Closed ThomasWaldmann closed 6 years ago
Hmm, interesting. I'm not sure how this broke with 1.9.0 -- it looks like this problem has been there for a long time. It's caused because sys.platform is sunos5
on this OS, and PosixDirEntry
is defined if ctypes is not None and have_dirent_d_type
, but used if ctypes is not None
without the extra clause.
However, can you please tell me what is the version of scandir installed when it works, i.e., the scandir<1.9.0
in your case?
It installed 1.8.0.
I found this while doing borgbackup testing and resurrecting the openindiana vagrant box now as the vagrant change that broke oi boxes was found. I'll try to reproduce this with a clean virtualenv in a minute...
In a clean python 3.4.7 virtual env:
started at 1.5.0, upgrading scandir via pip install scandir==V
for V in 1.6.0, 1.7.0, 1.8.0, 1.9.0.
it definitely does not traceback with <= 1.8.0, but tracebacks after upgrading to 1.9.0.
OK, guess I have the reason why it crashes:
OK, guess I have why the C ext building was broken:
Because vagrant did not support openindiana for a while and was just recently fixed, my openindiana box was a bit outdated. Some stuff (e.g. current Cython, maybe also scandir?) now wanted gcc-6 which somehow was not there (yet). So my first tries to build that stuff failed when it tried to compile C stuff - but I guess pip cached a wheel for scandir 1.9.0 made at that time (and thus: without _scandir C extension).
Later, i fixed by updating the system and also installing gcc-6.
But, as pip had a cached wheel for scandir 1.9.0, it continued using this and thus missed the C extension.
I killed ~/.cache/pip
later and now a freshly pip-installed scandir 1.9.0 does have the C extension and thus does not trigger the exception any more.
So seems like this rather was a caching issue than a build problem rooted in scandir setup code.
But at least we found a bug, so please merge #111.
Excellent, thanks. I just merged #111.