This is test.py:

def foo():
    print "Bla"

And in an interactive session:

>>> import inspect, test
>>> test.foo()
Bla
>>> inspect.getsource(test.foo)
'def foo():\n    print "Bla"\n'

Now I edit test.py to (without exiting the interactive
session):

def foo():
    print "Oh no!"

>>> reload(test)
<module 'test' from 'test.py'>
>>> test.foo()
Oh no!
>>> inspect.getsource(test.foo)
'def foo():\n    print "Bla"\n'

inspect should output the new source.

Logged In: YES
user_id=1188172

This is the fault of the linecache module which inspect
uses. It caches the file contents and does not reload it
accordingly. However, I assume it's pretty hard finding a
good solution to this.

I'm attaching a patch. Is there some hidden problem that it may cause ?
By the way, this issue is a duplicate of http://bugs.python.org/issue993580

Fix was committed to py3k but with no tests AFAIK. Here's a (failing)
test for trunk that runs OK after Guilherme's patch.

We might want to keep this behavior restricted for when the module was
reloaded.

Er, no, it wasn't committed to py3k, it was just me getting my patches
mixed. Attaching the py3k test with relative import fixed.

The change to inspect.py is two lines. Given that unit test patches are also supplied could someone with the knowledge give yes or no to accepting this patch?

Calling linecache.checkcache for every inspect.getsource call sounds like a fairly bad idea to me.

linecache.checkcache does a stat() of every single cached file.

linecache.checkcache does a stat() of every single cached file.

Ah, sorry. I didn't read carefully enough. I see that the patch passes in the filename and checkcache restricts the work it does in that case.

Something else to consider, though, is that this change means you'll also get the new source if you *don't* reload the module, too. So, what exactly is inspect.getsource() supposed to be doing? Giving you the current on-disk contents of the relevant source file? Or giving you the actual source corresponding to the object passed to it?

If the latter, then this actually needs to be more tightly integrated with module reloading somehow.

This issue appears implicated in https://bitbucket.org/pypa/setuptools/issue/201

Thanks, I'll take a look at the patch.

Note that there has never been a strict correspondence between linecache/getsource, the module in memory, and the code on disk. If the file is changed before a traceback is generated, for example, you will get the new source. That doesn't mean fixing the reload case is wrong, it's just something to keep in mind.

New changeset 8e2505d535c8 by Yury Selivanov in branch 'default':
inspect: Fix getsource() to load updated source of reloaded module
https://hg.python.org/cpython/rev/8e2505d535c8

Fixed in 3.5. Not sure if we need to backport this to 3.4 and 2.7.

Closing this issue. Thanks to Björn Lindqvist and Berker Peksag!

FWIW, I'd appreciate a backport to 3.4, given that 3.5 is scheduled for release in Sep 2015.

New changeset e52d8e888df1 by Yury Selivanov in branch '3.4':
inspect: Fix getsource() to load updated source of reloaded module
https://hg.python.org/cpython/rev/e52d8e888df1

@jason: done ;)