I can't quite sort this out, because it's difficult to see what is
intended. The documentation of xml.etree.ElementTree (19.11 in the
Library doc) uses terms like "iterator", "tree iterator", "iterable",
"list" in vague and perhaps not quite accurate ways. I can't tell from
the documentation which functions/methods return lists, which return a
generator, which return an unspecified kind of iterable, and so on.
Moreover, the results are different using ElementTree than they are
using cElementTree. In particular, getiterator() returns a list in
ElementTree and a generator in cElementTree. This can make a substantial
difference in performance when iterating over a large number of nodes
(in addition to cElementTree's parsing being what appears to be about
10x faster).

I think someone should go over the page and sort this out and make it
clear what the user can expect. (I don't think it's fair to
overgeneralize to things like "iterables" if the module is really meant
to be making a commitment to a list or a generator.) I also think that
the differences in the results of methods returned in the Python and C
versions of the module should be highlighted.

I stumbled on this trying to parses and extract individual bits of
information out of large XML files. I full well realize there are better
ways to do this (SAX, e.g.) and better ways to search than just iterate
over all the tags of the type I'm interested in, but I should still know
what to expect from ElementTree, especially because it is so wonderful!

This isn't just a documentation issue. A function named getiterator(),
for which the docs say that it returns an iterator, should return an
iterator, not just an iterable. They have different semantics and can't
be used interchangeably, so the behaviour of getiterator() in
ElementTree is wrong. I was using this in my program:

iterator = element.getiterator()
next(iterator)
subelement = next(iterator)

Which broke when I tried switching to ElementTree from cElementTree,
even though the docs tell me that I'll get an iterator there.

Also, for findall() and friends, is there any reason why we can't stick
to either an iterator or list, and not both? The API will be more clear
if findall() always returned a list, or always an iterator, regardless
of the implementation. It is currently not clear what will happen if I do:

for x in tree.findall(path):
     mutate_tree(tree, x)

There's many differences between both implementations.
I don't know if we can live with them or not.

~ $ ./python 
Python 3.1.1+ (release31-maint:76650, Dec  3 2009, 17:14:50) 
[GCC 4.3.2] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> from xml.etree import ElementTree as ET, cElementTree as cET
>>> from io import StringIO
>>> SAMPLE = '<root/>'
>>> IO_SAMPLE = StringIO(SAMPLE)

With ElementTree

>>> elt = ET.XML(SAMPLE)
>>> elt.getiterator()
[<Element root at 15cb920>]
>>> elt.findall('')  # or '.'
[<Element root at 15cb920>]
>>> elt.findall('./')
[<Element root at 15cb920>]
>>> elt.items()
dict_items([])
>>> elt.keys()
dict_keys([])
>>> elt[:]
[]
>>> IO_SAMPLE.seek(0)
>>> next(ET.iterparse(IO_SAMPLE))
('end', <Element root at 15d60d0>)
>>> IO_SAMPLE.seek(0)
>>> list(ET.iterparse(IO_SAMPLE))
[('end', <Element root at 15583e0>)]

With cElementTree

>>> elt_c = cET.XML(SAMPLE)
>>> elt_c.getiterator()
<generator object getiterator at 0x15baae0>
>>> elt_c.findall('')
[]
>>> elt_c.findall('./')
[<Element 'root' at 0x15cf3a0>]
>>> elt_c.items()
[]
>>> elt_c.keys()
[]
>>> elt_c[:]
Traceback (most recent call last):
TypeError: sequence index must be integer, not 'slice'
>>> IO_SAMPLE.seek(0)
>>> next(cET.iterparse(IO_SAMPLE))
Traceback (most recent call last):
TypeError: iterparse object is not an iterator
>>> IO_SAMPLE.seek(0)
>>> list(cET.iterparse(IO_SAMPLE))
[(b'end', <Element 'root' at 0x15cf940>)]

Proposed patch fixes most of the discrepancies between both implementations.

It restores some features that were lost with Python 3:

• cElement slicing and extended slicing
• iterparse, cET.getiterator and cET.findall return an iterator
  (as documented)

Some tests were added to check these issues.

I fixed it differently, using the upstream modules (Thank you Fredrik).

• ElementTree 1.3a3-20070912
• cElementTree 1.0.6-20090110

It works.
And it closes bpo-1143, too.

The patch should have doc updates for new functionality, if any.

I see some new features in the changelog.
I will try to update the documentation during the week.

(patch "py3k" fixed: support assignment of arbitrary sequences)

Patch for the documentation. (source: upstream documentation)

Small update of the patch for 3.2: the __cmp__method is replaced with
__eq__ method (on CommentProxy and PIProxy).

It would be nice to upgrade ElementTree for 2.7 and 3.2, at least.

Patch updated, with upstream packages:

• ElementTree 1.3a3-20070912
• cElementTree 1.0.6-20090110

Now all tests are identical for the ElementTree part:

• ElementTree 2.x
• cElementTree 2.x
• ElementTree 3.x
• cElementTree 3.x

Waiting for some developer kind enough to review and merge in 2.7 and 3.2.

Given the size of the patch, it's very difficult to review properly.
In any case, could you upload it to http://codereview.appspot.com/ ?

Ok, will do the upload to rietveld.

In addition to the straight review of the patch itself, you could:

• diff against the upstream source code (very few changes)
• diff between 2.x and 3.x
• review the test_suite (there's only additions, no real change)
• hunt refleaks

Btw, I've backported the last tests (bpo-2746, bpo-6233) to all 4 test files (ET and cET, 2.x and 3.x).

Here it is:

• http://codereview.appspot.com/207048/show

Update the 2.x patch with the last version uploaded to rietveld (patch set 5).

Improved test coverage with upstream tests and tests cases provided by Neil on issue bpo-6232.

Note: the patch for 3.x is obsolete.

Strip out the experimental C API.

Fixed on trunk with r78838.
Some extra work is required to port it to 3.x.

Thank you Fredrik and Antoine for reviewing this patch.

W00t!

Patch to merge ElementTree 1.3 in 3.x.

Merged in 3.x with r78942 and r78945.

See bpo-8047 for a discussion about the encoding argument of the serializer (used for .write() method and tostring() tostringlist() functions).
Currently the output is not encoded by default in 3.1 and 3.x.
It is encoded to ASCII in 2.6 and 2.x.