Re:Old GCC easter egg (aka implementation-defined behavior)
by pmjordan on Sunday October 03, 2010 @09:22AM
(#1752667)
Undefined and implementation-defined behaviour are tricky. They've been doing a better job of it recently, but I do wish compilers warned you more when encountering code with undefined behavior. Sometimes it's very hard or impossible to detect reliably, for example when doing pointer arithmetic acrobatics. But things like modifying a variable twice within a sequence point, or reading potentially uninitialised variables should have the compiler screaming at you by default.
Re:Old GCC easter egg (aka implementation-defined behavior)
by pmjordan on Sunday October 03, 2010 @09:58AM
(#1752709)
GCC 4 has seen big improvements in that area, especially with -Wall. Other compilers (except clang) not so much.
FWIW, GCC 4.2.1 on OSX 10.6 silently accepts
x += ++x;
by default. -Wall catches it, at least. I'm trying to think of a reason why the warning might not be on by default - I can only assume it's either lumped in with some other, less reliable undefined behaviour detection, such as aliasing. -Wall doesn't even catch trivial cases of aliasing though, such as
int* y = &x;
x += ++*y;
Maybe it's the default in a newer version, or this is Apple messing with me, not GNU.
Re:Old GCC easter egg (aka implementation-defined behavior)
by joe_bleau on Sunday October 03, 2010 @01:06PM
(#1753070)
PC-lint (http://www.gimpel.com/) gives a "Warning 564: variable 'x' depends on order of evaluation" for your first example, but is silent on the second.
FWIW, GCC 4.2.1 on OSX 10.6 silently accepts
by default. -Wall catches it, at least. I'm trying to think of a reason why the warning might not be on by default - I can only assume it's either lumped in with some other, less reliable undefined behaviour detection, such as aliasing. -Wall doesn't even catch trivial cases of aliasing though, such as Maybe it's the default in a newer version, or this is Apple messing with me, not GNU.