c++/6352: GCC 3.1 inliner regression from March

[gbeauchesne at mandrakesoft dot com]

>Number:         6352
>Category:       c++
>Synopsis:       GCC 3.1 inliner regression from March
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    unassigned
>State:          open
>Class:          ice-on-legal-code
>Submitter-Id:   net
>Arrival-Date:   Thu Apr 18 04:56:02 PDT 2002
>Closed-Date:
>Last-Modified:
>Originator:     Gwenole Beauchesne
>Release:        GNU C++ version 3.1 20020418 (prerelease) (i686-pc-linux-gnu)
>Organization:
>Environment:
i586-mandrake-linux-gnu
>Description:
The testcase below, distilled from Mesa 4.0.2 (si-glu/libnurbs/nurbtess/quicksort.cc), ICEs in current gcc-3.1-branch. This is a regression from gcc-3.0.4.

[gb@kernel gcc]$ ./cc1plus -O1 -finline-functions ~/vrac/qsort.cpp
 void f() void g() void f() void g()
/home/gb/vrac/qsort.cpp:11: Internal compiler error in genrtl_finish_function,
   at cp/semantics.c:2622
Please submit a full bug report,
with preprocessed source if appropriate.
See <URL:http://www.gnu.org/software/gcc/bugs.html> for instructions.
>How-To-Repeat:
Compile the following testcase with -O1 -finline-functions and the current gcc-3.1-branch.

[gb@kernel gb]$ cat ~/vrac/qsort.cpp
static void g();

void f()
{
  void g();
  g();
}

void g()
{
}

BTW, do you want the full preprocessed source from Mesa?
>Fix:
Either don't use inlining or remove the declaration of g() in f().

Ah, as I happen to have a few prebuilt CVS snapshots around, the following might help to spot of the regressing change.

1) It works with
GNU C++ version 3.1 20020315 (prerelease) (i686-pc-linux-gnu)

2) But fails as above-mentioned with 
GNU C++ version 3.1 20020316 (prerelease) (i686-pc-linux-gnu)
>Release-Note:
>Audit-Trail:
>Unformatted: