[Bug middle-end/41149] New: -fdump-tree-original and procedure pointer components

[Bug middle-end/41149] New: -fdump-tree-original and procedure pointer components

[janus at gcc dot gnu dot org]

There seems to be a bug in the tree dump related to procedure pointer
components ('PPC', a Fortran 2003 feature which would probably be described as
"function pointers as structure components" in C). In particular: When calling
such a PPC, the dump only shows the component name, while the name of the
object which contains the component is omitted. This is not language dependent
and is also observed for C code, as the following example shows:


#include <stdio.h>

typedef long funcPtr(char *);

long func(char * str)
{
  printf("%s\n", str);
  return 0;
}

int main(void)
{
  long l;
  struct
  {
    funcPtr *ppc;
  } t;
  t.ppc = func;
  l = t.ppc("Hello");
  return l;
}


The dump shows (with -fdump-tree-original):

  t.ppc = (long int (*<T617>) (char *)) func;
  l = ppc ((char *) "Hello");

In the second line I would expect to see 't.ppc' instead of just 'ppc'.

For a Fortran example see PR 41139.


-- 
           Summary: -fdump-tree-original and procedure pointer components
           Product: gcc
           Version: 4.5.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: janus at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41149

[Bug middle-end/41149] -fdump-tree-original and procedure pointer components

[rguenth at gcc dot gnu dot org]

------- Comment #1 from rguenth at gcc dot gnu dot org  2009-08-24 09:21 -------
Seems like by design, see tree-pretty-print.c:print_call_name

Likely for printing prettier member function names.  IMHO we should just
drop this function on the floor considering it prints t.w (...) for

  struct
    {
      struct
        {
      funcPtr *ppc;
        } w;
    } t;
  t.w.ppc = func;
  l = t.w.ppc("Hello");

I'll happily approve a patch that passes bootstrap and testing that
reduces the function to strip an address-taking operation (note the
CALL_EXPR operand is an address of the function, so &a (...) should
print as a (...) and p (...) should print as (*p) (...)).


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |rguenth at gcc dot gnu dot
                   |                            |org
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|0                           |1
   Last reconfirmed|0000-00-00 00:00:00         |2009-08-24 09:21:23
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41149

[Bug middle-end/41149] -fdump-tree-original and procedure pointer components

[janus at gcc dot gnu dot org]

------- Comment #2 from janus at gcc dot gnu dot org  2009-08-24 19:44 -------
(In reply to comment #1)
> Seems like by design, see tree-pretty-print.c:print_call_name

Thanks for pointing me at the right place.


> Likely for printing prettier member function names.  IMHO we should just
> drop this function on the floor

I wouldn't go as far as dropping the whole function, but at least the part
concerning COMPONENT_REF seems pretty braindead to me. I'd propose to just drop
this and replace it by something more sane, like here:


Index: gcc/tree-pretty-print.c
===================================================================
--- gcc/tree-pretty-print.c     (revision 151053)
+++ gcc/tree-pretty-print.c     (working copy)
@@ -2705,19 +2705,6 @@ print_call_name (pretty_printer *buffer,
       dump_generic_node (buffer, TREE_OPERAND (op0, 2), 0, flags, false);
       break;

-    case COMPONENT_REF:
-      /* The function is a pointer contained in a structure.  */
-      if (TREE_CODE (TREE_OPERAND (op0, 0)) == INDIRECT_REF ||
-         TREE_CODE (TREE_OPERAND (op0, 0)) == VAR_DECL)
-       dump_function_name (buffer, TREE_OPERAND (op0, 1), flags);
-      else
-       dump_generic_node (buffer, TREE_OPERAND (op0, 0), 0, flags, false);
-      /* else
-        We can have several levels of structures and a function
-        pointer inside.  This is not implemented yet...  */
-      /*                 NIY;*/
-      break;
-
     case ARRAY_REF:
       if (TREE_CODE (TREE_OPERAND (op0, 0)) == VAR_DECL)
        dump_function_name (buffer, TREE_OPERAND (op0, 0), flags);
@@ -2725,6 +2712,7 @@ print_call_name (pretty_printer *buffer,
        dump_generic_node (buffer, op0, 0, flags, false);
       break;

+    case COMPONENT_REF:
     case SSA_NAME:
     case OBJ_TYPE_REF:
       dump_generic_node (buffer, op0, 0, flags, false);


This seems to do the right thing (also for multiply nested function pointers as
in comment #1). Regtesting now ...


-- 

janus at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |janus at gcc dot gnu dot org
                   |dot org                     |
             Status|NEW                         |ASSIGNED
   Last reconfirmed|2009-08-24 09:21:23         |2009-08-24 19:44:09
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41149

[Bug middle-end/41149] -fdump-tree-original and procedure pointer components

[janus at gcc dot gnu dot org]

------- Comment #3 from janus at gcc dot gnu dot org  2009-08-25 09:08 -------
The patch in comment #2 was successfully bootstrapped and regtested. Ok for
trunk?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41149

[Bug middle-end/41149] -fdump-tree-original and procedure pointer components

[rguenther at suse dot de]

------- Comment #4 from rguenther at suse dot de  2009-08-25 09:13 -------
Subject: Re:  -fdump-tree-original and procedure pointer
 components

On Tue, 25 Aug 2009, janus at gcc dot gnu dot org wrote:

> ------- Comment #3 from janus at gcc dot gnu dot org  2009-08-25 09:08 -------
> The patch in comment #2 was successfully bootstrapped and regtested. Ok for
> trunk?

Ok with a changelog entry.

Richard.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41149

[Bug middle-end/41149] -fdump-tree-original and procedure pointer components

[janus at gcc dot gnu dot org]

------- Comment #5 from janus at gcc dot gnu dot org  2009-08-25 09:35 -------
Subject: Bug 41149

Author: janus
Date: Tue Aug 25 09:35:41 2009
New Revision: 151075

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=151075
Log:
2009-08-25  Janus Weil  <janus@gcc.gnu.org>

        PR middle-end/41149
        * tree-pretty-print.c (print_call_name): Print the correct call name
        for procedure pointer components.


Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/tree-pretty-print.c


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41149

[Bug middle-end/41149] -fdump-tree-original and procedure pointer components

[janus at gcc dot gnu dot org]

------- Comment #6 from janus at gcc dot gnu dot org  2009-08-25 09:38 -------
Fixed with r151075. Closing.


-- 

janus at gcc dot gnu dot org changed:

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41149

[Bug middle-end/41149] -fdump-tree-original and procedure pointer components

[steven at gcc dot gnu dot org]

------- Comment #7 from steven at gcc dot gnu dot org  2009-08-25 11:21 -------
Please drop the patch in gcc-patches as well (just post it with a comment that
you've committed it to fix this PR).


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41149