Re: elpa.gnu.org packages requiring external packages

[Eric Abrahamsen]

Stefan Monnier <address@hidden> writes:

>>> AFAICT the pinyin table is only kept in the quail table: quail tables
>>> are somewhat similar to keymaps, so they work well to incrementally map
>>> a sequence of ASCII chars to the possible chinese characters, but for
>>> pyim-hanzi2pinyin we'll need to traverse this table to build a "reverse
>>> table" (most likely represented as a char-table).
>> Okay, I'll take a look.
>
> Thanks.  I played a bit with quail tables in the past, so even though
> I don't understand them 100% I should be able to help if needed.

By dint of heroic grepping I found the simplest entrypoint to the
table:

(nth 2 (quail-package "chinese-py"))

But this value is already heavily processed in the *wrong* direction
(optimized for pinyin-to-character). It's built from the intermediate
file "lisp/leim/quail/PY.el", which is produced during `make' from
"leim/MISC-DIC/pinyin.map", which is the same source as Feng Shu uses
for the pyim package.

So the data is accessible in different formats at different stages:
build, compile, and load. Now I'm thinking that it might make sense to
build the reverse table earlier on -- anyway, I'll explore what's
possible there.

[...]

>>> I think it makes a lot of sense to fold helm-ebdb into the ebdb package,
>>> without adding `helm` as a dependency.  This way, users of Helm and EBDB
>>> won't have to additionally install helm-ebdb to enjoy the combination of
>>> the two: just by installing `ebdb` and `helm` they'll get `helm-ebdb`.
>> Oh I'm all in favor, but I assumed it would be a no-no from a
>> dependency-graph point of view.
>
> Not at all (actually, back when we got EBDB into elpa.git I mentioned
> this possibility).

Wish I'd been paying closer attention!

>> If it doesn't matter, I can fold this back into ebdb and just use
>> `declare-function' with "ext:" to quiet the compiler?
>
> Yes.
>
>> In which case, I might as well just do the same for
>> `helm-marked-candidates' and leave it as-is?
>
> Yes.
>
>> I've also got these other silly little packages -- counsel-ebdb,
>> company-ebdb -- etc, should those all be reabsorbed? There are an awful
>> lot of counsel-* packages in there already.
>
> I don't have an opinion on counsel-ebdb, but for company-ebdb an even
> better option is to turn the company backend into
> a completion-at-point-function (so it works not only for company users
> but also for old-style TAB-completion users).

company primarily does popup-on-a-timer; I think calling it as an
explicit completion command (while supported) isn't how it's expected to
be used.

>> Sorry, I'm just trying to figure out best policy for managing this kind
>> of stuff.
>
> And overall, I really wish someone could sit down, take the ivy,
> company, helm, and completion-at-point-function APIs and design a new
> API which can be used by all of those UIs so you don't have to implement
> N different slight variations of the same thing.

But they do their things in such different ways. Some work on a timer
with a popup (company), others tie into the existing completion
mechanisms (helm and ivy), and others provide their own versions of
basic Emacs commands (helm and counsel). How much unification is
possible?

Eric