this fails on python3.5-alpine and python3.6-alpine (works as fine in python2.7-alpine) cd /bug && ln -s /broken_path/to_nowhere broken python -c "import shutil; shutil.copytree('/bug', '/temp', symlinks=True)" Dockerfile example here: https://github.com/docker-library/python/issues/155 https://github.com/python/cpython/blob/c30098c8c6014f3340a369a31df9c74bdbacc269/Lib/shutil.py#L198 seem like its suppressing NotImplementedError, and in our case OsError with ENOSUP was raised

the failure looks like that: Traceback (most recent call last): File "<string>", line 1, in <module> File "/usr/local/lib/python3.5/shutil.py", line 359, in copytree raise Error(errors) shutil.Error: [('/bug/broken', '/temp/broken', "[Errno 95] Not supported: '/temp/broken'")]

Actually the symlinks don't need to be broken. It fails for any kind of symlink on musl. $ ls -l /tmp/symtest lrwxrwxrwx 1 mcrees mcrees 10 Apr 18 21:16 empty -> /var/empty -rw-r--r-- 1 mcrees mcrees 0 Apr 18 21:16 regular lrwxrwxrwx 1 mcrees mcrees 16 Apr 18 21:16 resolv.conf -> /etc/resolv.conf $ python3 >>> import shutil; shutil.copytree('/tmp/symtest', '/tmp/symtest2', symlinks=True) shutil.Error: [('/tmp/symtest/resolv.conf', '/tmp/symtest2/resolv.conf', "[Errno 95] Not supported: '/tmp/symtest2/resolv.conf'"), ('/tmp/symtest/empty', '/tmp/symtest2/empty', "[Errno 95] Not supported: '/tmp/symtest2/empty'")] $ ls -l /tmp/symtest2 total 0 lrwxrwxrwx 1 mcrees mcrees 10 Apr 18 21:16 empty -> /var/empty -rw-r--r-- 1 mcrees mcrees 0 Apr 18 21:16 regular lrwxrwxrwx 1 mcrees mcrees 16 Apr 18 21:16 resolv.conf -> /etc/resolv.conf The implication of these bugs mean that things like pip may fail if it calls shutil.copytree(..., symlinks=True) on a directory that contains symlinks(!) Attached is a patch that works around the issue but does not address why chmod is returning OSError instead of NotImplementedError.

This looks like it may be covered by Issue 31940, about the “shutil.copystat” API. See Anthony’s initial proposal at <https://bugs.python.org/issue31940#msg305528>. Max: I think you need the “else” branch to reraise the exception if “errno” doesn’t match. Perhaps a test case would have picked this up.