Re: [RFC] mi: add -a option to the "-data-disassemble" command

[Pedro Alves <palves at redhat dot com>]

On 08/10/2018 11:28 AM, Jan Vrany wrote:
> On Fri, 2018-08-10 at 10:47 +0100, Pedro Alves wrote:
>> On 08/10/2018 09:29 AM, Jan Vrany wrote:
>>> @@ -33095,6 +33101,9 @@ records, produced when trying to execute an undefined @sc{gdb/mi} command
>>>  @item exec-run-start-option
>>>  Indicates that the @code{-exec-run} command supports the @option{--start}
>>>  option (@pxref{GDB/MI Program Execution}).
>>> +@item data-disassemble-a-option
>>> +Indicates that the @code{-data-disassemble} command supports the @option{-a}
>>> +option (@pxref{GDB/MI Data Manipulation}).
>>>  @end ftable
>>
>> I'm curious about how you intend to use this new -list-features feature.
>> Are you enabling/disabling some UI element depending on presence of the
>> feature?  I.e., something that you wouldn't be able to do by just
>> trying the "-data-disassemble -a" command and looking for error?
> 
> Yes. Basically I have a menu item on frame that opens another window with
> disassembly of the coresponding function (showing current instruction, 
> (basic) blocks and alike). 
> 
> For programs with debug information, I can do that even without -a option. 
> For programs without debug information, I need -a option and - to make the
> frontend robust w.r.t UX - I disable the menu item. 
> 
> I can indeed just try and catch the error, but this seems to be too heavyweight.
> It would mean to disassemble function whenever user switches frames in the 
> UI or in CLI. I can cache the result, but still. This new feature just makes
> it trivial and fast enough (no MI roundtrip to enable/disable menu item)

Ahah, thanks.  (IMHO, that's useful info that include in a submission that
proposes a -list-features addition.)

> 
> That being said, I do not insist on having this feature. I'm close to the point
> of just saying "you need to use my patched GDB, anyway.  At least for "now" and 
> hope that my patches will eventually make it into official GDB release. Not a big 
> deal for me, not at all. 
> 
> So, if you (meaning you maintainers) prefer not having this feature, I'll just remove 
> it. Absolutely no problem! Just let me know. 

I was really just curious.  And TBC, I was just talking about the -list-features
entry, not the actual whole "-a" feature added by the patch, which I agree is useful.

> --- a/gdb/NEWS
> +++ b/gdb/NEWS
> @@ -3,6 +3,9 @@
>  
>  *** Changes since GDB 8.1
>  
> +* The '-data-disassemble' MI command now accepts the '-a' option to disassemble
> +  the whole function surrounding given program counter value or function name.

This should be moved to the "since 8.2" section.

Thanks,
Pedro Alves