Insight crashing when debugging invalid pointers

Insight crashing when debugging invalid pointers

[elmer.elmer]

Hi,

I'm using a cross insight built from recent CVS snapshot (10242005)
[host=cygwin, target=arm-linux]

Insight is running quite fine, but I experience some troubles when insight tries
to display variables with invalid memory references...

For example :

typedef struct {
    int a;
    int b;
} TEST_STRUCT;

TEST_STRUCT pt;
int         sum;

pt = NULL;
sum = pt->a + pt->b;

This code will obviously fail (if memory address 0 is not accessible on target)
!!!
But it is only an example ...

If you stop execution at line 'sum = pt->a + pt->b'
Using the GDB console :
p pt
gives -> $1 (TEST_STRUCT *) 0x0
p pt->a
gives -> Error: Cannot access Memory at address 0x0

But if insight is trying to display the same things (either using tooltips or
watch), the entire Insight is crashing !!!

I thing it is related variable evaluation like {$val value}
I checked the Insight scripts, all calls to {$val value} are secured with
'catch'.
Is "catch {$val value} result" supposed to catch GDB exceptions ???

I wonder if this behavior is specific to my built/target or more generic ...

This is very annoying, because when stepping, your mouse can go over over an
expression where pointers are not yet initialized. Insight then tries to
display variable value in tooltips and this leads to frequent Insight
crashes...

Thanks,
Elmer.

Re: Insight crashing when debugging invalid pointers

[Keith Seitz]

elmer.elmer@free.fr wrote:

> But if insight is trying to display the same things (either using tooltips or
> watch), the entire Insight is crashing !!!
> 
> I thing it is related variable evaluation like {$val value}
> I checked the Insight scripts, all calls to {$val value} are secured with
> 'catch'.
> Is "catch {$val value} result" supposed to catch GDB exceptions ???
> 
> I wonder if this behavior is specific to my built/target or more generic ...

Good guess. It's probably a problem in varobj (which is part of gdb). 
Can you run insight on gdb and get a backtrace of where it dies?

Keith

Re: Insight crashing when debugging invalid pointers

[elmer]

>> But if insight is trying to display the same things (either using 
>> tooltips or
>> watch), the entire Insight is crashing !!!
>>
>> I thing it is related variable evaluation like {$val value}
>> I checked the Insight scripts, all calls to {$val value} are secured 
>> with
>> 'catch'.
>> Is "catch {$val value} result" supposed to catch GDB exceptions ???
>>
>> I wonder if this behavior is specific to my built/target or more 
>> generic ...
>
>
> Good guess. It's probably a problem in varobj (which is part of gdb). 
> Can you run insight on gdb and get a backtrace of where it dies?
>
> Keith

Unfortunately, it is not very easy for me to have a native gdb to debug 
insight ....

But I think I have found the problem (you're right in varobj.c) :
- When gdb tries to evaluate a value, it calls many functions and 
subfunctions. One of these is "value_of_child"
When value_of_child cannot acces memory to evaluate a value, it simply 
return NULL (deletes the value context) instead of reporting the error.
I changed :
      if (!gdb_value_fetch_lazy (value))
         value = NULL;
to:
      if (!gdb_value_fetch_lazy (value))
         parent->error = 1;
All functions calling this one are testing the "->error" flag or NULL 
return value. So it works !

But this patch does not delete the value context as it was done 
previously....
So further use of the value (for types, names,...) are still possible.
Previously, Insight was crashing when trying to get the 'type' of this 
variable.

I made another patch in function "my_value_equal" (it assumes that 
variable can be accessed) :
I changed :
  gdb_assert (!value_lazy (val1));
to:
  if (value_lazy (val1))
    {
      *error2 = 1;
      return 0;
    }


With both these patches, Insight is able to display invalid pointers 
location with an error message set by gdbtk-varobj.c:variable_value
By the way, in case of error reported by varobj_get_value, 
variable_value sets an error messages and returns TCL_ERROR to the caller.

But it seems that "catch {$val value} result" does not catch the error ...
How to catch it ?

Thanks,
Elmer.

Re: Insight won't let me use the source any more

[Bruce Korb]

Hi,

Some strange thing has happened and now I cannot get Insight to
let me do source level debugging any more.  I saved aside a version
of ~/.gdbtkinit, but rewriting it with the saved version and (also)
deleting it completely accomplishes nothing.  Nothing I do brings
back source level debugging.  What's the secret?  :)  Thanks. - Bruce

$ insight --version
GNU gdb 6.4.0.20051202-cvs
<<blah, blah, blah>>
This GDB was configured as "i686-pc-linux-gnu"

Re: Insight won't let me use the source any more

[Bruce Korb]

Bruce Korb wrote:

>
> Hi,
>
> Some strange thing has happened and now I cannot get Insight to
> let me do source level debugging any more.  I saved aside a version
> of ~/.gdbtkinit, but rewriting it with the saved version and (also)
> deleting it completely accomplishes nothing.  Nothing I do brings
> back source level debugging.  What's the secret?  :)  Thanks. - Bruce
>
> $ insight --version
> GNU gdb 6.4.0.20051202-cvs
> <<blah, blah, blah>>
> This GDB was configured as "i686-pc-linux-gnu"
>
/home/bkorb/ag/ag/agen5/autogen.c:2: Error in sourced command file:
Undefined command: "".  Try "help".

    while executing
"gdb_cmd "source $file_name""
    (procedure "source_file" line 4)
    invoked from within
"source_file"
    invoked from within
".srcwin0.#srcwin0#srcwin#container#pane0#childsite#con#menubar#m.#srcwin0#srcwin#container#pane0#childsite#con#menubar#m#file 
invoke active"
    ("uplevel" body line 1)
    invoked from within
"uplevel #0 [list $w invoke active]"
    (procedure "tk::MenuInvoke" line 47)
    invoked from within
"tk::MenuInvoke 
.srcwin0.#srcwin0#srcwin#container#pane0#childsite#con#menubar#m.#srcwin0#srcwin#container#pane0#childsite#con#menubar#m#file 
1
"
    (command bound to event)errorCode is NONE

Re: Insight won't let me use the source any more

[Vale Group]

Bruce Korb wrote:
> ...I cannot get Insight to let me do source level debugging...
> What's the secret?

The first question that pops to mind is, did you switch the compilation to your
program-under-study to MinGW from Cygwin, for instance?

--
Chuck

Re: Insight won't let me use the source any more

[Bruce Korb]

Vale Group wrote:

>Bruce Korb wrote:
>  
>
>>...I cannot get Insight to let me do source level debugging...
>>What's the secret?
>>    
>>
>
>The first question that pops to mind is, did you switch the compilation to your
>program-under-study to MinGW from Cygwin, for instance?
>
No, but I did retype, "make" from the affected directory instead of
the top and it was a "VPATH" build.  I could see the sources again
once I linked the sources into the build directory, so the glitch is
more likely to do with some compile option or another.  So, I do
not know the cause, but my best guess at the moment is some funny
compiler option.  Thanks.  Regards, Bruce