http://docs.python.org/library/string.html#formatstrings
field_name        ::=  (identifier | integer) ("." attribute_name | "[" element_index "]")*
element_index     ::= integer

Subsequent text indicates __getitem__() is used but does not overtly say that a string can be used; but
http://docs.python.org/whatsnew/2.6.html#pep-3101-advanced-string-formatting
gives the example
>>> 'Content-type: {0[.mp4]}'.format(mimetypes.types_map)
and clearly '.mp4' is passed to __getitem__(); a string, not an integer.

Clearly one of these is wrong !
Given that the "what's new" doc goes into some detail about how the content of [...] gets parsed, I'm guessing it's right and the grammar is wrong.

I'm not exactly sure what wording to use here.

      element_index: `integer` | `identifier`

is not exactly correct, because it can be a non-identifier (as the example that eddy quotes points out. It's really "any sequence of characters except ']'".

Any ideas on the best way to express that? Maybe taking a clue from string literals, this would be:
      element_index: `integer` | index_string
      index_string: <any source character except "]"> +

Proposed patch attached. The rest of the documentation in the following 2 paragraphs looks correct. It refers to __getitem__, which is how either strings or integers is looked up.

Checked in.

trunk: r78444
release26-maint: r78445
py3k: r78446
release31-maint: r78447