Re: Moving completion functions

[Oliver Kiddle <okiddle@xxxxxxxxxxx>]

Sven wrote:

> If we remove the second level on installation, this should be a
> first-level directory like X, though.  And there's even stuff to fill
> a Network/Type directory.

It would depend on whether we remove the second-level on installation
or remove any directories which don't contain subdirectories on
installation. The latter would allow you to use Unix/Network if you
wanted. I don't really mind about this or about whether we do any
further subdivision - 84 functions in Unix/Command is not so much.

> Yes, similar foe Network if it's a first level directory.  There are
> network commands that need X and ones that don't.

I agree with Bart that anything needing X should be under X first.
Bart's point about the dependencies is also very much valid - an X
wrapper for a command-line network program is very likely to call
things in Network/Types.

Bart wrote:
> Let me jump back a bit and point out that it's not really correct to say
> that _use_lo is a "utility" -- _arguments is the utility, and _use_lo
> calls it; nobody would call _use_lo from some other function.

I'd agree with this. It was easy to miss that with _use_lo not being
the completion for any commands anymore. _gnu_generic is a good name.

> Aliasing is not the only reason for making compinit a function.  It also
> uses quite a few local parameters, "emulate -L zsh", and extendedglob.
> Those are all messy to deal with in a sourced file.

I remembered the aliasing as the main reason because there are ways
around the other problems such as the way you describe of using a
function defined within the sourced script. I'd prefer that way of doing
things. More user's would understand what they are doing when told to
source a file than the autoloading. And, it would avoid the zsh always
setting fpath stuff that I mentioned my dislike for before.

Oliver