The Python parser stores the filename as a byte string. But it decodes the filename on error because most Python functions now use unicode strings. Instead of decoding the filename at error, which may raise a new error, I propose to decode the filename on the creation of the parser object and only store the filename as unicode.

This issue would prepare the last part of the full unicode support (#3080).

I like the idea, but I don't like the trend that parser code continues to diverge from pgen.  I understand that most of the Python runtime is not available to pgen, but maybe a more elegant solution than changing the type conditional on PGEN can be found.  For example, maybe filename could be decoded from FS encoding to UTF-8?

> maybe a more elegant solution than changing the type conditional 
> on PGEN can be found

In pgen, the filename is only used to display the following warning, in indenterror():

   <filename>: inconsistent use of tabs and spaces in indentation

In pratical, this warning never occurs on Grammar/Grammar: this file doesn't use indentation at all, only continuation lines.

A better solution is maybe just to drop the filename for pgen. Anyway, pgen only compiles *one* file (Grammar/Grammar), so we don't need the input filename ;-)

When testing my patch, I found and fixed two bugs in pgen:
 - r87557: PGEN was not defined to compile pgenmain.c and printgrammar.c
 - r87558: pgen error was ignored on "make Parser/pgen.stamp" (when executing pgen to compile the grammar)

Version 2 of the patch:
 - remove filename attribute from perrdetail and tok_state structure in PGEN mode, and add a comment to explain why
 - rename filename_obj to filename
 - indenterror() no longer print the input filename in PGEN mode

err_clear() should set err->filename to NULL.

Version 3 of the patch to fix also #9319.

@Benjamin: You told me that you don't want two versions of pgen, but I don't remember why. As my work on #3080 is mostly done, I now plan to patch the Python parser to store the filename as Unicode. So could you please review the patch attached to this issue?

New changeset 6e9dc970ac0e by Victor Stinner in branch 'default':
Issue #10785: Store the filename as Unicode in the Python parser.
http://hg.python.org/cpython/rev/6e9dc970ac0e