This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: fixups for calls to gen_rtx()
- To: davem at jenolan dot rutgers dot edu (David S. Miller)
- Subject: Re: fixups for calls to gen_rtx()
- From: Joern Rennecke <amylaar at cygnus dot co dot uk>
- Date: Wed, 1 Oct 1997 19:07:27 +0100 (BST)
- Cc: rth at cygnus dot com, law at cygnus dot com, egcs at cygnus dot com
> From: Richard Henderson <rth@cygnus.com>
> Date: Mon, 22 Sep 1997 11:20:57 -0700 (PDT)
>
> I disagree; varargs is slow. There is no need to be dropping the
> last four register arguments into memory just so we can pull them
> out again.
>
> I honestly don't care how slow it is, I really want the type checking.
In some cases the types will be silently promoted to the correct type,
unless you use -Wtraditional (which would probably produce a lot of noise,
because the gcc source contains some ANSI specific code that is
conditionally used for compilation with an ANSI compiler).
Thus, to avoid problems bootstrapping with an K&R compiler, I suggest
that the functions are named differently for a K&R compilation, and
macros are defined that call the actual gen_rtx_* functions with
all arguments casted to the proper type - or if you want to get more
sophisticated, cast only to types that can be the result of a
prototype-guided argument promotion.