This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

bct_p *is* used in loop.c:strength_reduction


	I have removed the ATTRIBUTE_UNUSED designation from the bct_p
parameter in loop.c:strength_reduce().  That parameter *is* used depending
on ifdef's.

	I think that Kaveh is using *much* too limited a set of platforms
to test for no-return and unused parameters and does not appear to be
checking the results visually for correctness.  By inspection in both
loop.c:strength_reduction() and jump.c:sets_cc0_p(), one can see that
these parameters are used in certain cases.

	I just found another in combine.c:can_combine_p().  All parameters
used within HAVE_cc0 ifdefs were marked with ATTRIBUTE_UNUSED. 

	These warning cleanups are happening much too automatically and
with far too little consideration.  Mark Mitchell and I should not need to
wade through massive patches to discover that these mistakes are being
introduced. 

	This is starting to introduce bugs into the compiler due to
misapplied optimizations.  Previously it was removing labels which at
least would fail at compile time.

	If the warnings output of a build on one or a few platforms are
going to be used as the basis for transforming the compiler sources in
this manner, this must stop.

	I am extremely concerned about the entire set of Oct 14 cleanup
changes which make this type of assumption throughout.  Again, all
parameters used within HAVE_cc0 ifdefs have been marked with
ATTRIBUTE_UNUSED.  Either all of these changes need to be re-checked or
the patch should be reverted.  Please let me know which is going to
happen.

David


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]