nothrow attribute w/ gcc

[daniel.santos]

../../../source/lib/allocators/headerless.cpp:757: error: attributes are not allowed on a function-definition

My understanding of __attribute__ (nothrow) is that it is an optimization feature to prevent the compiler from generating unwind code. I would recommend consulting an expert on this, but I believe the c++ throw() clause will accomplish the same thing (as in MSVC) and you can (and actually, must) stick it on both the declaration and the definition, whereas gcc only accepts the attribute on the definition.

See http://gcc.gnu.org/o...Attributes.html for the vague details.

--- lib/code_annotation.h   	(revision 10384)
+++ lib/code_annotation.h   	(working copy)
@@ -63,7 +63,8 @@
  * smaller and more efficient code.
  **/
 #if GCC_VERSION >= 303
-# define NOTHROW __attribute__((nothrow))
+//# define NOTHROW __attribute__((nothrow))
+# define NOTHROW throw()   	// special meaning, equivalent to __declspec(nothrow)
 #elif MSC_VERSION
 # define NOTHROW throw()   	// special meaning, equivalent to __declspec(nothrow)
 #else

Alternately, you can make a separate macros for the declaration & definition.

Edited October 10, 2011 by daniel.santos

[fcxSanya]

Hello.

Thanks for posting this topic and information. I created ticket on Trac and sent message to Jan, who is working on this code.

[janwas]

Thanks for the report! I was not aware about the restriction on definitions and still haven't found any mention of it.

The closest is http://www.mail-archive.com/commits@stdcxx.apache.org/msg00960.html, which also use the separate DEFINE/DECLARE approach. It's ugly, but preferable to possible pessimization via throw(), so I'll go with that approach.

Am about to commit a fix; sorry about the breakage.

[Shield Bearer]

Sorry to go off topic, but you wouldn't be Daniel Santos from the Glest forums, would you?