GDB/MI - var-update/create BUG

GDB/MI - var-update/create BUG

[Eran Ifrah]

Hi All,

I found an annoying bug which I was not able solve (however, I found a
reasonable workaround)
This email is a bit long, so please execute me.

In brief:
If I have a local variable with the same name in different scopes,
when switching between the scopes the content of the variable object
does not reflect the new variable.

Here is a code snippet that demonstrates what I mean:

#include <string>
void foo() {
	std::string mystr = " second value ";
	mystr += " appended content";
}

int main(int argc, char **argv) {
	std::string mystr = " first value ";
	foo();
	mystr += " appended content";
	return 0;
}

In the above code snippet, the variable 'mystr' exists in both 'main'
and 'foo' functions.
If I create a variable object while inside 'main' and then I am
stepping into 'foo' the content of the variable object is not updated
to the new value.
(Obviously, I called -var-update var1 and -var-update * without a success)

It is as if 'var1' is pointing to the 'mystr' from the 'main' scope
and refuses to update. The only workaround is  to delete the variable
object and then recreate it.
However, to implement this, I need my code to keep track on the
current frame and depth level - which is a tedious task (not to
mention that the scope is only provided in the form of method name so
overloaded methods can break this logic as well)

Here is the debug session for the above code:

( starting from first line of main() )
(gdb)
-var-create - * mystr._M_dataplus._M_p
^done,name="var1",numchild="1",value="0x28ff28 \"h\\377(\"",type="char
*",thread-id="1",has_more="0"
(gdb)
-exec-next
(gdb)
*stopped,reason="end-stepping-range",frame={addr="0x004014a3",func="main",args=[{name="argc",value="1"},{name="argv",value="0xb21868"}],file="C:/Users/eran/src/TestArea/GDB_MI/main.cpp",fullname="C:/Users/eran/src/TestArea/GDB_MI/main.cpp",line="13"},thread-id="1",stopped-threads="all"
(gdb)
-var-update *
^done,changelist=[{name="var1",in_scope="true",type_changed="false",has_more="0"
}]
(gdb)
-var-evaluate-expression "var1"
^done,value="0xb218fc \" first value \""
(gdb)


// SO FAR all is OK, var1 contains the correct value, " first value "
// Next, I will enter the method 'foo' which also has a variable named
// 'mystr'

-exec-step
(gdb)
*stopped,frame={addr="0x0040134d",func="foo",args=[],file="C:/Users/eran/src/TestArea/GDB_MI/main.cpp",fullname="C:/Users/eran/src/TestArea/GDB_MI/main.cpp",line="5"},thread-id="1",stopped-threads="all"
(gdb)
-var-update *
^done,changelist=[]
(gdb)
-exec-next
^running
*running,thread-id="all"
(gdb)
*stopped,frame={addr="0x00401384",func="foo",args=[],file="C:/Users/eran/src/TestArea/GDB_MI/main.cpp",fullname="C:/Users/eran/src/TestArea/GDB_MI/main.cpp",line="6"},thread-id="1",stopped-threads="all"
(gdb)
-var-update *
^done,changelist=[]
(gdb)
-var-evaluate-expression "var1"
^done,value="0xb218fc \" first value \""
(gdb)

// At this point, we are still inside 'foo' method, however the
content of mystr is still " first value "

Any ideas what am I doing wrong?

-- 
Eran Ifrah
Cross platform, open source C++ IDE: http://www.codelite.org

RE: GDB/MI - var-update/create BUG

[Andrew Burgess]

> On 26 October 2010 23:02, Behalf Of Eran Ifrah wrote:
> 
> Here is the debug session for the above code:
> 
> ( starting from first line of main() )
> (gdb)
> -var-create - * mystr._M_dataplus._M_p
> ^done,name="var1",numchild="1",value="0x28ff28 \"h\\377(\"",type="char
> *",thread-id="1",has_more="0"

Using "*" ties the creation of the variable to the current frame. Changing to use "@" will create a floating variable that will use the current _active_ frame. This seems to produce the result you are looking for.

Thanks,
Andrew

Re: GDB/MI - var-update/create BUG

[Eran Ifrah]

On Wed, Oct 27, 2010 at 9:50 AM, Andrew Burgess <aburgess@broadcom.com> wrote:
>> On 26 October 2010 23:02, Behalf Of Eran Ifrah wrote:
>>
>> Here is the debug session for the above code:
>>
>> ( starting from first line of main() )
>> (gdb)
>> -var-create - * mystr._M_dataplus._M_p
>> ^done,name="var1",numchild="1",value="0x28ff28 \"h\\377(\"",type="char
>> *",thread-id="1",has_more="0"
>
> Using "*" ties the creation of the variable to the current frame. Changing to use "@" will create a floating variable that will use the current _active_ frame. This seems to produce the result you are looking for.
>

Thank you! this was exactly what I was overlooked.

I was using the wrong docs link:
http://www.delorie.com/gnu/docs/gdb/gdb_235.html

While I should have looked at:
http://sourceware.org/gdb/onlinedocs/gdb/GDB_002fMI-Variable-Objects.html#GDB_002fMI-Variable-Objects

> Thanks,
> Andrew
>
>



-- 
Eran Ifrah
Cross platform, open source C++ IDE: http://www.codelite.org

Re: GDB/MI - var-update/create BUG

[Eran Ifrah]

On Wed, Oct 27, 2010 at 7:03 PM, Jan Kratochvil
<jan.kratochvil@redhat.com> wrote:
> On Tue, 26 Oct 2010 18:01:30 -0400, Eran Ifrah wrote:
>> If I have a local variable with the same name in different scopes,
>> when switching between the scopes the content of the variable object
>> does not reflect the new variable.
>
> This is correct from the GDB's point of view,

As you can see from the my signature I am coming from the IDE point-of-view.

> normally you want to track that
> _variable_, not that _name_. Anywhere you are.
>
Not really, thats depends on what you are trying to achieve - since
variable object were designed to ease the task of the IDE, then I say
this is incorrect (at least to me).
In my case, I actually never want to track 'variable object' I want to
track 'expressions' in the code.

In codelite IDE, there are 3 main views which use the 'Variable Objects':
- The Locals view
- Watches (user adds expressions)
- The debugger tooltip (codelite uses a tree form tree which suits
perfectly with the variable object concept)

As for the 'Locals' view, I would expect the variable-object to
present the values based on the expression used during the creation of
the variable object, otherwise it does not make any sense of putting
it in the Locals view in the first place.

For the 'watches' view (which are more of a 'persistent' entries) it
make sense to use the scope of creation - but this is also debatable.
I am currently using floating variable object in the watches view and
it seems "natural" to me

The tooltip - well its nature is a transient window, so it does not
really matter - since the variable object is destroyed as soon as the
tool tip is dismissed.

> You should use '@' instead of '*' for frame if you want to track it by name,
> see "floating" in: info '(gdb)GDB/MI Variable Objects'
>

Already did and it is working as expected

>
> Thanks,
> Jan
>
>
> g++ -o 2 2.C -Wall -g
> gdb -nx -i=mi <2.cmd ./2
>
> ==> 2.C <==
> #include <string>
> void foo() {
>        std::string mystr = " second value ";
>        mystr += " appended content";
> }
> int main(int argc, char **argv) {
>        std::string mystr = " first value ";
>        foo();
>        mystr += " appended content";
>        return 0;
> }
>
> ==> 2.cmd <==
> -break-insert -t 8
> -exec-run
> -var-create - * mystr._M_dataplus._M_p
> -break-insert -t 4
> -exec-continue
> -var-update *
> -var-evaluate-expression "var1"
>
> ->
> ^done,value="0x601028 \" first value \""
>
> ==> 2-floating.cmd <==
> -break-insert -t 8
> -exec-run
> -var-create - @ mystr._M_dataplus._M_p
> -break-insert -t 4
> -exec-continue
> -var-update *
> -var-evaluate-expression "var1"
>
> ->
> ^done,value="0x601058 \" second value \""
>



-- 
Eran Ifrah
Cross platform, open source C++ IDE: http://www.codelite.org

Re: GDB/MI - var-update/create BUG

[Jan Kratochvil]

On Tue, 26 Oct 2010 18:01:30 -0400, Eran Ifrah wrote:
> If I have a local variable with the same name in different scopes,
> when switching between the scopes the content of the variable object
> does not reflect the new variable.

This is correct from the GDB's point of view, normally you want to track that
_variable_, not that _name_. Anywhere you are.

You should use '@' instead of '*' for frame if you want to track it by name,
see "floating" in: info '(gdb)GDB/MI Variable Objects'


Thanks,
Jan


g++ -o 2 2.C -Wall -g
gdb -nx -i=mi <2.cmd ./2

==> 2.C <==
#include <string>
void foo() {
        std::string mystr = " second value ";
        mystr += " appended content";
}
int main(int argc, char **argv) {
       	std::string mystr = " first value ";
       	foo();
        mystr += " appended content";
        return 0;
}

==> 2.cmd <==
-break-insert -t 8
-exec-run
-var-create - * mystr._M_dataplus._M_p
-break-insert -t 4
-exec-continue
-var-update *
-var-evaluate-expression "var1"

->
^done,value="0x601028 \" first value \""

==> 2-floating.cmd <==
-break-insert -t 8
-exec-run
-var-create - @ mystr._M_dataplus._M_p
-break-insert -t 4
-exec-continue
-var-update *
-var-evaluate-expression "var1"

->
^done,value="0x601058 \" second value \""

Re: GDB/MI - var-update/create BUG

[Robert Dewar]

On 10/27/2010 1:17 PM, Eran Ifrah wrote:

>> normally you want to track that _variable_, not that _name_.
>> Anywhere you are.

I agree with that viewpoint
>>
> Not really, thats depends on what you are trying to achieve - since
> variable object were designed to ease the task of the IDE, then I
> say this is incorrect (at least to me). In my case, I actually never
> want to track 'variable object' I want to track 'expressions' in the
> code.

OK, but this seems quite bizarre to me, and I would not want GDB
to do this by default