Re: interpreter_p && strcmp (interpreter_p, "mi") == 0

[Andrew Cagney <ac131313 at cygnus dot com>]

> Hello,
>> 
>> In going over the MI sources I've come across this little gem (it is
>> scattered through out both breakpoint.c and infrun.c).  I'm now trying
>> to remember what discussion went on about it back when it was introduced
>> and how to now go back and fix it (I know there was some but it would
>> have been Cygnus internal :-( ).
>> 
>> Briefly, there are a number of functions for which the behavour changes
>> when the MI output is enabled:
>> 
>> o       extra information not part of
>> the normal CLI output is added
>> 
>> o       the output is made fixed
>> 
>> The reason I noticed this is because I was playing around with having
>> ``-i=mi1'' for the old syntax and ``-i=mi'' for the final syntax.
>> 
>> Andrew
> 
> 
> The MI is not capable of generating output that was not also printed. 
> This must have been added by someone trying to circumvent that.


Yep, you, me and Elena :-)


> As far as I can remember, the way I recommended that to be fixed was by
> splitting the code that used to generate the console output into two
> functions -- the actual task (libgdb function) and a CLI printing part.
> Then, making the MI call the former would allow it to generate the
> output in whatever way it wants.


I don't remember, the libgdb doco paints the same picture.  I do 
remember that pratically from day one, refusing to include a GDB CLI on 
the (now dead) churchill IDE, because we knew we couldn't do it right.


> This, of course, is more work and may be tricky in some cases, but it is
> something that has to be done anyway so we get a UI-free libgdb.


Dig, dig, dig.  Elena sumarized the problem in this message:


> Those tests address two needs:
> 
> 1. We don't have a function that does the printing only in case of MI.
> We have a function (ui_out_text) that does printing only if CLI and
> not MI mode. We don't have the opposite. If we did, we wouldn't need
> those tests. We can write such a function and embed the test into it.
> Those prints I am referring to should be done (I believe) for any
> interpreter, the check that interpreter_p == MI is kind of redundant.


The first one I'm looking at is breakpoints. For that I think I'll add 
an extra parameter to table_begin() to indicate if the number of rows is 
nonzero. That way the cli and MI can each decide what to do when there 
are no rows.


> 2. The other role those conditionals have is to sligtly alter the
> behavior of gdb. For instance when gdb prints the location and source
> line of where it stopped at. Gdb in cli mode changes the way of
> printing that info depending on the command it was executing, while in
> mi mode, it always prints the info in the same consistent way (I
> needed to overwrite a flag for mi/interpreter mode only). In these
> cases, again I think we could maybe avoid checking whether it is mi.


Now that this code is public, such manipulation becomes possible.

	Andrew