Re: recursive bug in dwarf2read.c

[Gaius Mulley <gaius@glam.ac.uk>]

Daniel Jacobowitz <drow@false.org> writes:

> On Wed, Dec 09, 2009 at 10:55:30AM +0000, Gaius Mulley wrote:
>> 
>> Hi,
>> 
>> I think there is a bug in dwarf2read.c which can be exposed by
>> attempting to set a break point at RealIO_WriteFixed in this
>> executable.
>> 
>> I acknowledge the (gnu modula-2 generated executable) could probably
>> be at fault, nevertheless it does contain cyclic data structures which
>> are not explicitly created in the same way as C.  For example:
>> 
>> 
>> TYPE
>>   ScanState =  PROCEDURE (CHAR, VAR ScanClass, VAR ScanState);
>
> See e.g. this from read_structure_type:
>
>   /* We need to add the type field to the die immediately so we don't
>      infinitely recurse when dealing with pointers to the structure
>      type within the structure itself. */
>   set_die_type (die, type, cu);
>
> It looks as if read_subroutine_type does not do that, but now has to.

Hi Daniel,

here is a patch which fixes the bug above.  Embarrassingly after
checking my private gdb directories I found I fixed this last year
and sat on and forgot the patch.  Guess there is a moral here
somewhere..


2009-12-10  Gaius Mulley  <gaius@glam.ac.uk>

	* dwarf2read.c (read_subroutine_type): Add the subroutine
	  type to the die immediately to allow a parameter type to be
	  the same subroutine type.


--- src-orig/gdb/dwarf2read.c	2009-12-10 16:41:40.000000000 +0000
+++ src/gdb/dwarf2read.c	2009-12-10 17:04:16.000000000 +0000
@@ -5879,6 +5879,11 @@
      the default value DW_CC_normal.  */
   attr = dwarf2_attr (die, DW_AT_calling_convention, cu);
   TYPE_CALLING_CONVENTION (ftype) = attr ? DW_UNSND (attr) : DW_CC_normal;
+
+  /* We need to add the subroutine type to the die immediately so
+     we don't infinitely recurse when dealing with parameters
+     declared as the same subroutine type. */
+  set_die_type (die, ftype, cu);
   
   if (die->child != NULL)
     {


I've ran the testsuite on a AMD64 debian stable (lenny) system and and
it doesn't appear to cause any regression.  I've also some minor
Modula-2 language support improvement patches which I wrote in Sept
which I'll now push out to the mailing list..

regards,
Gaius