[Bug lto/52779] New: [4.6 only] LTO produces false warning with type merging

[Bug lto/52779] New: [4.6 only] LTO produces false warning with type merging

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52779

             Bug #: 52779
           Summary: [4.6 only] LTO produces false warning with type
                    merging
    Classification: Unclassified
           Product: gcc
           Version: 4.6.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: lto
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: pinskia@gcc.gnu.org


Take the following three files:
--- t0.c ----
void g1 (void) {
    __builtin_printf("+++\n");
}
--- CUT ---
--- t1.c ---
 const char glbl_str4[] = "New";
--- CUT ---
--- t2.c ---
extern const char glbl_str4[];

void g0 (void) {
    f0(glbl_str4);
}
--- CUT ---
Using "gcc -flto -O3 -r -nostdlib t0.c t1.c t2.c" produces the following
warning:
t2.c:1:19: warning: type of ‘glbl_str4’ does not match original declaration
[enabled by default]
t1.c:1:13: note: previously declared here
---- CUT ----
Note the order of the files is required so is the semi unrelated t0.c file.

[Bug lto/52779] [4.6 only] LTO produces false warning with type merging

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52779

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> 2012-03-30 01:39:59 UTC ---
A better understanding of what happens.
Arrays in both the C and C++ front-end are treated in an interesting way.  The
array's main variant is the array of the element types' main variant.
That is the const char[4]'s main variant is char[4].  But when we doing the
folding for printf to puts, this no longer holds for that const char[4] type.  
So when we do the merging of the type const char[4] for the first TU to the
second one, we have created an array type with a main variant which is
different from what the C/C++ front-end produces.
And then comes the last TU, where we have this main variant that is different. 
So we warn that char and const char are different, even if we don't care about
them being different variants because the front-end does not.

4.7 fixes it by using types_compatible instead of gimple_types_compatible.  Now
we still have this main variant issue that is latent in 4.7 and above.  Should
we fix that issue or just ignore it and how it does not matter?

[Bug lto/52779] [4.6 only] LTO produces false warning with type merging

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52779

Richard Guenther <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|                            |FIXED
   Target Milestone|---                         |4.7.0

--- Comment #2 from Richard Guenther <rguenth at gcc dot gnu.org> 2012-03-30 09:10:47 UTC ---
There is a duplicate of this bug which was fixed for 4.7.  There is even a
testcase somewhere.  We can't fix this in 4.6, thus, fixed for 4.7.