On Fri, 26 Feb 2010, Ralf Wildenhues wrote:
Thankfully, CFLAGS and CXXFLAGS are separate variables. So it sounds like
all we have to do is teach AC_PROG_CXX to decide if this is the Sun C++
compiler, and if so, use -g0 instead of -g. But I don't have quick access
to the Sun compiler, to write a reliable test for deciding between the two
ideas, so I'm relying on some help on that front.
I agree that also disabling inlining may be weird, and that the default
choices for CFLAGS and CXXFLAGS may make questionable assumptions about
bug distribution in (esp. past) compilers, but is it worth changing this
long-known semantics for that?
More, is it worth deviating from feature tests for this setting? Other
compilers accept -g0 too, for example GCC does and it doesn't do what
we would like to have.
The decision to pass -g as a default is strongly influenced by GCC's
almost unique ability to achieve full optimization with debugging
enabled. While Sun Studio will accept -g or -g0 (and provide a level
of debugging), either of these may have a negative influence on
optimizations. In my opinion, it does not make sense for debugging to
be applied to most non-GCC compilers as a default since the resulting
code may run slower and the default should provide well optimized
code.
There are other issues with Sun Studio as well. When I want to
profile GraphicsMagick I use CFLAGS '-g -xO3' because an optimization
level of less than -xO3 disables OpenMP support, and the profiler
needs -g in order to provide human readable output. Optimization
levels greater than -xO3 don't provide the profiler with the data it
needs.
Bob
--
Bob Friesenhahn
bfriesen@xxxxxxxxxxxxxxxxxxx, http://www.simplesystems.org/users/bfriesen/
GraphicsMagick Maintainer, http://www.GraphicsMagick.org/
_______________________________________________
Autoconf mailing list
Autoconf@xxxxxxx
http://lists.gnu.org/mailman/listinfo/autoconf
