gdb can't print array element for fortran

gdb can't print array element for fortran

[Chandru]

This mail pertains to the discussion had on gdb mailing list sometime
back here  http://sourceware.org/ml/gdb-patches/2009-01/msg00251.html.
After a small discussion with Jan I made the following changes and 
these didn't seem to work. Any inputs or thoughts would be helpful. 

thanks
Chandru

 
--- src/gdb/gdbtypes.c.orig	2010-02-25 00:00:19.000000000 +0530
+++ src/gdb/gdbtypes.c	2010-02-25 00:03:16.000000000 +0530
@@ -727,6 +727,14 @@ create_range_type (struct type *result_t
   if (low_bound >= 0)
     TYPE_UNSIGNED (result_type) = 1;
 
+  if (high_bound < low_bound)
+    {
+      if (high_bound == -1)
+        TYPE_HIGH_BOUND_UNDEFINED(result_type) = 1;
+      else
+        TYPE_HIGH_BOUND_UNDEFINED(result_type) = 0;
+    }
+
   return result_type;
 }
 
--- src/gdb/valarith.c.orig	2010-02-25 00:00:26.000000000 +0530
+++ src/gdb/valarith.c	2010-02-25 00:04:37.000000000 +0530
@@ -198,7 +198,7 @@ value_subscripted_rvalue (struct value *
   unsigned int elt_offs = elt_size * longest_to_int (index - lowerbound);
   struct value *v;
 
-  if (index < lowerbound || elt_offs >= TYPE_LENGTH (array_type))
+  if (index < lowerbound || ((elt_offs >= TYPE_LENGTH (array_type)) && !TYPE_HIGH_BOUND_UNDEFINED(array_type)))
     error (_("no such vector element"));
 
   v = allocate_value (elt_type);

Re: gdb can't print array element for fortran

[Joel Brobecker]

> This mail pertains to the discussion had on gdb mailing list sometime
> back here  http://sourceware.org/ml/gdb-patches/2009-01/msg00251.html.
> After a small discussion with Jan I made the following changes and 
> these didn't seem to work. Any inputs or thoughts would be helpful. 

Does anyone else have an opinion on this patch.  It looks wrong to me
(sorry Chandru): Arrays with a high bound that's less than the low bound
can exist, at least in Ada. It's just an empty array.

I haven't looked at Jan's work for the Fortran VLA stuff in a while,
so I can't remember how he dealt with this issue.  But I think his
approach might be more correct.

> --- src/gdb/gdbtypes.c.orig	2010-02-25 00:00:19.000000000 +0530
> +++ src/gdb/gdbtypes.c	2010-02-25 00:03:16.000000000 +0530
> @@ -727,6 +727,14 @@ create_range_type (struct type *result_t
>    if (low_bound >= 0)
>      TYPE_UNSIGNED (result_type) = 1;
>  
> +  if (high_bound < low_bound)
> +    {
> +      if (high_bound == -1)
> +        TYPE_HIGH_BOUND_UNDEFINED(result_type) = 1;
> +      else
> +        TYPE_HIGH_BOUND_UNDEFINED(result_type) = 0;
> +    }
> +
>    return result_type;
>  }
>  
> --- src/gdb/valarith.c.orig	2010-02-25 00:00:26.000000000 +0530
> +++ src/gdb/valarith.c	2010-02-25 00:04:37.000000000 +0530
> @@ -198,7 +198,7 @@ value_subscripted_rvalue (struct value *
>    unsigned int elt_offs = elt_size * longest_to_int (index - lowerbound);
>    struct value *v;
>  
> -  if (index < lowerbound || elt_offs >= TYPE_LENGTH (array_type))
> +  if (index < lowerbound || ((elt_offs >= TYPE_LENGTH (array_type)) && !TYPE_HIGH_BOUND_UNDEFINED(array_type)))
>      error (_("no such vector element"));
>  
>    v = allocate_value (elt_type);

-- 
Joel

Re: gdb can't print array element for fortran

[Tom Tromey]

>>>>> "Joel" == Joel Brobecker <brobecker@adacore.com> writes:

>> This mail pertains to the discussion had on gdb mailing list sometime
>> back here  http://sourceware.org/ml/gdb-patches/2009-01/msg00251.html.
>> After a small discussion with Jan I made the following changes and 
>> these didn't seem to work. Any inputs or thoughts would be helpful. 

Joel> Does anyone else have an opinion on this patch.  It looks wrong to me
Joel> (sorry Chandru): Arrays with a high bound that's less than the low bound
Joel> can exist, at least in Ada. It's just an empty array.

Additionally, if the proposed changes don't actually work, I guess some
other approach is needed anyhow.

Joel> I haven't looked at Jan's work for the Fortran VLA stuff in a while,
Joel> so I can't remember how he dealt with this issue.  But I think his
Joel> approach might be more correct.

I too have lost track of this, but my impression is that Jan's current
patch is rather larger.

Tom

Re: gdb can't print array element for fortran

[Jan Kratochvil]

On Mon, 15 Mar 2010 21:51:48 +0100, Tom Tromey wrote:
> Joel> I haven't looked at Jan's work for the Fortran VLA stuff in a while,
> Joel> so I can't remember how he dealt with this issue.  But I think his
> Joel> approach might be more correct.
> 
> I too have lost track of this, but my impression is that Jan's current
> patch is rather larger.

The VLA patch of mine does a real calculation of the dynamic sizes.  This
patch just considers dynamically-sized arrays as unbound.

This patch will be superseded by the VLA patch.  But I do not have the VLA
patch ready now in a suitable form for FSF GDB.  Fixed the Chandru's patch if
this partial functionality is enough on its own.

No regressions on {x86_64,x86_64-m32,i686}-fedora12-linux-gnu.  The
functionality is already tested by the testsuite so the patch does:
-FAIL: gdb.fortran/array-element.exp: print the first element of array a
+PASS: gdb.fortran/array-element.exp: print the first element of array a


Thanks,
Jan


2010-03-16  Jan Kratochvil  <jan.kratochvil@redhat.com>
	    Chandru <chandru@in.ibm.com>

	* dwarf2read.c (read_subrange_type): Set TYPE_HIGH_BOUND_UNDEFINED.
	* valarith.c (value_subscripted_rvalue): Suppress error if
	TYPE_ARRAY_UPPER_BOUND_IS_UNDEFINED.

--- a/gdb/dwarf2read.c
+++ b/gdb/dwarf2read.c
@@ -6074,6 +6074,9 @@ read_subrange_type (struct die_info *die, struct dwarf2_cu *cu)
 
   range_type = create_range_type (NULL, base_type, low, high);
 
+  if (attr && attr->form == DW_FORM_block1)
+    TYPE_HIGH_BOUND_UNDEFINED (range_type) = 1;
+
   name = dwarf2_name (die, cu);
   if (name)
     TYPE_NAME (range_type) = name;
--- a/gdb/valarith.c
+++ b/gdb/valarith.c
@@ -198,7 +198,8 @@ value_subscripted_rvalue (struct value *array, LONGEST index, int lowerbound)
   unsigned int elt_offs = elt_size * longest_to_int (index - lowerbound);
   struct value *v;
 
-  if (index < lowerbound || elt_offs >= TYPE_LENGTH (array_type))
+  if (index < lowerbound || (!TYPE_ARRAY_UPPER_BOUND_IS_UNDEFINED (array_type)
+			     && elt_offs >= TYPE_LENGTH (array_type)))
     error (_("no such vector element"));
 
   v = allocate_value (elt_type);

Re: gdb can't print array element for fortran

[Joel Brobecker]

> 2010-03-16  Jan Kratochvil  <jan.kratochvil@redhat.com>
> 	    Chandru <chandru@in.ibm.com>
> 
> 	* dwarf2read.c (read_subrange_type): Set TYPE_HIGH_BOUND_UNDEFINED.
> 	* valarith.c (value_subscripted_rvalue): Suppress error if
> 	TYPE_ARRAY_UPPER_BOUND_IS_UNDEFINED.

This seems like a reasonable workaround, especially since we already
have some special handling of attr->form == DW_FORM_block1 for range
types.  So I guess we never have block2/block4 or even block attributes
for the upper bound in practice?

Just one request:

>    range_type = create_range_type (NULL, base_type, low, high);
>  
> +  if (attr && attr->form == DW_FORM_block1)
> +    TYPE_HIGH_BOUND_UNDEFINED (range_type) = 1;

Can you add a quick comment, with a FIXME to show that this is temporary?
There's already some useful comments just slightly earlier in the code,
perhaps we just ought to refer to that, somehow.

Pre-approved with the addition of the comment.

-- 
Joel

Re: gdb can't print array element for fortran

[Jan Kratochvil]

On Tue, 16 Mar 2010 16:16:21 +0100, Joel Brobecker wrote:
> This seems like a reasonable workaround, especially since we already
> have some special handling of attr->form == DW_FORM_block1 for range
> types.  So I guess we never have block2/block4 or even block attributes
> for the upper bound in practice?

* I cannot imagine a case something else than DW_FORM_block1 would be used but
  sure GDB should handled all the DW_FORM_block* cases to be correct.

* The maintained offtrunk patch archer-jankratochvil-vla is using
  attr_form_is_block() for the full DW_FORM_block* support.

* As the existing FSF GDB code was already limited to DW_FORM_block1.

* The new patch code fragment application scope should match the scope of the
  existing FSF GDB code.

* This patch will need to be reverted/merged-out on future import of some form
  of the archer-jankratochvil-vla branch therefore I wanted to minimize any
  effort for its import as it should have been spent rather on preparing
  archer-jankratochvil-vla for the merge instead.

=> Chose the minimal patchset satisfying Chandru's requirements.


> >    range_type = create_range_type (NULL, base_type, low, high);
> >  
> > +  if (attr && attr->form == DW_FORM_block1)
> > +    TYPE_HIGH_BOUND_UNDEFINED (range_type) = 1;
> 
> Can you add a quick comment, with a FIXME to show that this is temporary?
> There's already some useful comments just slightly earlier in the code,
> perhaps we just ought to refer to that, somehow.

Done (forgot the FIXME keyword I see now, sorry).


> Pre-approved with the addition of the comment.

Checked-in.



Thanks,
Jan


http://sourceware.org/ml/gdb-cvs/2010-03/msg00158.html

--- src/gdb/ChangeLog	2010/03/16 18:47:14	1.11491
+++ src/gdb/ChangeLog	2010/03/16 20:51:23	1.11492
@@ -1,3 +1,10 @@
+2010-03-16  Jan Kratochvil  <jan.kratochvil@redhat.com>
+	    Chandru <chandru@in.ibm.com>
+
+	* dwarf2read.c (read_subrange_type): Set TYPE_HIGH_BOUND_UNDEFINED.
+	* valarith.c (value_subscripted_rvalue): Suppress error if
+	TYPE_ARRAY_UPPER_BOUND_IS_UNDEFINED.
+
 2010-03-16  Holger Hans Peter Freyther  <zecke@selfish.org>
 
 	* linux-record.c (record_linux_msghdr): Remove unintended semicolons.
--- src/gdb/dwarf2read.c	2010/03/15 20:49:53	1.368
+++ src/gdb/dwarf2read.c	2010/03/16 20:51:23	1.369
@@ -6074,6 +6074,12 @@
 
   range_type = create_range_type (NULL, base_type, low, high);
 
+  /* Mark arrays with dynamic length at least as an array of unspecified
+     length.  GDB could check the boundary but before it gets implemented at
+     least allow accessing the array elements.  */
+  if (attr && attr->form == DW_FORM_block1)
+    TYPE_HIGH_BOUND_UNDEFINED (range_type) = 1;
+
   name = dwarf2_name (die, cu);
   if (name)
     TYPE_NAME (range_type) = name;
--- src/gdb/valarith.c	2010/02/11 21:45:25	1.81
+++ src/gdb/valarith.c	2010/03/16 20:51:23	1.82
@@ -198,7 +198,8 @@
   unsigned int elt_offs = elt_size * longest_to_int (index - lowerbound);
   struct value *v;
 
-  if (index < lowerbound || elt_offs >= TYPE_LENGTH (array_type))
+  if (index < lowerbound || (!TYPE_ARRAY_UPPER_BOUND_IS_UNDEFINED (array_type)
+			     && elt_offs >= TYPE_LENGTH (array_type)))
     error (_("no such vector element"));
 
   v = allocate_value (elt_type);

Re: gdb can't print array element for fortran

[Nick Roberts]

> This mail pertains to the discussion had on gdb mailing list sometime
> back here  http://sourceware.org/ml/gdb-patches/2009-01/msg00251.html.
> After a small discussion with Jan I made the following changes and 
> these didn't seem to work. Any inputs or thoughts would be helpful. 

Hmm, the example given in January works with GDB 6.8.  It looks like a
regression due to this change:


2008-04-22  Markus Deuling  <deuling@de.ibm.com>

	* eval.c (evaluate_subexp_standard): Use value_subscripted_rvalue for
	multi_f77_subscript to support values from registers.
	* valarith.c (value_subscripted_rvalue): Remove prototype and static.
	* value.h (value_subscripted_rvalue): Add prototype.

	* f-typeprint.c (f_type_print_base): Add support for TYPE_CODE_UNION.
	Fix output.
	* f-valprint.c (f_val_print): Likewise.

-- 
Nick                                           http://users.snap.net.nz/~nickrob