[rfc patch] nomem: internal_error -> error

[rfc patch] nomem: internal_error -> error

[Jan Kratochvil]

Hi,

unfortunately I see this problem reproducible only with the
archer-jankratochvil-vla branch (VLA = Variable Length Arrays - char[var]).
OTOH this branch I hopefully submit in some form for FSF GDB later.

In this case (a general problem but tested for example on Fedora 13 i686):

int
main (int argc, char **argv)
{
  char a[argc];
  return a[0];
}

(gdb) start
(gdb) print a
../../gdb/utils.c:1251: internal-error: virtual memory exhausted: can't allocate 4294951689 bytes.

It is apparently because boundary for the variable `a' is not initialized
there.  Users notice it due to Eclipse-CDT trying to automatically display all
the local variables on each step.


Apparentl no regressions on {x86_64,x86_64-m32,i686}-fedora13-linux-gnu.
But is anone aware of the reasons to use internal_error there?
I find simple error as a perfectly reasonable there.
(history only tracks it since the initial import)

IIRC this idea has been discussed with Tom Tromey, not sure of its origin.

I understand it may be offtopic for FSF GDB but from some GDB crashes I am not
sure if it can happen only due to the VLA variables.


Thanks,
Jan


gdb/
2010-06-01  Jan Kratochvil  <jan.kratochvil@redhat.com>
	    Tom Tromey  <tromey@redhat.com>

	* utils.c (nomem): Change internal_error to error.

--- a/gdb/utils.c
+++ b/gdb/utils.c
@@ -1265,15 +1265,9 @@ void
 nomem (long size)
 {
   if (size > 0)
-    {
-      internal_error (__FILE__, __LINE__,
-		      _("virtual memory exhausted: can't allocate %ld bytes."),
-		      size);
-    }
+    error (_("virtual memory exhausted: can't allocate %ld bytes."), size);
   else
-    {
-      internal_error (__FILE__, __LINE__, _("virtual memory exhausted."));
-    }
+    error (_("virtual memory exhausted."));
 }
 
 /* The xmalloc() (libiberty.h) family of memory management routines.

Re: [rfc patch] nomem: internal_error -> error

[Mark Kettenis]

> Date: Tue, 1 Jun 2010 18:48:08 +0200
> From: Jan Kratochvil <jan.kratochvil@redhat.com>
> 
> Hi,
> 
> unfortunately I see this problem reproducible only with the
> archer-jankratochvil-vla branch (VLA = Variable Length Arrays - char[var]).
> OTOH this branch I hopefully submit in some form for FSF GDB later.
> 
> In this case (a general problem but tested for example on Fedora 13 i686):
> 
> int
> main (int argc, char **argv)
> {
>   char a[argc];
>   return a[0];
> }
> 
> (gdb) start
> (gdb) print a
> ../../gdb/utils.c:1251: internal-error: virtual memory exhausted: can't allocate 4294951689 bytes.
> 
> I find simple error as a perfectly reasonable there.

I disagree; this internal-error really is a bug in GDB.  We should
never try to allocate that much memory.

Re: [rfc patch] nomem: internal_error -> error

[Jan Kratochvil]

On Tue, 01 Jun 2010 19:55:40 +0200, Mark Kettenis wrote:
> > (gdb) print a
> > ../../gdb/utils.c:1251: internal-error: virtual memory exhausted: can't allocate 4294951689 bytes.
> > 
> > I find simple error as a perfectly reasonable there.
> 
> I disagree; this internal-error really is a bug in GDB.  We should
> never try to allocate that much memory.

So that even if the object is really so large in the inferior GDB should be
fixed to read+process the object in multiple parts?  That makes sense.


Thanks,
Jan