[RFC] Use of create_new_frame from 'frame'.

[RFC] Use of create_new_frame from 'frame'.

[Andrew Burgess]

The 'frame' command can cause 'create_new_frame' in some conditions:

  - When two arguments are passed, or
  - When one argument is passed that is not:
    - A valid index into the backtrace, or
    - A valid stack frame base address.

Personally, I dislike this behaviour because:

  1. It's easy to make a mistake using the frame command and create a
     new frame.  It's not obvious (I think) how to recover from this
     situation.  [ The answer is just to use 'frame' again with a
     valid frame specifier argument. ]

  2. As a backtrace can be possibly any depth then depending on where
     we stop alters which frames we can create.  This seems a little
     arbitrary.

We could improve #1 by not allowing small indexes that are close to
the number of frames in the backtrace, however, this feels like a bit
of a hack, and would be a little annoying on small targets where low
value addresses are actually valid.

I wonder then if there's a better way to offer this feature.  The two
possible choices (other suggestions welcome) are:

  1. Create a new command 'create-frame' that takes one or two
     arguments, and always creates and selects a new frame new frame.
     The frame command no longer creates new frames.

  2. Make 'frame' take an (optional) flag (like 'x', or
     'disassemble'), so a user can now say:
        frame /c STACK-ADDR PC-ADDR
     in order to create (and select) a new frame.  Without the /c flag
     no new frames are created.  With the flag only new frames are
     created, old frames are not selected.

What do people think?  Any alternative suggestions?

Thanks,
Andrew

Re: [RFC] Use of create_new_frame from 'frame'.

[Pedro Alves]

On 07/08/2015 06:09 PM, Andrew Burgess wrote:
> The 'frame' command can cause 'create_new_frame' in some conditions:
> 
>   - When two arguments are passed, or
>   - When one argument is passed that is not:
>     - A valid index into the backtrace, or
>     - A valid stack frame base address.
> 
> Personally, I dislike this behaviour because:
> 
>   1. It's easy to make a mistake using the frame command and create a
>      new frame.  It's not obvious (I think) how to recover from this
>      situation.  [ The answer is just to use 'frame' again with a
>      valid frame specifier argument. ]
> 
>   2. As a backtrace can be possibly any depth then depending on where
>      we stop alters which frames we can create.  This seems a little
>      arbitrary.
> 

Totally agreed.  See here for my previous suggestions:

  https://sourceware.org/ml/gdb/2014-11/msg00028.html

Related bugs:

  https://sourceware.org/bugzilla/show_bug.cgi?id=18074
  https://sourceware.org/bugzilla/show_bug.cgi?id=8864

> We could improve #1 by not allowing small indexes that are close to
> the number of frames in the backtrace, however, this feels like a bit
> of a hack, and would be a little annoying on small targets where low
> value addresses are actually valid.

Yeah, yuck.  :-)

Thanks,
Pedro Alves

Re: [RFC] Use of create_new_frame from 'frame'.

[Eli Zaretskii]

> Date: Wed, 8 Jul 2015 18:09:44 +0100
> From: Andrew Burgess <andrew.burgess@embecosm.com>
> 
> I wonder then if there's a better way to offer this feature.  The two
> possible choices (other suggestions welcome) are:
> 
>   1. Create a new command 'create-frame' that takes one or two
>      arguments, and always creates and selects a new frame new frame.
>      The frame command no longer creates new frames.
> 
>   2. Make 'frame' take an (optional) flag (like 'x', or
>      'disassemble'), so a user can now say:
>         frame /c STACK-ADDR PC-ADDR
>      in order to create (and select) a new frame.  Without the /c flag
>      no new frames are created.  With the flag only new frames are
>      created, old frames are not selected.

You could also ask the user for confirmation, when the command is
going to create a new frame.

Re: [RFC] Use of create_new_frame from 'frame'.

[Pedro Alves]

On 07/08/2015 08:42 PM, Eli Zaretskii wrote:
>> Date: Wed, 8 Jul 2015 18:09:44 +0100
>> From: Andrew Burgess <andrew.burgess@embecosm.com>
>>
>> I wonder then if there's a better way to offer this feature.  The two
>> possible choices (other suggestions welcome) are:
>>
>>   1. Create a new command 'create-frame' that takes one or two
>>      arguments, and always creates and selects a new frame new frame.
>>      The frame command no longer creates new frames.
>>
>>   2. Make 'frame' take an (optional) flag (like 'x', or
>>      'disassemble'), so a user can now say:
>>         frame /c STACK-ADDR PC-ADDR
>>      in order to create (and select) a new frame.  Without the /c flag
>>      no new frames are created.  With the flag only new frames are
>>      created, old frames are not selected.
> 
> You could also ask the user for confirmation, when the command is
> going to create a new frame.

I'd rather the explicit approach, because querying works when
in interactive mode, but not with scripts.

(and because I see no benefit to the current overloading
of number/address, only downsides.)

Thanks,
Pedro Alves

Re: [RFC] Use of create_new_frame from 'frame'.

[Eli Zaretskii]

> Date: Wed, 08 Jul 2015 20:46:17 +0100
> From: Pedro Alves <palves@redhat.com>
> CC: gdb-patches@sourceware.org
> 
> > You could also ask the user for confirmation, when the command is
> > going to create a new frame.
> 
> I'd rather the explicit approach, because querying works when
> in interactive mode, but not with scripts.

We've got the same situation with umpteen other commands, and we also
have a solution for scripts: we implicitly assume a NO answer.

But I'm not going to insist.  I just wanted to remind us that there's
one more alternative.

Re: [RFC] Use of create_new_frame from 'frame'.

[Andrew Burgess]

* Eli Zaretskii <eliz@gnu.org> [2015-07-08 22:42:08 +0300]:

> > Date: Wed, 8 Jul 2015 18:09:44 +0100
> > From: Andrew Burgess <andrew.burgess@embecosm.com>
> > 
> > I wonder then if there's a better way to offer this feature.  The two
> > possible choices (other suggestions welcome) are:
> > 
> >   1. Create a new command 'create-frame' that takes one or two
> >      arguments, and always creates and selects a new frame new frame.
> >      The frame command no longer creates new frames.
> > 
> >   2. Make 'frame' take an (optional) flag (like 'x', or
> >      'disassemble'), so a user can now say:
> >         frame /c STACK-ADDR PC-ADDR
> >      in order to create (and select) a new frame.  Without the /c flag
> >      no new frames are created.  With the flag only new frames are
> >      created, old frames are not selected.
> 
> You could also ask the user for confirmation, when the command is
> going to create a new frame.

That's certainly a possible option.  I didn't list it as it does not
in itself solve the problem of different length backtraces preventing
the creation of frames; a backtrace of length 100 means the user can't
create a frame at address 100.

Thanks,
Andrew