[Bug lto/43891] New: ICE in expand_call_inline, at tree-inline.c:3820 with -fwhopr, missed fatal diagnostic

["rguenth at gcc dot gnu dot org" <gcc-bugzilla at gcc dot gnu dot org>]

For

1.c
------
extern float f(void);

int main(void)
{
  return f();
}

2.c
------
int f(void)
{
  return 0;
}

we ICE when building with -O2 -fwhopr because we inline f() and the
return value compatibility check does not trigger there (see comment
in fixed PR43455).

When inlining we use a FLOAT_EXPR to convert the return value to
the expected (float) type - this is bogus in general.  There are
two cases: if the ABI returns both types in the same way then
a VIEW_CONVERT_EXPR should be used to make inline vs. non-inline
behavior the same, if the ABI returns in different places then
the value returned is undefined.

We should try to emit a diagnostic here.

Runtime behavior of the non-LTO build should be preserved by not
inlining such calls also with -fwhopr.


-- 
           Summary: ICE in expand_call_inline, at tree-inline.c:3820 with -
                    fwhopr, missed fatal diagnostic
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Keywords: lto
          Severity: normal
          Priority: P3
         Component: lto
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: rguenth at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43891