Re: [RFA] completer test [was Re: [RFC] Cleanup for make_source_files_completion_list]

[Pedro Alves <palves@redhat.com>]

On 05/16/2013 12:47 AM, Keith Seitz wrote:
> On 05/15/2013 03:33 PM, Sergio Durigan Junior wrote:
>> Wouldn't it be better to use the "complete" command?  Here is what I see
>> when I use it:
>>
>>      (gdb) complete break filesy
>>      break filesym
>>      break filesym.c
> 
> Is that necessarily "better" than testing what a user would actually type? I don't know. gdb.base/completion.exp uses both forms.

Yeah.  Testing just "complete" doesn't cover everything
necessary.  See thread around:

 http://sourceware.org/ml/gdb-patches/2011-05/msg00032.html

"complete" is useful for frontends, so it's good to test it too.

An idea would be to have a mechanism (procedure, whatever) in (or used by)
completion.exp that given an input string and the expected completions
exercises both "complete" and "\t" with a single call.  E.g.,:

test_completion "break filesy" { "break" "break filesym.c" }

Probably, the completion.exp tests already test the specifics of gdb<->readline
interaction that only the \t code paths trigger, independent of specific
completer, so for testing specific completers it's probably fine to just
go with using the simpler "complete" way only.  If we had the mechanism
suggested above, then we wouldn't have to think about considering whether
all bases are covered though.  We'd just always use it.  :-)

This patch seems to involve testing sort of actual completion mechanics,
so it looks like a specific case where testing (at least) the \t variant
is good.

> This is the first I've heard of send_gdb being deprecated. As far as I can tell, there is no other way to directly test completion this way. I do see, though, that completion.exp uses gdb_test_multiple instead of gdb_expect... If it truly is deprecated, I would expect send_gdb to be made "private" in some way. [deprecated_send_gdb?] Or at least mentioned in lib/gdb.exp.

Yeah.  Testing completion is one of the cases that need to use it.
If we had a higher level mechanism/procedure that simplified \t completion
testing, we could hide those send_gdb calls though.  :-)

> If there is a preference for one or the other [or an actual policy], I will certainly make necessary changes.
> 
> I'm using a similar test strategy for my explicit completion tests, which I am about to submit...

Please use gdb_test_multiple.  See e.g., how completion.exp itself
simplified much when converted to it:

 http://sourceware.org/ml/gdb-patches/2011-05/msg00063.html

-- 
Pedro Alves