Re: [RFC] Fixing gdb.base/completion.exp (PR testsuite/12649)

[Pedro Alves <pedro at codesourcery dot com>]

On Wednesday 27 April 2011 16:05:29, Joel Brobecker wrote:
> > Thus, my point is that we could replace those "send_gdb + sleep +
> > gdb_expect" sequences with just one gdb_test{,multiple,no_output}.  I
> > don't know yet if this transformation is possible for every test in
> > the completion.exp file.  Maybe the changes would be quite dramatical.
> > However, this test would be _much_ simpler and much faster.  Also, the
> > current formatting is ugly ;).
> > 
> > So, do you think this is a good idea?  Is there something I'm missing?
> 
> I don't know the history of the testcase, and this is only my own
> opinion, but I tend to agree with you.   I think we should keep one
> test with \t, to make sure that a tab does trigger the completion,
> but the rest of the testcase should be using the "complete" command.
> That's what we do at AdaCore anyways...

How to fix the race that Marek is seeing in that leftover \t instance?

Marek wrote:

> The '\t's do not work well with char-wise read1() and thus
> they're occasionally causing problems.

What are these problems exactly?

I also wonder what's the rationale for the sleeps in the
current implementation?

> # tests for command completion
> #
> # Here are some useful test cases for completion.  
> # They should be tested with both M-? and TAB.

An idea would be for the test to exercise all supported completion
methods (using a convenience procedure, not duplicating
the tests!).

-- 
Pedro Alves