Re: [PATCH v3 07/34] Make the intepreters output to all UIs

[Pedro Alves <palves at redhat dot com>]

On 05/19/2016 04:15 PM, Simon Marchi wrote:
> On 16-05-06 08:34 AM, Pedro Alves wrote:

>> +  /* Traverse through all UI, and switch the current UI to the one
>> +     being iterated.  */
>> +#define SWITCH_THRU_ALL_UIS(STATE)		\
>> +  for (switch_thru_all_uis_init (&STATE);		\
>> +       switch_thru_all_uis_cond (&STATE);		\
>> +       switch_thru_all_uis_next (&STATE))		\
> 
> The last backslash is not necessary I think.

Thanks, indeed.  I've removed it.

> 
> I was wondering why you did not name this "ALL_UIS", using the same pattern
> as ALL_INFERIORS & al, but then I realized it's because this one actually
> sets current_ui before each iteration (and restores it at the end).

Exactly.

> 
> I guess the reason you need to do this is because, again, everything relies on
> accessing the global current_ui, whereas it should be passed down as a parameter.

It goes deeper than straight access of current_ui directly.  It goes all the
way to gdb_stdout, etc., which are really fields of current_ui.  I don't
ever see us passing down a current_ui to _all_ functions that might want
to do I/O.  That'd mean passing it around _everywhere_.  So basically, if
you're calling a function that may do output, and you want that output
to be duplicated to all UIs, you need to use SWITCH_THRU_ALL_UIS.
Note, we can't just call such functions once, with output redirected to
a buffer, and then print that buffer on all UI's streams, because each
UI's ui_out may want to print differently, and also run through distinct
code paths, due to ui_out_is_mi_like_p.

> I am not saying it should be done in this series (it's a big task in itself), but
> we can probably tackle it after.  They way you did things should make it easy to
> improve things later on.

Thanks,
Pedro Alves