Use elisp--xref-backend in *Apropos* and *Help* (Was Re: [ELPA] New package: transient)

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

Since we're on the subject of Elisp development help, can we restart the
discussion on how to get the `elisp--xref-backend` activated in *Apropos*
and help?  The default `etags--xref-backend` isn't very useful there.  Plus,
when one is used to M-. and M-, it's a little strange to have to switch to C-h o
to follow the very commonly occurring references to symbols.

I recall correctly, we had already settles that it's a decently good idea,
but we ran into some loading precedence problem and the effort stalled.
Is that the right recollection?  It doesn't seem to make a lot of sense,
since all we apparently need is `add-hook`.

Anyway, it can be argued that `elisp--xref-backend` isn't the _best_ choice
for those buffers, and that we should be taking the user to say, the manual
when searching from those places.  Or maybe, the `elisp--xref-backend` could
also collect matches from the manual, why not?

Thanks,
João

On Sat, May 2, 2020 at 11:57 AM Philippe Vaucher
<philippe.vaucher@gmail.com> wrote:
>>
>> > My original use case is: I want to copy an association list, I know the function will probably have "copy" in its
>> > name and "alist", let's C-h f for "alist TAB" and fail because no results. Start again with "copy TAB" and filter
>> > the lots of "copy-*" function until I find copy-alist.
>>
>> "C-u C-h a copy alist RET" finds that function as the single hit.  As
>> does "C-u C-h a alist copy RET".
>
>
> Right, that works well. Thanks. It's not the most obvious keybinding but "C-h d copy alist" also works.
>
>
>>
>> > Now imagine if this particual function was named "asscpy"
>> > instead how frustrated I'd be :-)
>>
>> You will be frustrated because you use the wrong command.  "C-h f" is
>> not your friend unless you have a very good idea of the function's
>> name.  That's why "apropos" family of commands were invented: to help
>> you find something you don't know by name.  There's a best tool for
>> each job, and "C-h f" is not a good tool for this job.
>>
>> But I already said that, several times.  And since you still disagree,
>> then I must insist that my response to Richard was spot-on: he asked
>> why doesn't "C-h d" do the job, and my response was, in a nutshell,
>> that what "C-h d" does is not relevant for you and other users who
>> think and are accustomed to work like you do.  You disagreed with my
>> conclusion, but we now made one more full circle, and established that
>> my conclusion was correct after all.
>
>
> I think we are mixing two concepts here. To search for a single function, where you already know the keywords, yes "C-h d" works. To get a curated list of the API of a topic, not so much. Because my examples mix both finding a specific function and getting the curated list, I guess we were each talking about the specific function example and I want talking about the curated list.
>
> I still think "C-h d alist", because it does not give me assq and assoc is not doing a good job.
>
> Because alist is controversial let's take "C-h d regexp match": it does not give me string-match nor match-string. If I search for "regex match" then I find match-string.
>
> But overall I understand your point better now, you are not supposed to use "C-h d" to get a curated list of the api of a topic, it's just to find a single function. For that I agree it's a good tool.
>
>
>>
>> > Also imagine how self-documenting and self-discoverable Emacs Lisp would
>> > be if things where properly namespaced, like we ask for all packages to be on MELPA/ELPA and friends.
>>
>> I'm not objected to have aliases that would make it easier to find out
>> the function's name using simple completion, but I think you greatly
>> overestimate the usefulness of that in many practical situations.
>> Meanwhile there are existing features designed specifically for these
>> use cases, which do their job much more efficiently _today_, and you
>> choose to disregard them or treat them as second-class citizens.  I
>> think it's a mistake, but I can only make suggestions and point out
>> that those features do exist.
>
>
> I'm still not very satisfied about the tools you showed for listing an overview of the API to manipulate regexp, or files, or processes (etc). What you are basically saying is that wanting to have this overview is overrated, but for me it helps me understand the landscape of how the API works and is designed. What functions would be available should I need them in the future, what parts is missing, etc.
>
> For now the only tool I have is the manual page with "M-x keep-lines -- function".
>
>
>>
>> > As said earlier, probably that my way of thinking is not common around here, but I'd not be surprised if it was
>> > common for many developpers, maybe outside Emacs Lisp.
>>
>> My earnest advice for those developers is to try the features I
>> mentioned, they might find them useful in situations similar to the
>> one you described, and they might then decide to use them more than
>> what they do today.
>
>
> I'll certainly start to use it more.
>
> Thanks,
> Philippe



-- 
João Távora