.util.find_library('%s')" % libname
    res = __scan_code_for_ctypes(code, monkeypatch)
    assert res

[rokm]

Well, the good news is that in the full run, only those three tests fail:

FAILED tests/unit/test_depend_utils.py::test_ctypes_util_find_library - asser...
FAILED tests/functional/test_apple_events.py::test_osx_custom_protocol_handler
FAILED tests/functional/test_apple_events.py::test_osx_event_forwarding - sub...
= 3 failed, 622 passed, 100 skipped, 46 xfailed, 10 xpassed, 4 warnings in 1051.30s (0:17:31) =

And yeah, test_ctypes_util_find_library is expected to fail (maybe we can add compat.is_macos11 and xfail it). Will check if I can reproduce the other two locally.

[Legorooj]

Nice 👍.