On 11/08/2008 04:46 PM, Tom Tromey wrote:
>>>>>> "Jiri" == Jiri Slaby <[EMAIL PROTECTED]> writes:
>
> Jiri> while compiling slightly augmented cc1, I've found a poison
> Jiri> pragma bug. Here comes a trimmed example of the bug:
>
> Jiri> $ echo -e '#pragma GCC poison malloc\nstruct { int malloc; };'|gcc - -E
> -o/dev/null
> Jiri> <stdin>:2:14: error: attempt to use poisoned "malloc"
>
> Could you explain what you think the bug is?
Either its use in gcc sources or the way how the #pragma keyword is handled
(this can be hardly changed).
> My first reaction is that this is not a bug. #pragma poison works on
> identifiers, no matter what context they appear in (outside of system
> headers). For this reason, poison is not always useful.
> __attribute__((deprecated)) might be closer to what you want, though
> it can be hard to use that with system-declared functions like malloc.
I want to extend cpp to extract some more info (for backward mapping from
preprocessed_code into source_code+line+column) into xml while preprocessing and
I use gdome2 library. If I try to compile it, it fails due to:
struct _GMemVTable
{
gpointer (*malloc) (gsize n_bytes);
gpointer (*realloc) (gpointer mem,
gsize n_bytes);
void (*free) (gpointer mem);
/* optional; set to NULL if not used ! */
gpointer (*calloc) (gsize n_blocks,
from /usr/include/glib-2.0/glib/gmem.h. If I remove this structure compeltely,
everything is OK (it's unused anyway).
At least some option to disable these pragma checks inside cpp would be great.
(I suppose there is none, I found it neither in gcc.info nor cpp.info nor
anywhere else...)
