This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Request for suggestions: xm-romp.h
- To: gcc at gcc dot gnu dot org
- Subject: Request for suggestions: xm-romp.h
- From: "Zack Weinberg" <zackw at stanford dot edu>
- Date: Fri, 16 Mar 2001 14:04:07 -0800
xm-romp.h's complete functional content is
#ifdef __HIGHC__
#define FUNCTION_CONVERSION_BUG
#endif
This is the only definition of FUNCTION_CONVERSION_BUG in the entire
source tree. Its only use is in expr.h:
#ifdef FUNCTION_CONVERSION_BUG
/* Some compilers fail to convert a function properly to a
pointer-to-function when used as an argument.
So produce the pointer-to-function directly.
Luckily, these compilers seem to work properly when you
call the pointer-to-function. */
#define GEN_FCN(CODE) (insn_data[(int) (CODE)].genfun)
#else
#define GEN_FCN(CODE) (*insn_data[(int) (CODE)].genfun)
#endif
I would like to solicit suggestions on what to do with this. With the
two patches I just sent, xm-romp.h is the only host cpu header left,
and I'd like to kill it too. Options that occur to me include:
- move the above #ifdef __HIGHC__ to system.h.
- move it to expr.h. (Possibly cleaner given it's only used there and
this doesn't come up much.)
- write an autoconf test. (Massive overkill; we're talking about an
obscure bug in a single ancient vendor compiler.)
- drop support for ROMP. (I'd rather not; I have a soft spot for the
things, since I salvaged an IBM RT from the old student union at
Columbia and used it as a terminal for awhile. And it's not like
the port is otherwise broken, far as I can tell, at least for C.
But on the other hand it is a chip that was discontinued in 1991...)
zw