target_flags conflict in gcc/config/mips/mips.h

[Gavin Romig-Koch <gavin at cygnus dot com>]

Chris G. Demetriou writes:
 > Reading the mips GCC configuration bits in the 2000403 snapshot, i
 > noticed the following in gcc/config/mips/mips.h:
 > 
 > #define MASK_UNINIT_CONST_IN_RODATA 0x10000000  /* Store uninitialized
 >                                                    consts in rodata */
 > 
 > #define MASK_DEBUG_C    0x10000000      /* don't expand seq, etc. */
 > 
 > #define TARGET_DEBUG_C_MODE     (target_flags & MASK_DEBUG_C)
 > 
 > #define TARGET_UNINIT_CONST_IN_RODATA   (target_flags & MASK_UNINIT_CONST_IN_RODATA)                             
 > 
 > I don't think that'll have a desirable effect.  8-)

No, not desirable at all. 

 > I'm new to the whole set of mips config bits, so i dunno what the
 > right patch is.  (move the debug flags up a bit?  what happens when you run over 32 bits?)

The MASK_DEBUG_C flag should have been set to zero like a number of
other MASK_DEBUG_<x> flags, when TARGET_UNINIT_CONST_IN_RODATA was
added.

If you decide to fix this yourself (and I encorage you to do so),
you might as well zero out all the MASK_DEBUG_<x>s.

A few years ago I made the decision to keep the framework for
these DEBUG flags, and simply start zeroing out the MASK flags
as we needed bits, on the theory that eventually one or more
of these flags might turn out to be useful.  As yet none of
them have, at least for me anyway.  Eventually we will need to
clean this up, and someone (me if no one else does it) will have
to decide which of these to keep, and get renamed to something
useful, and which should just go.

When we run over 32, we have to rewrite the whole mess.




                                          -gavin...