Re: [ELPA] New package: transient

["João Távora" <joaotavora@gmail.com>]

On Mon, May 4, 2020 at 12:47 AM Dmitry Gutov <dgutov@yandex.ru> wrote:
>
> On 03.05.2020 22:49, João Távora wrote:
> > We really should work on the completion experience instead of naming.
> > If I can't drive straight, I'm not going to twist the road to compensate.
> I can suggest a different analogy: if the road is twisted and full of
> holes, it doesn't matter how fast or reliable your car is. :-) You'll be
> late to your destination anyway.

... and I'll stretch it some more: if you build more and more road you'll
just be left with cross-roads and no clear path.  So ditch the car,
get a mountain bike.

> I'm sure there's something to improve there, and more capability is
> always good, but when the user has to compensate for the API difficulty
> with specially-tweaked search terms, that's not too great either.

No special tweaking.  I mean users type "string" and they don't
miss any important string-only function and doesn't get
zibity-bob-stringy-string in the first results.  Emacs -Q +  fido-mode
I type C-f string and the only less relevant result i see is lgstring-char,
a super specific function from composite.el used multilangual
support.  _That_ should be renamed composite-lgstring-char.

Should the "string" results I get be grouped by arity, destructiveness
or some other sub-criteria? Maybe. flex knows nothing about that.
But it could if it really is useful.

I can even see a completion system where you type "alist" and "assq"
appears in the list by considering some source of truth (the manual).

>  > Reasonably simple
>  > ML techniques come to mind for relevancy scoring, for example.
> Usage-based sorting? I don't like that in general, and it doesn't fit
> all situations.

Don't see why. But let's for sake of argument say that I agree. What is
the problem we're trying to solve here?  Is it that there are functions
operators missing or just inconsistently named so newbies can't
find them? Eli has already shown they are grouped in the manual.
But some say newbies don't read manuals. Fine. You contend that
newbies use completion. Fine, then mix in info from the manual
into the sorting/grouping of completion results. Newbies prefer
API lists by topic? Then let's make those pretty lists from the
manual, again.

Really, let's first try that and have a good look at the results before
we decide renaming is the way to go, because that has serious
costs in mental overhead.

-- 
João Távora