Re: [PATCH] New MI notification "=tsv-modified"

[Pedro Alves <palves at redhat dot com>]

On 02/02/2013 08:28 AM, Yao Qi wrote:
> On 02/02/2013 04:38 AM, Pedro Alves wrote:
>> Should we emit a notification when GDB detects the target
>> changes the variable too?
>>
> 
> Yes, we should, but I am still evaluating the pros vs. cons. of using
> async remote notification on tsv changes in the target or let GDB to
> check the tsv changes.  I need more time to get the answer, but in
> parallel, I'd like the "=tsv-modified" notification goes in to notify
> MI front-end that a tsv is modified by command.

Ack.  Hmm.  This makes me realize that I think all three =tsv
notifications should be changed in one aspect.

> because "what is the new" is added into the changelog entry in gdb/doc/ChangeLog, don't have to replicate it again in gdb/ChangeLog to mention it again for NEWS.

But those are different ChangeLog files.  Reading gdb/ChangeLog
in isolation should make sense.  More below.

> Here is the new version below, to address your comments.  I don't 
> insist on my original changelog entry, as either is OK to me.

> diff --git a/gdb/doc/gdb.texinfo b/gdb/doc/gdb.texinfo
> index 4b51228..b6b0439 100644
> --- a/gdb/doc/gdb.texinfo
> +++ b/gdb/doc/gdb.texinfo
> @@ -27954,6 +27954,10 @@ value @var{value}.
>  Reports that the trace state variable @var{name} is deleted or all
>  trace state variables are deleted.
>
> +@item =tsv-modified,name=@var{name},value=@var{value}
> +Reports that the trace state variable @var{name} is modified with
> +value @var{value}.

> +@deftypefun void tsv_modified (const char *@var{name}, LONGEST @var{value})
> +The trace state value @var{name} is modified with value
> +@var{value}.
> +@end deftypefun

_Which_ value is being talked about here is not explicit.
Trace state variables have _two_ values.  The initial value,
and the current value (omitted if doesn't exist yet).
If we list tsvs with MI's -trace-list-variables, we'll indeed
see an "init" and a "current" attribute, for each tsv, and no
attribute named "value".  So I think it'd be very good to fix this
before the release, and make the output of the notifications
consistent with the tsv listing output, and the docs clearer.

E.g.:
(gdb) interpreter-exec mi "-trace-list-variables"
^done,trace-variables={nr_rows="1",nr_cols="3",
                       hdr=[{width="15",alignment="-1",col_name="name",colhdr="Name"},
                            {width="11",alignment="-1",col_name="initial",colhdr="Initial"},
                            {width="11",alignment="-1",col_name="current",colhdr="Current"}],
                       body=[variable={name="$a",initial="1"},
                             variable={name="$b",initial="2",current="3"}]

SO IOW, =tsv-created should be

 =tsv-created,name=@var{name},initial=@var{value}

instead of the current

 =tsv-created,name=@var{name},value=@var{value}


and =tsv-modified should be

 =tsv-modified,name=@var{name},initial=@var{value}

instead of the proposed

 =tsv-modified,name=@var{name},value=@var{value}

Maybe it'd be a good idea to factor out the bits in
tvariables_info_1 that dump a tsv into a separate
function to use the in modify case as well?  We'd also
output a "current" attribute as well in the =tsv-modified
case, but I'd argue that actually makes sense?  Then this is
very much like the breakpoint-modified notification, and
interface consistency, at all the levels, is good all around.

-- 
Pedro Alves