Re: use of __builtin_trap

[Paul Eggert]

Bruno Haible wrote:
> Would it be possible to use abort () instead of __builtin_trap ()?

It would, but I'm not sure it'd be a good idea, as assume(X) is not intended to 
be a safety check: it's intended to be a message from the programmer to static 
analysis tools such as GCC's optimizer and/or warner.

assume(X) uses __builtin_trap only in GCC versions 3.3.4 through 4.4.7, which 
are so old that it's fair to call them obsolescent. In more-recent versions of 
GCC, assume(X) uses __builtin_unreachable, which means that when X is false the 
machine code can do anything at all. The main reason __builtin_trap is called on 
older GCCs is to pacify them when warnings are enabled. 'abort' is no better for 
that purpose, and to some extent is worse as it brings in stdlib.h and typically 
generates even more more useless bloat in the executable.

I suppose we could remove the call to __builtin_trap on older GCCs, and have 
assume (X) be a complete no-op with GCC 3.3.4 through 4.4.7. But this would 
cause even more chatter when people compile with warnings on these obsolescent 
platforms (a mistake if you ask me, but people do it...), and this cost exceeds 
any benefit I can see from removing the call.