Re: allow plugins to disable poison

[Paolo Bonzini <bonzini at gnu dot org>]

> > While I don't disagree with this change in principle, I think malloc
> > and realloc should be the only poisonings disabled. Also, it would be
> > nice if this was done automatically for all plugins, for example by
> > doing in gcc-plugin.h
> I disagree. GCC poisons a lot of identifiers and the risk of collisions
> with other headers is fairly high.

You're right in that these could also be left aside

 #pragma GCC poison calloc strdup
 #pragma GCC poison ANSI_PROTOTYPES PTR_CONST LONG_DOUBLE VPARAMS VA_OPEN \
  VA_FIXEDARG VA_CLOSE VA_START
 #pragma GCC poison bcopy bzero bcmp rindex

(together with the #undef that they follow), but not the eliminated 
parts of the GCC API.  This may be moot now, but not as soon as plugins 
will have been part of GCC for a few versions.

Paolo