Bugs item #1545668, was opened at 2006-08-24 03:14 Message generated for change (Comment added) made by jwhowarth You can respond by visiting: https://sourceforge.net/tracker/?func=detail&atid=105470&aid=1545668&group_id=5470
Please note that this message will contain a full copy of the comment thread, including the initial issue submission, for this request, not just the latest update. Category: Python Interpreter Core Group: Python 2.4 Status: Open Resolution: None Priority: 5 Submitted By: Jack Howarth (jwhowarth) Assigned to: Tim Peters (tim_one) Summary: gcc trunk (4.2) exposes a signed integer overflows Initial Comment: While building python 2.4.3 with the current gcc trunk (soon to be 4.2), I uncovered a signed integer overflows bug in Python with the help of one of the gcc developers. The bug I observed is documented in this gcc mailing list message... http://gcc.gnu.org/ml/gcc/2006-08/msg00436.html The gcc developer comments about its origin are in the messages... http://gcc.gnu.org/ml/gcc/2006-08/msg00434.html http://gcc.gnu.org/ml/gcc/2006-08/msg00442.html which in short says... It *is* a bug in python, here is the proof: https://codespeak.net/viewvc/vendor/cpython/Python-r243/dist/src/ Objects/intobject.c?revision=25647&view=markup Function * i_divmod*(*register* *long* x, *register* *long* y, the following lines: / /* (-sys.maxint-1)/-1 is the only overflow case. *// *if* (y == -1 && x < 0 && x == -x) *return* DIVMOD_OVERFLOW; If overflow is barred then x==-x may happen only when x==0. This conflicts with x<0, which means that the compiler may assume that x<0 && x==-x always yields false. This may allow the compiler to eliminate the whole if statement. Hence, clearly python is at fault. ---------------------------------------------------------------------- >Comment By: Jack Howarth (jwhowarth) Date: 2006-08-24 15:22 Message: Logged In: YES user_id=403009 Everyone involved in reviewing this patch should definitely read the following sequence of gcc mailing list messages which show the process by which this patch was arrived at... http://gcc.gnu.org/ml/gcc/2006-08/msg00434.html http://gcc.gnu.org/ml/gcc/2006-08/msg00436.html http://gcc.gnu.org/ml/gcc/2006-08/msg00437.html http://gcc.gnu.org/ml/gcc/2006-08/msg00443.html http://gcc.gnu.org/ml/gcc/2006-08/msg00446.html http://gcc.gnu.org/ml/gcc/2006-08/msg00447.html http://gcc.gnu.org/ml/gcc/2006-08/msg00449.html So we have had lots of gcc developer eyes on this problem and they all agree on the flaw and the fix as posted. It's unfortunate that I had to abuse their mailing list to get this addressed before python 2.5 gets released. ---------------------------------------------------------------------- Comment By: Neal Norwitz (nnorwitz) Date: 2006-08-24 15:00 Message: Logged In: YES user_id=33168 I grepped for ' == .*-[^1>]' and ' != .*-[^1>]' and didn't spot any other occurrences. ---------------------------------------------------------------------- Comment By: Sjoerd Mullender (sjoerd) Date: 2006-08-24 14:54 Message: Logged In: YES user_id=43607 Just a comment, I'm not claiming this bug. The test (x < 0 && x == -x) tests whether x is equal to the smallest negative number. If x is equal to -2147483648, then -x is also equal to -2147483648 due to overflow. What does this version of gcc do with this code when x = -2147483648? ---------------------------------------------------------------------- Comment By: Neal Norwitz (nnorwitz) Date: 2006-08-24 14:54 Message: Logged In: YES user_id=33168 Assigning to Tim, he likes these problems. :-) He recently fixed a similar problem in another piece of code. I'm going to try to grep and see if I can find more occurrences of this. ---------------------------------------------------------------------- Comment By: Jack Jansen (jackjansen) Date: 2006-08-24 14:37 Message: Logged In: YES user_id=45365 Hmm... intobject.c doesn't really have a "champion"... Neal, I'm assigning this to you purely on the ground that you're the last person to have edited this file. Feel free to pass on (but not back:-). ---------------------------------------------------------------------- Comment By: Jack Howarth (jwhowarth) Date: 2006-08-24 11:22 Message: Logged In: YES user_id=403009 The was a few other comments from the gcc developers on the proposed fix... http://gcc.gnu.org/ml/gcc/2006-08/msg00448.html http://gcc.gnu.org/ml/gcc/2006-08/msg00449.html ...so that since x is a long the more correct fix is... --- Python-2.4.3/Objects/intobject.c.org 2006-08-24 07:06:51.000000000 -0400 +++ Python-2.4.3/Objects/intobject.c 2006-08-24 07:08:06.000000000 -0400 @@ -479,7 +479,7 @@ return DIVMOD_ERROR; } /* (-sys.maxint-1)/-1 is the only overflow case. */ - if (y == -1 && x < 0 && x == -x) + if (y == -1 && x < 0 && ((unsigned long)x) == -(unsigned long)x) return DIVMOD_OVERFLOW; xdivy = x / y; xmody = x - xdivy * y; I have tested this form of the patch and it works as well. My main concern is that we get this fix in python 2.5 before release. Jack, could you reassign this to the person you think might be most appropriate out of the list of python developers? I really should be a pretty simple review for the patch. ---------------------------------------------------------------------- Comment By: Jack Jansen (jackjansen) Date: 2006-08-24 09:14 Message: Logged In: YES user_id=45365 I don't think this is a mac-specific bug, and I'm also not really the right person to look into this... ---------------------------------------------------------------------- Comment By: Jack Howarth (jwhowarth) Date: 2006-08-24 04:13 Message: Logged In: YES user_id=403009 As suggested by another gcc developer in... http://gcc.gnu.org/ml/gcc/2006-08/msg00446.html ...the following patch eliminates the error when python is built with gcc trunk... --- Python-2.4.3/Objects/intobject.c.org 2006-08-23 23:49:33.000000000 -0400 +++ Python-2.4.3/Objects/intobject.c 2006-08-23 23:52:01.000000000 -0400 @@ -479,7 +479,7 @@ return DIVMOD_ERROR; } /* (-sys.maxint-1)/-1 is the only overflow case. */ - if (y == -1 && x < 0 && x == -x) + if (y == -1 && x < 0 && ((unsigned)x) == -(unsigned)x) return DIVMOD_OVERFLOW; xdivy = x / y; xmody = x - xdivy * y; This change allows python to completely pass its make check now when built with gcc trunk. ---------------------------------------------------------------------- You can respond by visiting: https://sourceforge.net/tracker/?func=detail&atid=105470&aid=1545668&group_id=5470 _______________________________________________ Python-bugs-list mailing list Unsubscribe: http://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
