http://anonsvn.wireshark.org/viewvc/viewvc.cgi?view=rev&revision=53556
User: guy
Date: 2013/11/24 10:51 PM
Log:
According to
http://gcc.gnu.org/onlinedocs/gcc-4.8.2/gcc/Code-Gen-Options.html#Code-Gen-Options
-ftrapv "generates traps for signed overflow on addition, subtraction,
multiplication operations." and -fwrapv "instructs the compiler to
assume that signed arithmetic overflow of addition, subtraction and
multiplication wraps around using twos-complement representation."
Those seem mutually-exclusive to me, and we probably want wrapping, not
traps, as there's probably a fair bit of code out there that explicitly
or implicitly assumes wrapping. (Actually, we really want to avoid
signed arithmetic for the cases that most matter, such as offsets and
lengths, but, unfortunately, we currently have API conventions that
allow negative values for lengths, either with -1 meaning "to the end"
or with negative values meaning "relative to the end".) In addition,
there seem to be some bugs complaining that -ftrapv doesn't always cause
traps on signed integer overflow.
We seem to be seeing crashes in Lemon on the Solaris buildbot subsequent
to adding -ftrapv; I don't know whether that's an overflow being
detected, a bug in the compiler, or something unrelated, especially
given that we're using Sun C, not GCC, on the Solaris buildbot.
However, we'll try removing -ftrapv, to see if it fixes the problem; the
MIT CSAIL paper in question wasn't really recommending all the GCC
options it mentioned (which, as noted, wouldn't make sense, as -ftrapv
and -fwrapv appear to be mutually-exclusive).
Directory: /trunk/
Changes Path Action
+0 -1 CMakeLists.txt Modified
+0 -1 configure.ac Modified