[Python-Dev] serious bug in new import X as Y code

[Thomas Wouters]

On Fri, Aug 18, 2000 at 11:42:38AM +0200, Sjoerd Mullender wrote:

> Your changes for the import X as Y feature introduced a serious bug:
> I can no longer run Python at all.

> The problem is that in line 2150 of compile.c com_addopname is called
> with a NULL last argument, and the firat thing com_addopname does is
> indirect off of that very argument.  On my machine (and on many other
> machines) that results in a core dump.

Hm. That's very strange. Line 2150 of compile.c calls com_addopname with
'CHILD(CHILD(subn, 0), 0)' as argument. 'subn' is supposed to be a
'dotted_as_name', which always has at least one child (a dotted_name), which
also always has at least one child (a NAME). I don't see how dotted_as_name
and dotted_name can be valid nodes, but the first child of dotted_name be
NULL.

Can you confirm that the tree is otherwise unmodified ? If you have local
patches, can you try to compile and test a 'clean' tree ? I can't reproduce
this on the machines I have access to, so if you could find out what
statement exactly is causing this behaviour, I'd be very grateful. Something
like this should do the trick, changing:

                        } else
                                com_addopname(c, STORE_NAME,
                                              CHILD(CHILD(subn, 0),0));

into

                        } else {
                                if (CHILD(CHILD(subn, 0), 0) == NULL) {
                                        com_error(c, PyExc_SyntaxError,
                                                  "NULL name for import");
                                        return;
                                }
                                com_addopname(c, STORE_NAME,
                                              CHILD(CHILD(subn, 0),0));
                        }

And then recompile, and remove site.pyc if there is one. (Unlikely, if a
crash occured while compiling site.py, but possible.) This should raise a
SyntaxError on or about the appropriate line, at least identifying what the
problem *could* be ;)

If that doesn't yield anything obvious, and you have the time for it (and
want to do it), some 'print' statements in the debugger might help. (I'm
assuming it works more or less like GDB, in which case 'print n', 'print
n->n_child[1]', 'print subn', 'print subn->n_child[0]' and 'print
subn->n_child[1]' would be useful. I'm also assuming there isn't an easier
way to debug this, like you sending me a corefile, because corefiles
normally aren't very portable :P If it *is* portable, that'd be great.)

> In case it helps, here is the stack trace.  The crash happens when
> importing site.py.  I have not made any changes to my site.py.

Oh, it's probably worth it to re-make the Grammar (just to be sure) and
remove Lib/*.pyc. The bytecode magic changes in the patch, so that last
measure should be unecessary, but who knows :P

breaky-breaky-ly y'rs,
-- 
Thomas Wouters <thomas@xs4all.net>

Hi! I'm a .signature virus! copy me into your .signature file to help me spread!