Here's a first draft of a small addon to shutil.copytree.

This patch allows excluding some folders or files from the copy, given
glob-style patterns. A callable can also be provided instead of the
patterns, for a more complex filtering.

I didn't upgrade Doc/shutil.rst yet in this patch, as this can be done
when the change will be accepted and in its final shape I guess.

On the interface, I would suggest renaming 'exclude' to 'ignore' for 
consistency with filecmp.dircmp. Also consider detecting file separator 
in the patterns and interpreting them as an absolute (if 
pattern.startswith(pathsep)) or relative with respect to src. 

On the implementation, consider making 'exclude_files' a set for a 
faster lookup.   It should also be possible to refactor the code to 
avoid checking the type of 'exclude' on every file and every recursion.

I changed the patch based on all remarks. For the absolute path, I was
wondering if it would be useful since calls are recursive, relative to
the visited directory.

Is there any reason for rmtree also to not support this exclusion
feature? Both copytree and rmtree explicitly iterate over list of names
and as I see it, this exclusion is really about which names to ignore.
Already, copytree and rmtree have inconsistencies (rmtree has 'onerror'
while 'copytree' doesn't) and it would be nice to not add more.

Agreed, rmtree should have it as well. I'll add that in the patch as well,

while working on the patch to add the same feature in rmtree, I realized
this is a non sense since the root folder itself is removed at the end
of the function when all its content is removed.

So, unless we change this behavior, which I doubt it is a good idea, it
won't be possible.

Maybe another API could be added in shutil, in order to do any kind of
treatment in a tree, like removing files, or whatever, and without
copying it like copytree does.

I have thaught of various ways to write this new API for the deletion
use case, but I think nothing makes it easier and shorter than a simple
os.walk call.

This patch includes the documentation for shutils.rst as well. (I
removed the older patches)

My update with email failed so I am just copying my response here:

>  while working on the patch to add the same feature in rmtree, I realized
>  this is a non sense since the root folder itself is removed at the end
>  of the function when all its content is removed.

Indeed. Sorry about that.

>  So, unless we change this behavior, which I doubt it is a good idea, it
>  won't be possible.

I agree. But in general, it would be nice to separate file list
generation and the actual operation. Something similar to shell where
it resolves the pattern while the actual command itself cares only
about the files passed to it. This is not necessarily a comment on
this patch which I am hoping I can check it out soon.

The patch looks good to me.

I forgot to add that the example provided in rst doc is incorrect. The
copytree() in that example should be given destination path as well. In
addition, the docstring for copytree mentions "which is a directory
list". "directory list" is a bit vague and should ideally be replaced by
something like "list of elements" (which is what appears in the doc) or
"list of entries".

Right, thanks.

I have corrected the doc, and pushed some examples at the bottom of the
module documentation.

patch with the new trunk

Hi Tarek,

here's a review:

* The new docs are not very clear about ignore_patterns being a function
factory. E.g.:

"""The callable must return a list of folder and file names relative to
the path, that will be ignored in the copy process.
:func:`ignore_patterns` is an example of such callable."""

Rather, the *return value* of ignore_patterns is an example of such a
callable.

* The new docs should also note that copytree is called recursively, and
therefore the ignore callable will be called once for each directory
that is copied.

* Instead of "path and its elements" the terminology should be
"directory" and "the list of its contents, as returned by os.listdir()".
Likewise, "folder" should be "directory".

* The second new example makes me wonder if *ignore* is the correct name
for the parameter. Is *filter* better?

* A nit; the signature should be "copytree(src, dst[, symlinks[, ignore]])".

* The patch adds a space in the definition of rmtree().

Thanks Georg, I have changed the patch accordingly.

There's one issue left: the name of the parameter (ignore)

I have renamed it like this on Alexander suggestion, for consistency
with filecmp.dircmp which uses ignore. 

By the way, I was wondering: do we need to used reStructuredText as well
in function doctstrings ?

Committed in r64722. Thanks everyone!