On Tuesday 19 February 2008 00:44:33 Nick Roberts wrote:
>  > > mi_info_threads is nearly the same function as thread_command.  
>  > 
>  > You probably meant info_threads_command?
> 
> Yes.
> 
>  > > Can this be 
>  > > used in a dual way just as -break-list uses breakpoint_1?
>  > 
>  > I don't know -- I actually have no idea what magic makes breakpoint_1 to
>  > work both for MI and CLI. Do you happen to know?
> 
> It just seems to be a kind of overloading where the output format of
> functions like ui_out_field_string depends on the interpreter used (value
> of uiout).  

Okay, I've tried to do that. Now, I'm using ui_out_text to ensure empty
spaces between fields. breakpoint.c uses ui_out_table_begin_end, but
I don't think the formatting that one produces is good here.

> In cases where only one interpreter outputs anything 
> ui_out_is_mi_like_p is used.
> 
>  > > Also Denis Pilat has already proposed a patch for -thread-info:
>  > > http://sourceware.org/ml/gdb-patches/2007-03/msg00167.html
>  > 
>  > I did not notice that.
>  > 
>  > > 
>  > > How does your patch compare?  (assuming the problem of return type is solved
>  > > as has been done for thread_select).
>  > 
>  > It appears that my patch:
>  > 
>  > 1. Does not bother with making non-throwing function, as MI top-level can
>  > handle exceptions.
>  > 
>  > 2. Allows to print information about all threads.
> 
> I see.  Denis also proposed something for -thread-list-all-threads
> (http://sourceware.org/ml/gdb-patches/2007-03/msg00144.html)

I think having one command is a bit better.

> Do you not want to print the frame level?  

Probably not; this command is useful right after stop, when the level is always 0.

> Also do you want to 
> catch errors in print_stack_frame?

Yes, why not? A typical error is when we have a char* parameter pointing to
something not very accessible, in which case I'd prefer print_stack_frame
to catch this error, as opposed to -thread-info outright dying. 


> Rather than: 
> 
> +      print_stack_frame (get_selected_frame (NULL), 0, LOCATION);
> 
> Would it be better to use:
> 
> +      print_frame_info (get_selected_frame (NULL), 1, LOC_AND_ADDRESS, 0);
> 
> ?
> 
> Note that with MI (following the thread with Denis) that the address gets
> printed anyway when print_stack_frame is used:
> 
>   args.print_what = ui_out_is_mi_like_p (uiout) ? LOC_AND_ADDRESS : print_what;

Then, using LOC_AND_ADDRESS in print_stack_frame invocation will only
be more clear, without affecting behaviour? But I guess it's better, indeed.

> 
>  > I don't know what problem of return type you refer to -- can you clarify?
> 
> That, in Denis' patch, gdb_thread_info returns type enum gdb_rc but
> catch_exceptions_with_msg returns type int.  (just like gdb_breakpoint and
> break_command_really did for a while).

This problem does not exist in my patch, as catch_exceptions_with_msg is
just not used.

I attach the revised version of the patch, including docs. Eli, is the
doc part OK? I intend to commit the code part in a week unless there are
objections.

- Volodya

	[gdb]
	Implement -thread-info.
        * gdb.h (mi_info_threads): Declare.

        * thread.c (mi_info_threads): New, extracted
        from info_threads_command and adjusted to
        work for CLI and MI.
        (info_threads_command): Use mi_info_threads.

        * mi/mi-cmds.c (mi_cmds): Specify a handler
        for -thread-info.
        * mi/mi-cmds.h (mi_cmd_thread_info): Declare.
        * mi/mi-main.c (mi_cmd_thread_info): New.
        (mi_cmd_list_features): Include 'thread-info'.

        [gdb/doc]
        * gdb.texinfo (Thread Commands): Document
        -thread-info.  Remove -thread-list-all-threads.