[RFC] detecting "file not found" and "directory not found" exceptions on Windows

[Mark Hammond]

I was tracking down some test failures on Windows, a'la:

| WindowsError: [Error 267] The directory name is invalid:
u'O:/window~1/testbzr-3ml2mo.tmp/nd_conflicts_wrong_parent_kind/work/parent\
\*.*'

and came across an issue that might be worth fixing in a slightly broader
context.

The short story is that code in osutils.py and transform.py is assuming
os.listdir raises an exception with errno set to ENOTDIR - but this isn't
true on Windows.  Looking a little further, I see that John and Alex already
struck this issue in a slightly different context - eg, workingtree_4.py has
the block:

"""
...
            except OSError, e:
                # on win32, python2.4 has e.errno == ERROR_DIRECTORY, but
                # python 2.5 has e.errno == EINVAL,
                #            and e.winerror == ERROR_DIRECTORY
                e_winerror = getattr(e, 'winerror', None)
                win_errors = (ERROR_DIRECTORY, ERROR_PATH_NOT_FOUND)
                # there may be directories in the inventory even though
                # this path is not a file on disk: so mark it as end of
                # iterator
                if e.errno in (errno.ENOENT, errno.ENOTDIR, errno.EINVAL):
                    current_dir_info = None
                elif (sys.platform == 'win32'
...
"""

All os.listdir() calls, and potentially others, should be wrapped with an
exception handler similar to this.  The listdir calls mentioned are easy to
fix - but a quick grep shows other possibly suspect cases.  Sometimes we are
sure the parent directory exists meaning ENOENT is OK (such as
fancy_rename), but sometimes its not completely clear how it will be used
(eg, Transport._translate_error()).

I've attached a patch that fixes just the test failures.  Before this patch
my test suite says:

| Ran 13687 tests
| FAILED (failures=66, errors=201, known_failure_count=14)
| 960 tests skipped

and after I see:

| Ran 13687 tests
| FAILED (failures=51, errors=106, known_failure_count=14)
| 960 tests skipped

While this might be a good approach for the 1.6 branch, I think that causing
such code to spread may be error prone long term (and there are already
other inconsistencies in the handling).  I'm wondering it we should try and
abstract this a little, introducing a new (osutils?) helper function that
takes an exception object and indicates if it means "directory not found",
then consolidating the similar code in workingtree_4, osutils and transform?
Note that the "file not found" handling does seem portable between windows
and linux, but that might be worth abstracting too?  If so, any advice on
names, places, etc?  Even better, anyone want to take it on <wink>?

Thanks,
 
Mark

-------------- next part --------------
A non-text attachment was scrubbed...
Name: listdir_exceptions.patch
Type: application/octet-stream
Size: 3948 bytes
Desc: not available
Url : https://lists.ubuntu.com/archives/bazaar/attachments/20080809/c0bbddf3/attachment.obj