This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

G++ defines _GNU_SOURCE, implies C99

I've been reading the gcc archive, gnats bug reports (2082), 
and the FAQ about _GNU_SOURCE always being defined in g++.
In features.h, if _GNU_SOURCE is defined, then _USE_ISOC99 is 
defined, which means that headers will contain extensions from

The FAQ only mentions this being a problem on Solaris, but
it is also a problem on Linux -- for example, _GNU_SOURCE is defined
in gcc/config/rs6000/linux.h, with the comment that this is a hack 
which will be corrected.   (By the way, the FAQ is misleading in 
this respect:  g++ -E -dM - < /dev/null does not show that _GNU_SOURCE 
is predefined; but g++ -E -dM empty.c will.) 

Oddly enough, IMO, PR 2082 (and subsequent duplicates) are closed.
The explaination for why this bug is closed is really not clear to
me:  features.h has to be processed with the correct switches. 

G++ 3.2 breaks the following conforming C++ code:

	#include <math.h>        (or <cmath>)
	int round;

C99 introduced a new function named round, as well as others.
ISO C++ does not define a function named round, and, as far
as I'm aware, the C++ standard doesn't incorporate C99 changes.

So, can anyone elucidate why _GNU_SOURCE is defined by G++ and 
not by GCC and/or why it implies _USE_ISOC99?  And why this is 
not a bug?

Michael Eager	408-328-8426	
MontaVista Software, Inc. 1237 E. Arques Ave., Sunnyvale, CA  94085

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]