Re: PATCH: Re: expansion

[Sven Wischnowsky <wischnow@informatik.hu-berlin.de>]

Oliver Kiddle wrote:

> Sven Wischnowsky wrote:
> > 
> > Yes, I think I said that the patterns could do with a little improvement.
> 
> Yes, and there is still scope for more improvements. Does anyone know
> which characters terminate a parameter reference which doesn't use
> braces?

Anything but [a-zA-Z0-9_], except for the special one-character
parameters and the `:...' modifiers if we want to deal with them.

> > 2) Not using `recexact' as the style name was, of course, intentional,
> >    because it really isn't about `recognising', is it? It's about
> >    accepting the exact match (it will always take an exact match as
> >    one of the possible matches[1]).
> >    And accept-exact has the same meaning as recexact, we don't need to
> >    reverse it.
> 
> Functionally, I can't see that there is a difference between recexact
> and accept-exact. Are you basically saying that acceptexact would have
> been a better name for the option as well?

No, that was taken from elsewhere, I think (csh? tcsh?). But we are
talking about `accepting' a match elsewhere, so...

> > [1] Has anyone else wished he had a style to say `complete anything
> >     *but* an exact match'?
> >     We already have ignore-line... we could extend that.
> 
> It took me a little while to work out what you mean by that but I can
> see that it might be useful because someone might feel that they are
> only going to press tab if they want to complete something.
> 
> It seems to me that the choices are that if the current line is an exact
> but ambiguous match, there are three choices: match only the current
> exact match, match all matches or match all but the current exact match
> so maybe a style with three possible values would be better.

I want to integrate it with (the already existing) ignore-line style,
three choices there: ignore all words from the line, ignore only the
current word, ignore all words except the current one.

> I actually got quite confused while trying to work out what happens in
> different situations because of the line in _expand which prevents
> substitutions expanding if they expand to an empty string and I was
> using an empty variable. Maybe _expand should display a message in this
> case.

Or make it configurable... (this is getting out of hand).

> > I didn't change that even now because I've regretted these
> > `mathematically conditional' styles for quite some time now. They
> > should be turned into boolean styles and the condition should come
> > from somewhere else, e.g. the -e option to zstyle I suggested (see
> > 11691).
> 
> I meant to reply to that at the time because it seems like a good idea.
> It would certainly be a lot cleaner if any sort of evaluations are built
> in to zstyle and don't have to be repeated in completion functions.
> 
> What exactly do you mean by the word evaluated when you say that 'the
> resulting string will be evaluated' in the documentation in the patch:
> is the idea that you would use a function.

The -e option makes the values given when defining the style be
concatenated with spaces between them and the resulting string eval'ed 
(i.e. exactly what the eval builtin does).

> How would this affect the
> functions which can be used for tag-order (which, if you remember I use
> to avoid completing commands with an empty word).

Not at all. Whatever the eval'ed code stores in $value/$reply is
returned as the value of the style when it is looked up. In a certain
sense it's just an extra level of indirection.

> For styles whose
> values are arrays, it would be useful to be able to return $value/$reply
> as an array.

Of course the patch already did that. Since I only used $value, not
$VALUE, you may have guessed that the eval'ed code can return *only*
arrays (well, ok, the contents of $value is only used as an array),
just like styles do. Before the stuff is eval'ed, $value is unset
(explicitly, by the zstyle-code). If it is left unset, this is as if
the style weren't set, if $value is set to some value, it is treated
as if that/those string/s were defined for the style without the -e
option.

> On the subject of the _expand not doing brace expansion, Sven wrote:
> 
> > Change (e)? Use `eval echo ...'? Urgh?
> 
> I think it would be a bad idea to change (e) and the eval is going to be
> slow. I think I said it before a while ago but it would be nice if there
> was clearly defined ways to do the various types of expansion,
> individually or together. A (b) to do brace expansion might solve the
> brace problem but we would maybe have fewer problems in _expand if there
> was syntax like (e:pab:) to do parameter, arithmetic and brace expansion
> with other letters for things like ~, =, history, command and whatever
> other expansions we have.

Well, history expansion at least has to be done before completion is
tried because it's done by the lexer.

The other thing: could get complicated to implement, I think (and
using e:...: might cause trouble with existing uses unless we allow
only a few special character as the separator; is E used?).

> This would have some problems though: expansions which expand to things
> which can be further expanded for example but I'd have thought these
> have all be addressed in the code before. The other trouble with
> expansion as a part of completion is that completion by definition deals
> with things which are incomplete so we have the problem of what to do
> with things which aren't syntactically correct yet. For example, try:
> a(<tab> and you will get '_expand:78: bad pattern: a(' because of the
> lack of a closing bracket.

It doesn't anymore, (e) is silent nowadays. By using eval we could get 
that, though (that's why I said `careful').

Bye
 Sven


--
Sven Wischnowsky                         wischnow@informatik.hu-berlin.de