Roundup Robot added the comment:
New changeset 76184b5339f2 by Brett Cannon in branch 'default':
Issue #18416: Have importlib.machinery.PathFinder treat '' as the cwd
http://hg.python.org/cpython/rev/76184b5339f2
--
nosy: +python-dev
___
Python
Brett Cannon added the comment:
I went with option 1 (changed PathFinder to consider '' the cwd) which allowed
me to flat-out remove the special-casing for '' in FileFinder.
Thanks for the initial patch, Madison!
--
resolution: - fixed
stage: test needed - committed/rejected
status:
Roundup Robot added the comment:
New changeset 33844153cd02 by Brett Cannon in branch 'default':
Issue #18416: Fix various os calls in importlib.machinery.FileFinder
http://hg.python.org/cpython/rev/33844153cd02
--
___
Python tracker
Changes by Brett Cannon br...@python.org:
--
assignee: - brett.cannon
___
Python tracker rep...@bugs.python.org
http://bugs.python.org/issue18416
___
___
Madison May added the comment:
Here's a minor revision to that patch removing an unnecessary
@skip_if_dont_write_bytecode decorator from the test I added to test_import.py.
No docs changes are included in the current patch -- I'm guessing this should
probably wait until we have all the
Changes by Madison May madison@students.olin.edu:
Added file: http://bugs.python.org/file31248/Issue18416_v2.patch
___
Python tracker rep...@bugs.python.org
http://bugs.python.org/issue18416
___
Brett Cannon added the comment:
To answer Eric's question: yes.
Since no one seems to be screaming that sys.path be the only place to have the
concept of a relative path, then let's use only absolute paths except in
sys.path for ''.
--
___
Python
Madison May added the comment:
I'm currently leaning towards having sys.path_importer_cache store
the actual directory name.
Exactly what I meant by Creating a new entry in sys.path_importer_cache after
changing directories and importing a new module, but expressed much more
succinctly :)
Brett Cannon added the comment:
I'm currently leaning towards having sys.path_importer_cache store the actual
directory name.
--
___
Python tracker rep...@bugs.python.org
http://bugs.python.org/issue18416
___
Eric Snow added the comment:
I'm currently leaning towards having sys.path_importer_cache store
the actual directory name.
Makes sense to me. It the problem that sys.path will still have '' in it and
it may be hard to figure out what it means (and how it maps to
sys.path_importer_cache)?
Nick Coghlan added the comment:
It seems to me that by *not* doing that, we may have a bug if the cwd changes
and the import machinery returns a stateful importer that remembers the path
name. Using the actual filesystem path for everything other than sys.path
sounds like a much better
Madison May added the comment:
I quickly ran the tests with each of the above edits to see what bits of the
test suite would break.
With option 1 (edits to PathFinder only):
``1 test failed:
test_importlib``
With option 2 (edits to FileFinder only):
``3 tests failed:
test_import
Brett Cannon added the comment:
So this is bringing up a sticky situation that I ran across when initially
implementing all of this: what should sys.path_importer_cache use as a key? ''
would be what happens with Madison's option 3, and with option 1 it would be
os.getcwd(). Now if you
Madison May added the comment:
Nick, it was definitely a good thing to mention. I had to learn the edit,
build, test cycle the hard way my first time. It took me a good 15-20 minutes
to figure out why none of my edits seemed to change anything :)
Anyhow, here's how I see the issue. It
Brett Cannon added the comment:
I actually meant FileFinder but PathFinder is a good point.
In FileFinder you need to change ``self.path = path or '.'`` to use the cwd.
--
___
Python tracker rep...@bugs.python.org
http://bugs.python.org/issue18416
Madison May added the comment:
A few days ago I tried the change: ``self.path = path or _os.cwd()``, but the
problem didn't seem to resolve itself.
``./python -c import blah; print(blah.__file__)`` still returned a relative
path on my system. The tie between FileFinder and the __file__
Nick Coghlan added the comment:
Make sure to run make to rebuild the frozen module - editing
importlib._bootstrap.py involves a C style edit, build, test cycle rather
than the typical Python edit, test cycle.
--
___
Python tracker
Madison May added the comment:
Thanks for the heads up, Nick. I've worked with _bootstrap.py before, so I'm
familiar with the cycle.
--
___
Python tracker rep...@bugs.python.org
http://bugs.python.org/issue18416
Nick Coghlan added the comment:
Figured it was worth mentioning, since I've been caught by forgetting
that myself :)
I suspect you're right that it's just a case of '.' passing the truth
test, even though it still results in a relative path.
--
___
Madison May added the comment:
PathFinder or FileFinder? Changing PathFinder._path_importer_cache(cls, path)
seems to fix the issue.
See line 1302 in _bootstrap.py.
if path == '':
-path = '.'
+path = _os.getcwd()
$ touch blah.py; ./python -c import blah;
Ronald Oussoren added the comment:
Isn't this because the first entry of sys.path is '' when python is started
without a script? All other paths on sys.path are already absolute paths
(tested with a relative path in $PYTHONPATH).
If that's correct, just adding os.getcwd() instead of '' as the
Brett Cannon added the comment:
That's exactly why it currently happens. What I would do is change importlib to
just associate '' with os.getcwd() in FileFinder. That locks down where the
module was found to an absolute path while still allowing '' to function as the
cwd and be dynamically
Nick Coghlan added the comment:
Brett's plan sounds good to me. We should also check we make sure
__main__.__file__ is always absolute.
--
___
Python tracker rep...@bugs.python.org
http://bugs.python.org/issue18416
New submission from Brett Cannon:
$ touch blah.py; ./python -c import blah; print(blah.__file__)
./blah.py
Should really change that to be an absolute path since the file's location is
not accurate if you call os.chdir().
--
components: Interpreter Core
messages: 192772
nosy:
Changes by Arfrever Frehtes Taifersar Arahesis arfrever@gmail.com:
--
nosy: +Arfrever
___
Python tracker rep...@bugs.python.org
http://bugs.python.org/issue18416
___
Changes by Brett Cannon br...@python.org:
--
nosy: +ncoghlan
___
Python tracker rep...@bugs.python.org
http://bugs.python.org/issue18416
___
___
Python-bugs-list mailing
26 matches
Mail list logo
