This is the mail archive of the gcc@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]
Other format: [Raw text]

[tree-ssa] mainline merge as of 2003-08-17


This merge brings in the new pretty printer from mainline to replace
c-pretty-print.c.  The GENERIC/GIMPLE pretty printer is not changed, but
it uses the new pp_* functions.

New regressions in 00testsuite-linux-gnu/20030819/gcc.sum.gz:
	 FAIL: gcc.c-torture/execute/builtins/string-9.c execution
	 FAIL: gcc.c-torture/execute/builtins/string-9.c execution
	 FAIL: gcc.c-torture/execute/builtins/string-9.c execution
	 FAIL: gcc.c-torture/execute/builtins/string-9.c execution
	 FAIL: gcc.c-torture/execute/builtins/string-9.c execution
	 FAIL: gcc.c-torture/execute/builtins/string-9.c execution

These are the usual builtin regressions for simplifications we miss in
the branch.  They are not a high priority right now.


	 FAIL: gcc.dg/tree-ssa/20030807-3.c scan-tree-dump-times if  3
	 FAIL: gcc.dg/tree-ssa/20030807-4.c scan-tree-dump-times if  4

There must've been some FE or fold change in mainline that is causing
GIMPLE to emit different trees.  Previously, we would emit code like
this:

     if (overflow_2 != 0)
         {
           iftmp.12_22 = 1
         }
       else
         {
           iftmp.12_23 = 0
         };
       iftmp.8_24 = iftmp.12_7;
     };
     if (iftmp.12_7)
     {
         foo ()
	...

After the merge, we get:

   T.10_13 = overflow_2 != 0;
   T.11_14 = T.10_13;
   if (T.10_13)
     {
       foo ()

I adjusted the patterns expected by the dump scanner, so these tests
should still pass.


New regressions in 00testsuite-linux-gnu/20030819/g++.sum.gz:
	 FAIL: g++.dg/ext/stmtexpr1.C (test for excess errors)

In this testcase, the gimplifier is trying to create a temporary
variable of type 'struct A<6>', which is explicitly rejected because the
type is not complete at that point.


	 FAIL: g++.dg/template/error2.C (test for excess errors)

This one is also present in mainline.  It's an additional error not
expected by dejagnu.


	 FAIL: g++.dg/template/warn1.C (test for excess errors)

This is an extra "statement with no effect" warning emitted by the
tree-ssa compiler.


	 FAIL: g++.old-deja/g++.brendan/crash30.C (test for excess errors)

Another gimplifier problem.  It's getting confused by a
WITH_CLEANUP_EXPR.

I spoke with Jason about these failures and he agreed that they were not
merge blockers.


Fixed regressions in 00testsuite-linux-gnu/20030819/g++.sum.gz:
	 FAIL: g++.dg/opt/template1.C scan-assembler-not foo1


There were no other regressions nor fixes in the remaining testsuites.


Diego.


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