[cdt-dev] RE: [cdt-debug-dev] Bring Back "Add Watchpoint (C/C++)" for CD

["Daryl Spitzer" <DSPITZER@xxxxxxxxxx>]

Alain,

> User enter "0x800fa4" in the text field ...
> (gdb) watch 0x800fa4                <----- REALLY BAD
> (gdb) watch *0x800fa4               <----- GOOD
>
> Note: To make this more clear, maybe this action should be name "Add
> Watch Expression"
>      Or we could do some basic sanity check on the expression, if an
> integer value, do the right
>      thing in gdb/mi.

That sounds good to me.  Could we squeeze that sanity check into CDT
2.1.1?

> If the selection is on a global variable ... we are good to go,  if
> it is on a local variable ... we could go from warning the user to
> implementing more complex scenarios ...
>
> in this satisfactory or you still would like to have a liberal
> "add Watch expression" that the user could enter any expression ?

So you're suggesting that we remove the "Add Watch Expression" action
(added in CDT 2.1.1) from CDT 3.0, and introduce watchpoints on local
variables (first with a warning that they essentially only work until
out of scope--maybe with some indication in the Breakpoint view for
watchpoints on locals out of scope--and later re-setting the watchpoint
when we detect that we're within scope)?  I shouldn't speak for our
customers (so I'll ask our product planning team to do that), but this
sounds good to me.  I don't think our customers will miss the
expressions, especially since they'll still be able to get at them
through the GDB console.

Is there someone out there who is able to reintroduce this "Add Watch
Expression" into CDT 2.1.1?  If not I'm willing, though I'll have a bit
of a learning curve to climb.  (So it might take me longer.)

--
Daryl Spitzer
Santa Cruz Technology Center
Altera Corporation
408-544-8539
dspitzer@xxxxxxxxxx


-----Original Message-----
From: cdt-debug-dev-admin@xxxxxxxxxxx
[mailto:cdt-debug-dev-admin@xxxxxxxxxxx] On Behalf Of Alain Magloire
Sent: Monday, January 31, 2005 15:07
To: cdt-debug-dev@xxxxxxxxxxx
Cc: cdt-dev@xxxxxxxxxxx
Subject: Re: [cdt-debug-dev] Bring Back "Add Watchpoint (C/C++)" for CDT
2.1.1?

> 
> The Run->Add Watchpoint (C/C++) menu item (in CDT 1.x) was removed
from
> CDT 2.0.  We (Altera) shipped our Nios II 1.1 IDE (last summer) based
on
> CDT 1.2.1.  Now we're in the middle of upgrading to CDT 2.1 for our
next
> release, and we don't want to take watchpoints (on local variables)
away
> from our users.
> 
> I understand that "Toggle Watchpoint" has been added for global
> variables in the Outline view.  But our customers used the Add
> Watchpoint feature to set watchpoints on local variables.  I also
> understand that watchpoints on local variables are not persistent
> between debugging sessions.  We're aware of the "bug" where
watchpoints
> on local variables remain in the Breakpoints view, but don't work in
> subsequent debugging sessions.  But we feel that it's more important
to
> keep this buggy functionality in than remove it all together.  (I
would
> like to try to fix this bug in the future and, of course, contribute
the
> fix back to the CDT.)
> 
> Would anyone object to this being added back for the CDT 2.1.1
release?
> 
> I exchanged email with Mikhail Khodjaiants about this, and he
suggested
> that we add it back in as an "Add Watchpoint" action in the
Breakpoints
> view, so there won't be two similar editor actions when "Toggle
> Watchpoint" is added as an editor action in CDT 3.0.  This makes sense
> to me.
> 

Cc: cdt-dev, not all folks follow cdt-debug-dev

CDT-2.1.1:
  I would have prefer, the issue to be raise in CDT-2.1 ... but hey! ...
  No objection on my part to put an "Add Watch Expression" action, so
late in the game.
  I guess, we can consider this to be a ... regression fix 8-)

========
A few more points:
(a)
The Add action did not provide any validation and was not well
integrated in term of UI:
- for example, user would enter in the text dialog an address ...
crashing gdb or the embedded target
  because address watch point has a different semantic, then expression
watchpoint.

User enter "0x800fa4" in the text field ...
(gdb) watch 0x800fa4                <----- REALLY BAD
(gdb) watch *0x800fa4               <----- GOOD
 
Note: To make this more clear, maybe this action should be name "Add
Watch Expression"
      Or we could do some basic sanity check on the expression, if an
integer value, do the right
      thing in gdb/mi.

(b)
Local variables watchpoint are not very well understood by the Users.
When the variable,
or when a variable in the expression, becomes out of scope, the
watchpoint is effectively 
deleted (meaning you may not stop if you reenter the function).
And as you pointed out, when the user restarts the debug session, unless
the variables are in scope .. the watchpoints can not be reset ... more
frustration.

Note: We may have some scheme that can make this transparent for the
user by implanting
     some __hidden__ breakpoint and do some work under the cover with
the AST parser.
     But ...

So for the future(CDT-3.0):
When "Toggle Action" action is trigger, we will use the AST parser to
examine
the selection in the CEditor and some analysis on the best course of
action.
If the selection is on a global variable ... we are good to go,  if it
is on
a local variable ... we could go from warning the user to implementing
more complex scenarios ...

in this satisfactory or you still would like to have a liberal
"add Watch expression" that the user could enter any expression ?


_______________________________________________
cdt-debug-dev mailing list
cdt-debug-dev@xxxxxxxxxxx
http://dev.eclipse.org/mailman/listinfo/cdt-debug-dev