This is the mail archive of the
gdb-patches@sourceware.org
mailing list for the GDB project.
Re: [PATCH]: MI -var-set-format
On Sat, May 06, 2006 at 11:29:57AM +1200, Nick Roberts wrote:
> > > I would like to add the value in the (new) current format, which I find
> > > much more useful:
> > >
> > > -var-set-format var1 hexadecimal
> > > ^done,format="hexadecimal",value="0x12"
> > >
> > > OK to apply if I update the testsuite accordingly?
> >
> > This seems reasonable; I doubt anyone will object.
> >
> > I was wondering how to handle potential error conditions. I see at
> > least two:
> >
> > - The return of varobj_get_value can be NULL. You should check for
> > that.
>
> That often happens already. I don't see a real problem: the null string
> gets returned and nothing gets printed for the value.
So ui_out_field_string will handle a NULL string OK. I didn't realize
that. I see that it seems to be true, though I couldn't see even any
comments to that effect - but both the CLI and MI output routines do
check, so you're right.
> > - common_val_print might fail to transform the struct value * into
> > a string for some reason. I believe it may call error() if that
> > happens.
>
> I don't know what the reason would be. However, the value is printed via
> varobj_get_value for many other MI commands e.g -var-evaluate-expression,
> -var-list-children --all-values, -var-update --all-values -- and I've
> not seen it call an error yet
There were plenty of reasons - we were talking about this in another
thread at the same time :-) For instance, it might try to read from
inaccessible memory to dereference a bad pointer. DEREF_REF was set in
this call.
But, I've just changed that. Now it won't error any more after the
patch I just committed. So, I suppose your patch is now OK.
It's a little disconcerting that in some cases it will omit value and
in others print a <message>, but hey, things will work out OK.
Could you resubmit the patch including any necessary testsuite changes,
please?
--
Daniel Jacobowitz
CodeSourcery