[RFA 1/2] Add completion for COMMAND in 'thread apply all|ID... COMMAND'

[Pedro Alves]

On 5/1/19 11:18 AM, Pedro Alves wrote:
> On 5/1/19 5:53 AM, Philippe Waroquiers wrote:
>> On Tue, 2019-04-30 at 09:13 -0600, Tom Tromey wrote:
>>>>>>>> "Philippe" == Philippe Waroquiers <philippe.waroquiers@skynet.be> writes:
>>>>> One example is if the input text is "thread apply all -ascen".
>>>>> It seems like this should complete to "-ascending " (with the trailing space).
>>>
>>> Philippe> Currently, the unpatched GDB HEAD does not complete -ascen.
>>> Philippe> I agree it should complete but it does not :(.
>>> Philippe> (the GDB behaviour regarding command option syntax is wildly inconsistent).
>>>
>>> Yeah, what I meant here is that it would be good if the new completion
>>> code did this, not that this was something that currently occurred.
>> Yes, effectively, the fact that GDB does not complete arguments is surprising
>> (while e.g. bash has completion for arguments of almost all commands, such as:
>>       ls --alm<TAB>
>>   =>  ls --almost-all
>> :).
>>
>> I will take a look at how to have completions for command arguments,
>> but IMO that is better done as a separate patch to provide a more general
>> solution (i.e. not only for commands that are launching other commands,
>> which is what this patch is providing).
> 
> I'm going to post the "print -OPT" patch soon, which includes a generic
> framework for command options, and integrates with completion.  It's the
> patch that I mentioned to you a while ago when we discussed the "qvcs" flags
> patches.  I've pushed it to the users/palves/cli-options branch on
> sourceware last night, if you guys want to take a look meanwhile.  I'd
> like to post it for discussion before we decide how to go forward on the
> "/" patches.  I'm on holiday today, and likely OOO for most of it, so not
> sure I'll be able to post it today.
> 
> Funny enough, "thread apply all -ascen[TAB]" is one of the options
> that I converted to the new scheme in the branch.

A status update -- I've been working on this as much as I'm able.  I fixed
a ton of details on the branch, plus hooked in more options to the "frame apply"
commands, wrote a testcase (need to extend it some more), wrote some
docs.  And also, today I made a replacement for your patch (this thread) that
is quite neat.  Instead of only completing on the command name itself,
my version recurses into the completion machinery, and thus completes
on the command's options correctly.  I.e., you get this:

 (gdb) thread apply all -[TAB]
 -ascending  -c          -q          -s          
 (gdb) thread apply all print -[TAB]
 -address         -array-indexes   -null-stop       -pretty          -static-members  -union           
 -array           -elements        -object          -repeats         -symbol          -vtbl            
 (gdb) thread apply all print -pretty -- [TAB]
 __init_array_start                      global                                  pthread_self@got.plt                    
 __libc_csu_fini                         global_neg                              pthread_self@plt                        
 __libc_csu_init                         int                                     pthread_setname_np                      
 ...

I force-pushed what I have to the branch.  I have not re-run the testsuite yet,
there may be regressions, though the new test exercising the new options framework
passes, at least.

Oh, one change I made was that for "print", if you want to specify - options,
then you must end the options with "--".  There was one testcase that regressed
without this, and so I concluded that this was safer.  I also noticed
that lldb's expr command does the same thing.

Note, you can do "help print" to see the available options.  That bit is
auto-generated.

I'm pretty happy and excited with the result.

My TODO says:

  - comments throughout
  - finish docs
  - NEWS
  - split patch into chunks

  - add smoke tests for "frame apply -TAB", etc.

  - The new options.exp testcase should validate parsed options, instead of being mostly focused 
    exercising completion.
    E.g., make the new maint commands print them out.

  - Hook options to "frame apply level"

  - Fix "print -element 123[TAB]".  Shouldn't complete commands unless
    there's a space after the number.

Thanks,
Pedro Alves