[Bug lto/42528] New: ICE with -flto and -fsigned-char

[Bug lto/42528] New: ICE with -flto and -fsigned-char

[anton at samba dot org]

A gcc build from svn today (20091228) gets an ICE on the following code when
using -flto and -fsigned-char:

char *foo;
int main()
{
        foo = "bar";
}

# /gcc-test/bin/gcc -flto -fsigned-char foo.c
In function 'main':
lto1: error: type mismatch in address expression
<unnamed-unsigned:8> *

<unnamed-signed:8>

foo = &"bar"[0];

lto1: internal compiler error: verify_stmts failed


It passes with either option on its own.


-- 
           Summary: ICE with -flto and -fsigned-char
           Product: gcc
           Version: 4.5.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: lto
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: anton at samba dot org
GCC target triplet: powerpc64-linux-gnu-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42528

[Bug lto/42528] ICE with -flto and -fsigned-char

[rguenth at gcc dot gnu dot org]

------- Comment #1 from rguenth at gcc dot gnu dot org  2009-12-31 15:08 -------
Confirmed on i?86-linux with -funsigned-char instead.

I'll have a looksee.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |rguenth at gcc dot gnu dot
                   |dot org                     |org
             Status|UNCONFIRMED                 |ASSIGNED
     Ever Confirmed|0                           |1
           Keywords|                            |ice-on-valid-code
   Last reconfirmed|0000-00-00 00:00:00         |2009-12-31 15:08:45
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42528

[Bug lto/42528] ICE with -flto and -fsigned-char

[rguenth at gcc dot gnu dot org]

------- Comment #2 from rguenth at gcc dot gnu dot org  2010-01-03 14:33 -------
This is because va_list_type_node is char * on some targets and so
pointer-to char types get globbed into it.  But va_list_type_node is
re-constructed as char * at IL read time and thus we read back
unsigned char * as signed char *.

Oops.

I guess va_list_type_node better should be _not_ dependent on
-f[un]signed-char.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42528

[Bug lto/42528] ICE with -flto and -fsigned-char

[rguenth at gcc dot gnu dot org]

------- Comment #3 from rguenth at gcc dot gnu dot org  2010-01-03 17:31 -------
The same problem exists for builtin functions that return or take arguments
of type char *.  They change signature according to -f[un]signed-char which
causes for example

FAIL: gcc.c-torture/execute/builtins/strcpy-chk.c compilation,  -O2 -flto 
(internal compiler error)

when running with -funsigned-char on x86:

char p[32] = "";
int main ()
{
  if (__builtin___strcpy_chk (p + 1, "vwxyz",
                              __builtin_object_size (p + 1, 0)) != p + 1)
    __builtin_abort ();
  return 0;
} 

$ /obj/trunk-g/gcc/xgcc -B/obj/trunk-g/gcc/ strcpy-chk.3.i -O -flto
-funsigned-char
In file included from :0:0:
strcpy-chk.3.i: In function 'main':
strcpy-chk.3.i:2:5: error: non-trivial conversion at assignment
<unnamed-unsigned:8> *
<unnamed-signed:8> *
D.2023_3 = D.2027_5;

strcpy-chk.3.i:2:5: internal compiler error: verify_stmts failed
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42528

[Bug lto/42528] ICE with -flto and -fsigned-char

[rguenth at gcc dot gnu dot org]

------- Comment #4 from rguenth at gcc dot gnu dot org  2010-01-03 18:12 -------
Which seems to be because of CCP which folds

D.2023_3 = __builtin___strcpy_chk (&p[1], &"vwxyz"[0], 31);

with unsigned LHS to

D.2023_3 = (<unnamed-signed:8> *) __builtin_memcpy (&p[1], &"vwxyz"[0], 6);

which requires a temporary and thus exposes the mismatched types to the
verifier (we don't verify function assignments).

Which is also because lto1 doesn't understand -f[un]signed-char and always
has flag_signed_char set to the target default because

1044      /* Share char_type_node with whatever would be the default for the
target.
1045         char_type_node will be used for internal types such as
1046         va_list_type_node but will not be present in the lto stream.  */
1047      char_type_node
1048        = DEFAULT_SIGNED_CHAR ? signed_char_type_node :
unsigned_char_type_node;

the comment is true, but still units with different setting of
flag_signed_char will have different va_list_type_node and different
function signatures for builtin string functions.

Now the above together with LTO1 not recognizing -f[un]signed-char
even breaks consistent but non-standard flag_signed_char setting.
Whatever is more important ...


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42528

[Bug lto/42528] ICE with -flto and -fsigned-char

[rguenth at gcc dot gnu dot org]

------- Comment #5 from rguenth at gcc dot gnu dot org  2010-01-03 18:21 -------
Created an attachment (id=19455)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=19455&action=view)
patch

Patch I am going to test.  Further testing appreciated.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42528

[Bug lto/42528] ICE with -flto and -fsigned-char

[anton at samba dot org]

------- Comment #6 from anton at samba dot org  2010-01-07 11:01 -------
Thanks Richard, it passes my tests.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42528

[Bug lto/42528] ICE with -flto and -fsigned-char

[rguenth at gcc dot gnu dot org]

------- Comment #7 from rguenth at gcc dot gnu dot org  2010-01-08 16:58 -------
Subject: Bug 42528

Author: rguenth
Date: Fri Jan  8 16:57:59 2010
New Revision: 155740

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=155740
Log:
2010-01-08  Richard Guenther  <rguenther@suse.de>

        PR lto/42528
        * c.opt (fsigned-char): Also let LTO handle this option.
        (funsigned-char): Likewise.

        lto/
        * lto-lang.c (lto_handle_option): Handle -f[un]signed-char.
        (lto_init): Do not init char_type_node in a standard way
        but according to flag_signed_char.

        * gcc.dg/lto/20100103-1_0.c: New testcase.
        * gcc.dg/lto/20100103-2_0.c: Likewise.

Added:
    trunk/gcc/testsuite/gcc.dg/lto/20100103-1_0.c
    trunk/gcc/testsuite/gcc.dg/lto/20100103-2_0.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/c.opt
    trunk/gcc/lto/ChangeLog
    trunk/gcc/lto/lto-lang.c
    trunk/gcc/testsuite/ChangeLog


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42528

[Bug lto/42528] ICE with -flto and -fsigned-char

[rguenth at gcc dot gnu dot org]

------- Comment #8 from rguenth at gcc dot gnu dot org  2010-01-08 16:58 -------
Fixed.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|                            |FIXED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42528