bug in expr.c?

bug in expr.c?

[Robert Lipe]

I was cleaning up warnings expr.c and believe this to be a bug.   Can
anyone that actually understands this confirm or deny this?

expand_assignment()

      if (TREE_CODE (to) == COMPONENT_REF
          && TREE_READONLY (TREE_OPERAND (to, 1)))
        {
          if (offset = 0)
            to_rtx = copy_rtx (to_rtx);

          RTX_UNCHANGING_P (to_rtx) = 1;
        }

That copy_rtx will never ever be called, right?

Is the "obvious" fix of "offset == 0" correct?

Thanx,
RJL

Re: bug in expr.c?

[Jeffrey A Law]

  In message < 19980209233414.16572@dgii.com >you write:
  > I was cleaning up warnings expr.c and believe this to be a bug.   Can
  > anyone that actually understands this confirm or deny this?
  > 
  > expand_assignment()
  > 
  >       if (TREE_CODE (to) == COMPONENT_REF
  >           && TREE_READONLY (TREE_OPERAND (to, 1)))
  >         {
  >           if (offset = 0)
  >             to_rtx = copy_rtx (to_rtx);
  > 
  >           RTX_UNCHANGING_P (to_rtx) = 1;
  >         }
  > 
  > That copy_rtx will never ever be called, right?
  > 
  > Is the "obvious" fix of "offset == 0" correct?
Correct.  If you look a little above that line you'll see something like
this:

              /* When the offset is zero, to_rtx is the address of the
                 structure we are storing into, and hence may be shared.
                 We must make a new MEM before setting the volatile bit.  */
              if (offset == 0)
                to_rtx = copy_rtx (to_rtx);

              MEM_VOLATILE_P (to_rtx) = 1;

It's basically the same issue, except that we're setting the readonly bit.


jeff

Re: bug in expr.c?

[Robert Lipe]

At Jeff's suggestion I've now committed this fix into EGCS.  Only the
development branch is affected.   1.0.2 didn't inherit this typo from
a 2.8.x import.

RJL


Mike Stump wrote:
> Here is what kenner said:
> 
> > Date: Thu, 12 Feb 98 10:10:50 EST
> > From: kenner@vlsi1.ultra.nyu.edu (Richard Kenner)
> > Message-Id: <9802121510.AA08266@vlsi1.ultra.nyu.edu>
> > To: mrs@wrs.com
> > Cc: robertl@dgii.com
> 
> > Yes, that was a clear typo.  Thanks.  I fixed it.
> 
> Code now reads:
> 
>         {
>           if (offset == 0)
>             to_rtx = copy_rtx (to_rtx);
>  
>           RTX_UNCHANGING_P (to_rtx) = 1;
>         }
> 
> at the FSF.
> 
> 

-- 
Robert Lipe       http://www.dgii.com/people/robertl       robertl@dgii.com

Re: bug in expr.c?

[Mike Stump]

Here is what kenner said:

> Date: Thu, 12 Feb 98 10:10:50 EST
> From: kenner@vlsi1.ultra.nyu.edu (Richard Kenner)
> Message-Id: <9802121510.AA08266@vlsi1.ultra.nyu.edu>
> To: mrs@wrs.com
> Cc: robertl@dgii.com

> Yes, that was a clear typo.  Thanks.  I fixed it.

Code now reads:

        {
          if (offset == 0)
            to_rtx = copy_rtx (to_rtx);
 
          RTX_UNCHANGING_P (to_rtx) = 1;
        }

at the FSF.

Re: bug in expr.c?

[Mike Stump]

This work was put in by kenner at the FSF, he might be able to say for
sure:

Thu Nov 20 16:37:36 1997  Richard Kenner  <kenner@vlsi1.ultra.nyu.edu>
 
        * expr.c (expand_assignment): If assigning to readonly field,
        mark rtx as unchanging.

revision 1.463
date: 1997/11/20 23:38:15;  author: kenner;  state: Exp;  lines: +9 -0
(expand_assignment): If assigning to readonly field, mark rtx as unchanging.

> Date: Mon, 9 Feb 1998 23:34:14 -0600
> From: Robert Lipe <robertl@dgii.com>
> To: egcs@cygnus.com

> I was cleaning up warnings expr.c and believe this to be a bug.   Can
> anyone that actually understands this confirm or deny this?

> expand_assignment()

>       if (TREE_CODE (to) == COMPONENT_REF
>           && TREE_READONLY (TREE_OPERAND (to, 1)))
>         {
>           if (offset = 0)
>             to_rtx = copy_rtx (to_rtx);

>           RTX_UNCHANGING_P (to_rtx) = 1;
>         }

> That copy_rtx will never ever be called, right?

> Is the "obvious" fix of "offset == 0" correct?