[tree-ssa] labels after returns

law@redhat.com law@redhat.com
Mon Apr 12 16:46:00 GMT 2004


In message <1081787227.3970.71.camel@localhost.localdomain>, Diego Novillo writ
es:
 >On Mon, 2004-04-12 at 12:23, law@redhat.com wrote:
 >> In message <1081772243.3970.5.camel@localhost.localdomain>, Diego Novillo w
 >rite
 >> s:
 >>  >On Mon, 2004-04-12 at 02:24, Andrew Pinski wrote:
 >>  >
 >>  >> testcase from gcc.c-torture/compile/20021108-1.c:
 >>  >> int
 >>  >> main()
 >>  >> {
 >>  >> l1:
 >>  >>    return &&l1-&&l2;
 >>  >> l2:;
 >>  >> }
 >>  >> 
 >>  >I thought this was invalid and thus fair game to remove?
 >> No, they are not fair game to remove.  The labels are fair game to 
 >> move though.  [A user label which is not used in control flow altering
 >> ways must be preserved, but its location need not be preserved. ]
 >> 
 >Regardless.  Are they not invalid? 
It is valid code.

 >  Didn't you go through FC packages fixing these things up?
Err, I certainly didn't trip over this building the FC2 stuff.  If you look
at Andrew's testcase it's actually from c-torture and clearly there's some
target dependent lameness involved since we're not tripping over this 
(at least not in a clearly visible way from the testsuite's point of view)
on i686-pc-linux-gnu.

jeff
 >
 >
 >Diego.
 >




More information about the Gcc mailing list