[Bug tree-optimization/21399] New: libstdc++ 12077.cc ICE

[Bug tree-optimization/21399] New: libstdc++ 12077.cc ICE

[dje at gcc dot gnu dot org]

libstdc++-v3/testsuite/ext/stdio_sync_filebuf/wchar_t/12077.cc ICEs when
compiled on AIX:

libstdc++-v3/testsuite/ext/stdio_sync_filebuf/wchar_t/12077.cc
: In function 'void test01()':
libstdc++-v3/testsuite/ext/stdio_sync_filebuf/wchar_t/12077.cc
:24: error: BB 3 can not throw but has EH edges
libstdc++-v3/testsuite/ext/stdio_sync_filebuf/wchar_t/12077.cc
:24: internal compiler error: verify_flow_info failed

-- 
           Summary: libstdc++ 12077.cc ICE
           Product: gcc
           Version: 4.1.0
            Status: UNCONFIRMED
          Keywords: ice-on-valid-code, ice-checking
          Severity: normal
          Priority: P2
         Component: tree-optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: dje at gcc dot gnu dot org
                CC: gcc-bugs at gcc dot gnu dot org
 GCC build triplet: powerpc-ibm-aix5.2.0.0
  GCC host triplet: powerpc-ibm-aix5.2.0.0
GCC target triplet: powerpc-ibm-aix5.2.0.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21399

[Bug tree-optimization/21399] libstdc++ 12077.cc ICE

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |pinskia at gcc dot gnu dot
                   |                            |org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21399

[Bug tree-optimization/21399] libstdc++ 12077.cc ICE

[dje at gcc dot gnu dot org]

------- Additional Comments From dje at gcc dot gnu dot org  2005-05-05 14:34 -------
Created an attachment (id=8824)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=8824&action=view)
pre-processed source for 12077.cc


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21399

[Bug tree-optimization/21399] libstdc++ 12077.cc ICE

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-05-05 14:35 -------
Also happens on ppc-darwin.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21399

[Bug tree-optimization/21399] libstdc++ 12077.cc ICE

[dje at gcc dot gnu dot org]

------- Additional Comments From dje at gcc dot gnu dot org  2005-05-05 14:37 -------
The failure appeared April 25.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21399

[Bug tree-optimization/21399] [4.1 Regression] libstdc++ 12077.cc ICE

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-05-05 14:55 -------
Confirmed, reduced testcase:
typedef struct __FILE FILE;
extern "C" int fputs(const char *, FILE *);
struct a { ~a(); };
void f(FILE* file)
{
  a b;
  const char* str = "a";
  fputs(str, file);
}

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
   Last reconfirmed|0000-00-00 00:00:00         |2005-05-05 14:55:48
               date|                            |
            Summary|libstdc++ 12077.cc ICE      |[4.1 Regression] libstdc++
                   |                            |12077.cc ICE
   Target Milestone|---                         |4.1.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21399

[Bug tree-optimization/21399] [4.1 Regression] libstdc++ 12077.cc ICE

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-05-05 15:01 -------
A slightly more reduced testcase:
typedef struct __FILE FILE;
extern "C" int fputs(const char *, FILE *);
void f(FILE* file) throw()
{
  const char* str = "a";
  fputs(str, file);
}

The problem is somehow converting fputs to fputc does not update the exceptions regions correctly.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21399

[Bug tree-optimization/21399] [4.1 Regression] libstdc++ 12077.cc ICE

[steven at gcc dot gnu dot org]

------- Additional Comments From steven at gcc dot gnu dot org  2005-05-16 00:10 -------
this fails for me with -O in CCP.  In .t18.copyrename1 we have: 
;; Function void f(FILE*) (_Z1fP6__FILE) 
 
 
Partition map 
 
 
void f(FILE*) (fileD.2015) 
Eh tree: 
   1 allowed_exceptions tree_label:<L0> 
{ 
  const charD.3 * strD.2018; 
 
  # BLOCK 0 
  # PRED: ENTRY (fallthru) 
  strD.2018_1 = &"a"[0]; 
  #   TMT.1D.2025_4 = V_MAY_DEF <TMT.1D.2025_3>; 
  fputs (strD.2018_1, fileD.2015_2); 
  goto <bb 2> (<L1>); 
  # SUCC: 1 (ab,eh) 2 (fallthru) 
 
  # BLOCK 1 
  # PRED: 0 (ab,eh) 
<L0>:; 
  #   VUSE <TMT.1D.2025_4>; 
  __cxa_call_unexpected (<<<exception object>>>); 
  # SUCC: 
 
  # BLOCK 2 
  # PRED: 0 (fallthru) 
<L1>:; 
  return; 
  # SUCC: EXIT 
 
} 
 
 
In .t18.ccp, we have the following replacement: 
Replaced fputs (str_1, file_2); 
 with __builtin_fputc (97, file_2); 
which is probably to blame for this. 
 

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21399

[Bug tree-optimization/21399] [4.1 Regression] libstdc++ 12077.cc ICE

[steven at gcc dot gnu dot org]

------- Additional Comments From steven at gcc dot gnu dot org  2005-05-16 00:13 -------
At the point of the ICE, I have this (from gdb's "p dump_tree_cfg(0)"): 
void f(FILE*) (fileD.2015) 
{ 
  const char * str; 
 
  # BLOCK 0 
  # PRED: ENTRY (fallthru,exec) 
  str_1 = &"a"[0]; 
  __builtin_fputc (97, file_2); 
  goto <bb 2> (<L1>); 
  # SUCC: 1 (ab,eh,exec) 2 (fallthru,exec) 
 
  # BLOCK 1 
  # PRED: 0 (ab,eh,exec) 
<L0>:; 
  __cxa_call_unexpected (<<<exception object>>>); 
  # SUCC: 
 
  # BLOCK 2 
  # PRED: 0 (fallthru,exec) 
<L1>:; 
  return; 
  # SUCC: EXIT 
 
} 

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21399

[Bug tree-optimization/21399] [4.1 Regression] libstdc++ 12077.cc ICE

[steven at gcc dot gnu dot org]

------- Additional Comments From steven at gcc dot gnu dot org  2005-05-16 00:16 -------
For CCP there is code to remove EH edges in tree-ssa-propagate.c, but 
it does not trigger: 
 
Breakpoint 3, substitute_and_fold (prop_value=0xf3e3a0) at 
tree-ssa-propagate.c:1063 
1063                  if (maybe_clean_eh_stmt (stmt)) 
(gdb) p debug_generic_stmt(stmt) 
#   TMT.1D.2025_4 = V_MAY_DEF <TMT.1D.2025_3>; 
__builtin_fputc (97, fileD.2015_2); 
 
$2 = void 
(gdb) p maybe_clean_eh_stmt (stmt) 
$3 = 0 '\0' 
(gdb)     

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21399

[Bug tree-optimization/21399] [4.1 Regression] libstdc++ 12077.cc ICE

[steven at gcc dot gnu dot org]

------- Additional Comments From steven at gcc dot gnu dot org  2005-05-16 00:26 -------
We have this code in tree-ssa-propagate.c:1052-1065: 
 
          if (did_replace) 
            { 
              fold_stmt (bsi_stmt_ptr (i)); 
              stmt = bsi_stmt(i); 
 
              /* If we folded a builtin function, we'll likely 
                 need to rename VDEFs.  */ 
              mark_new_vars_to_rename (stmt); 
 
              /* If we cleaned up EH information from the statement, 
                 remove EH edges.  */ 
              if (maybe_clean_eh_stmt (stmt)) 
                tree_purge_dead_eh_edges (bb); 
            } 
 
But maybe_clean_eh_stmt returns false.  Why?? 
 
Because we replace the entire statement apparently.  So we get to tree-eh.c: 
 
 
Breakpoint 4, substitute_and_fold (prop_value=0xf3e3a0) at 
tree-ssa-propagate.c:1054 
1054                  fold_stmt (bsi_stmt_ptr (i)); 
(gdb) p *bsi_stmt_ptr(i) 
$12 = 0x2a95896550 
(gdb) p debug_generic_stmt (*bsi_stmt_ptr(i)) 
#   TMT.1D.2025_4 = V_MAY_DEF <TMT.1D.2025_3>; 
fputs (&"a"[0], fileD.2015_2); 
$13 = void 
(gdb) cont 
Continuing. 
 
Breakpoint 3, substitute_and_fold (prop_value=0xf3e3a0) at 
tree-ssa-propagate.c:1063 
1063                  if (maybe_clean_eh_stmt (stmt)) 
(gdb) p stmt 
$14 = 0x2a95896d70 
1063                  if (maybe_clean_eh_stmt (stmt)) 
(gdb) p debug_generic_stmt(stmt) 
#   TMT.1D.2025_4 = V_MAY_DEF <TMT.1D.2025_3>; 
__builtin_fputc (97, fileD.2015_2); 
$8 = void 
(gdb) step 
maybe_clean_eh_stmt (stmt=0x2a95896d70) at tree-eh.c:2040 
2040      if (!tree_could_throw_p (stmt)) 
(gdb) next 
2041        if (remove_stmt_from_eh_region (stmt)) 
(gdb) 
2043      return false; 
(gdb)     
 
So we try to remove the _new_ statement from the EH region, and leave the 
old one right there. 
 

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21399

[Bug tree-optimization/21399] [4.1 Regression] libstdc++ 12077.cc ICE

[steven at gcc dot gnu dot org]

------- Additional Comments From steven at gcc dot gnu dot org  2005-05-16 00:40 -------
The same problem is in tree-ssa-dom.c:2975:  
      /* Try to fold the statement making sure that STMT is kept  
         up to date.  */  
      if (fold_stmt (bsi_stmt_ptr (si)))  
        {  
          stmt = bsi_stmt (si);  
  
Later on it calls maybe_clean_eh_stmt on the new stmt.  
  

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21399

[Bug tree-optimization/21399] [4.1 Regression] libstdc++ 12077.cc ICE

[steven at gcc dot gnu dot org]

------- Additional Comments From steven at gcc dot gnu dot org  2005-05-16 00:42 -------
We probably need something like, 
 
bool 
maybe_clean_eh_after_replacing_stmt (tree old_stmt, tree new_stmt) 
{ 
  if (!tree_could_throw_p (new_stmt)) 
    if (remove_stmt_from_eh_region (old_stmt)) 
      return true; 
  return false; 
} 
 
and use it in the two places I pointed out.  I think the others are safe. 
 

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21399

[Bug tree-optimization/21399] [4.1 Regression] libstdc++ 12077.cc ICE

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-05-16 00:44 -------
This patch works for me on this testcase, I have not done a full bootstrap/test yet:
Index: tree-ssa-propagate.c
===============================================================
====
RCS file: /cvs/gcc/gcc/gcc/tree-ssa-propagate.c,v
retrieving revision 2.19
diff -u -p -r2.19 tree-ssa-propagate.c
--- tree-ssa-propagate.c        11 May 2005 08:14:24 -0000      2.19
+++ tree-ssa-propagate.c        16 May 2005 00:43:18 -0000
@@ -558,7 +558,7 @@ get_rhs (tree stmt)
 bool
 set_rhs (tree *stmt_p, tree expr)
 {
-  tree stmt = *stmt_p, op;
+  tree stmt = *stmt_p, op, new_stmt;
   enum tree_code code = TREE_CODE (expr);
   stmt_ann_t ann;
   tree var;
@@ -621,7 +621,17 @@ set_rhs (tree *stmt_p, tree expr)
       /* Replace the whole statement with EXPR.  If EXPR has no side
         effects, then replace *STMT_P with an empty statement.  */
       ann = stmt_ann (stmt);
-      *stmt_p = TREE_SIDE_EFFECTS (expr) ? expr : build_empty_stmt ();
+      new_stmt = TREE_SIDE_EFFECTS (expr) ? expr : build_empty_stmt ();
+      {
+        /* Update the eh region for the new statement.  */
+        int old_eh_region = lookup_stmt_eh_region (stmt);
+       if (old_eh_region >= 0)
+         {
+           remove_stmt_from_eh_region (stmt);
+           add_stmt_to_eh_region (new_stmt, old_eh_region);
+         }
+      }
+      *stmt_p = new_stmt;
       (*stmt_p)->common.ann = (tree_ann_t) ann;
 
       if (TREE_SIDE_EFFECTS (expr))



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21399

[Bug tree-optimization/21399] [4.1 Regression] libstdc++ 12077.cc ICE

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-05-16 01:23 -------
Neither patches are fully complete and here is why,
even though fold creates a new CALL_EXPR, we still need to copy the eh_region the function can throw 
so Steven's patch is not complete, we don't check for this right now but it is hard to test for (maybe).
Second if we have a throw to a throw to a non throw function, we need to clean up the regions, so my 
patch in comment #12 is not complete.

I have a fix which I am writting up right now which combines the two problems, we have to change the 
interface for set_rhs and fold_stmt so they return if you need to removal of eh edges.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21399

[Bug tree-optimization/21399] [4.1 Regression] libstdc++ 12077.cc ICE

[rth at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |rth at gcc dot gnu dot org
                   |dot org                     |
             Status|NEW                         |ASSIGNED
   Last reconfirmed|2005-05-05 14:55:48         |2005-05-16 19:46:27
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21399

[Bug tree-optimization/21399] [4.1 Regression] libstdc++ 12077.cc ICE

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2005-05-16 23:14 -------
Subject: Bug 21399

CVSROOT:	/cvs/gcc
Module name:	gcc
Changes by:	rth@gcc.gnu.org	2005-05-16 23:14:02

Modified files:
	gcc            : ChangeLog tree-eh.c tree-flow.h tree-ssa-ccp.c 
	                 tree-ssa-dom.c tree-ssa-pre.c 
	                 tree-ssa-propagate.c 

Log message:
	PR tree-opt/21399
	* tree-eh.c (maybe_clean_or_replace_eh_stmt): Rename from
	maybe_clean_eh_stmt; take old stmt parameter.  Update EH
	region data structure to match replacement.
	* tree-flow.h: Update to match.
	* tree-ssa-ccp.c (execute_fold_all_builtins): Likewise.
	* tree-ssa-dom.c (optimize_stmt): Likewise.
	* tree-ssa-pre.c (eliminate): Likewise.
	* tree-ssa-propagate.c (substitute_and_fold): Likewise.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&r1=2.8806&r2=2.8807
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/tree-eh.c.diff?cvsroot=gcc&r1=2.38&r2=2.39
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/tree-flow.h.diff?cvsroot=gcc&r1=2.104&r2=2.105
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/tree-ssa-ccp.c.diff?cvsroot=gcc&r1=2.70&r2=2.71
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/tree-ssa-dom.c.diff?cvsroot=gcc&r1=2.110&r2=2.111
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/tree-ssa-pre.c.diff?cvsroot=gcc&r1=2.88&r2=2.89
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/tree-ssa-propagate.c.diff?cvsroot=gcc&r1=2.19&r2=2.20



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21399

[Bug tree-optimization/21399] [4.1 Regression] libstdc++ 12077.cc ICE

[rth at gcc dot gnu dot org]

------- Additional Comments From rth at gcc dot gnu dot org  2005-05-16 23:16 -------
Fixed.

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21399