Re: [RFA] stabsread.c (define_symbol): Guard against bad stabstring input.

[Michael Snyder <msnyder@vmware.com>]

[-- Attachment #1: Type: text/plain, Size: 799 bytes --]

Pedro Alves wrote:
> On Saturday 05 March 2011 00:42:59, Pedro Alves wrote:
>> On Saturday 05 March 2011 00:37:25, Michael Snyder wrote:
>>> I think I heard someone say that gdb should be proof against all inputs?
>> Yes, but bad inputs aren't gdb bugs, so those should be errors
>> or warnings, not internal_errors.
> 
> for completeness: ... or "complaints", of which the symbol
> readers are the biggest users, exactly to complain about
> bad input debug info.  complaints have the nice property
> that they warn once, and go quiet the next time the problem
> appears --- when we see a malformed debug info problem, we
> tend to see a lot of instances of the same in the same run,
> so that feature prevents terminal flooding with a bunch of
> duplicated warnings.
> 

Thanks.  How about this?



[-- Attachment #2: null13b.txt --]
[-- Type: text/plain, Size: 719 bytes --]

2011-03-04  Michael Snyder  <msnyder@vmware.com>

	* stabsread.c (define_symbol): Guard against bad stabstring input.

Index: stabsread.c
===================================================================
RCS file: /cvs/src/src/gdb/stabsread.c,v
retrieving revision 1.136
diff -u -p -r1.136 stabsread.c
--- stabsread.c	28 Feb 2011 23:32:27 -0000	1.136
+++ stabsread.c	11 Mar 2011 21:11:44 -0000
@@ -636,6 +636,12 @@ define_symbol (CORE_ADDR valu, char *str
     {
       p += 2;
       p = strchr (p, ':');
+      if (p == NULL)
+	{
+	  complaint (&symfile_complaints, 
+		     _("Bad stabs string '%s'"), string);
+	  return NULL;
+	}
     }
 
   /* If a nameless stab entry, all we need is the type, not the symbol.