[PATCH 2/3] gdb/varobj: Fix use after free in varobj

[Lancelot SIX]

 >>> +    if ((*var_handle)->root->exp == nullptr)
 >>> +      return nullptr;
 >>> +
>>
>> I notice that non of the tests in either this patch, or the next one,
>> exercise this condition.
>>
>> Is it possible to create a test for this case?
> 
> Will do.  Issuing a "-var-update" after the dlclose call in the test 
> should exercise this.  I did that manually, but forgot to include it in 
> the testcaes.
> 

Actually, after double checking, this case is not possible.  We cannot 
reach this point if the the varobj is invalid.  The only way to have
`exp == nullptr` and a valid varobj is if the varobj floating.

In this particular case, code a couple of lines above ensure that either 
the expression is re-parsed (ensuring exp is not nullptr), or exit the 
all function early:

   if (var->root->floating)
     {
       struct varobj *tmp_var;

       tmp_var = varobj_create (NULL, var->name.c_str (), (CORE_ADDR) 0,
			       USE_SELECTED_FRAME);
       if (tmp_var == NULL)
	{
	  return NULL;
	}
       ...
     }

I'll drop this part of the patch in V2.

I will still update the testcase to still exercise the case where:

     var->root->floating && var->root->exp == nullptr

on entry of this function.

Best,
Lancelot.