RE: set record query <on|off>

[Marc Khouzam <marc dot khouzam at ericsson dot com>]

> -----Original Message-----
> From: Tom Tromey [mailto:tromey@redhat.com] 
> Sent: September-11-09 6:07 PM
> To: Marc Khouzam
> Cc: 'gdb-patches@sourceware.org'
> Subject: Re: set record query <on|off>
> 
> >>>>> "Marc" == Marc Khouzam <marc.khouzam@ericsson.com> writes:
> 
> Marc> here is a patch to allow to enable/disable the queries
> Marc> used in Record.  It is relevant to:
> Marc> http://sourceware.org/ml/gdb/2009-09/msg00165.html
> Marc> What do you think?
> 
> It seems like a strange approach to me.

I was hoping for something more generic for queries, but in the few
email exchanges, things kept coming back to such a specific new 
Record command.

> I don't think I know the whole background here.  Is this to help users
> who enter commands at a console?  Or MI?  Or CLI commands sent by
> Eclipse, but not from a user console (that is, in response to some GUI
> control)?

Sorry, I should have summarized the issue in the mail.  Here goes.

When Eclipse is interacting with GDB, there is a single "channel of
communication" which is used for everything (MI, CLI when no MI equivalent,
and Eclipse console).  This channel triggers the GDB code paths
that are "!input_from_terminal_p ()" and this makes every query
be automatically answered with the default answer.

This was not a problem before because there was only a single
use of nquery() and it was never triggered by Eclipse
(it was for pending breakpoints).
With PRecord though, there are new uses of nquery(), and these, from Eclipse,
keep being answer 'N' automatically.

Only one case is actually a problem right now:
it is when changing memory (or registers) while replaying execution; in this
case there is an nquery(), and since it is automatically answered, I just
cannot set memory during an replay.  Being able to change memory during
replay is one of the cool features of PRecord, and that is why I'm trying
really hard to get this to work for 7.0.
BTW, to attempt to set memory, we use -var-assign so I cannot use the 
"server" prefix.

> For MI, it seems like there should be arguments to the 
> internal function
> and then the MI commands should disable querying that way.  
> That is, you
> don't need a new user-visible setting for this.

Disabling queries based on MI is something that was brought up.
The idea was to have queries check if the interpreter was MI and in
this case, answer 'Y' to any query.
I like this because it is more future-proof.
Is that what you are suggesting too?
 
> For the CLI-commands-sent-by-Eclipse case, it seems like you could use
> the "server" prefix.

This may be useful for other cases, but in the current case (-var-assign) 
it does not work.
 
> For the console case... it seems to me that queries are just 
> a necessary part of that.

It would be nicer to have queries in this case but not only does it
not work because we are in the case !input_from_terminal_p (), but having
a query would block the "communication channel" waiting for user input, 
which would somewhat lock-up the Eclipse GUI.

So, to summarize the summary, I need a way to get GDB to answer 'Y' for every
query, even nquery().  And right now, the only time this is needed is for PRecord.

Thanks for any more guidance.

Marc