Re: Remove __builtin_classify_type and typeclass.h

[Gabriel Dos Reis <gdr at integrable-solutions dot net>]

Zack Weinberg <zack at codesourcery dot com> writes:

| Phil Edwards <phil at jaj dot com> writes:
| 
| > On Sun, Mar 09, 2003 at 01:44:10PM -0800, Zack Weinberg wrote:
| >> Phil Edwards <phil at jaj dot com> writes:
| >> 
| >> > On Sun, Mar 09, 2003 at 12:48:06PM +0000, Joseph S. Myers wrote:
| >> >> On Sat, 8 Mar 2003, Phil Edwards wrote:
| >> >> 
| >> >> > The typeclass.h header exists only to support __builtin_classify_type,
| >> >> > which used to be used for implementing variable argument lists, but is no
| >> >> > longer used in any platform's implementation.  Any reason to keep it?
| >> >> 
| >> >> __builtin_classify_type is used by glibc's <tgmath.h>.  (I don't know
| >> >> whether the uses of it are necessary - I only wrote the code to generate
| >> >> the correct tgmath type without using __builtin_classify_type
| >> >> (__tgmath_real_type etc.) - but the installed base has this use of the
| >> >> function.)
| >> >
| >> > Franz Sirl just pointed out the same thing.  Thanks, patch withdrawn.
| >> 
| >> You can at least remove typeclass.h.
| >
| > Nope, not if the builtin is staying.  Expanding and folding the builtin
| > uses the enumeration values defined in typeclass.h, even if the header
| > isn't installed (see builtin.c).
| 
| So it does.  Ugh.  I would suggest moving the enum into builtins.c
| except I'm not sure that's actually an improvement.  Other people have
| thoughts?

>From my perspective, the right long term thing to do is to remove that
file entirely.  Once the compile-time reflection stuff is in good
shape, I believe its benefits would be obvious to everybody, even
to those religiously against extensions  ;-)

-- Gaby