Calling eldoc-message directly

Calling eldoc-message directly

[Clément Pit-Claudel]

Hi emacs-devel,

The changelog for Emacs 27.1 says "For ElDoc support, you should set 'eldoc-documentation-function' instead of calling 'eldoc-message' directly".

At least one of my packages calls eldoc-message directly, and I'm not sure how to get rid of it.

In that package, I use a server process to get information about the current buffer.  I can use it to asynchronously retrieve documentation about identifiers, among other things.  This pattern seems pretty common (I think that's how tide, elpy, cider, and multiple other modes work).

I currently have an eldoc-documentation-function that sends an asynchronous query, waits for a few milliseconds in the hope to get a quick answer, and otherwise registers a callback that calls eldoc-message directly.  That callback checks that the buffer hasn't been modified and that the point hasn't moved, and if so calls eldoc-message.

Can someone recommend a proper implementation, not based on eldoc-message? (I have vague recollections of a related thread on emacs-devel, but I can't seem to find it)

Thanks,
Clément.

Re: Calling eldoc-message directly

[João Távora]

Clément Pit-Claudel <cpitclaudel@gmail.com> writes:

> Hi emacs-devel,
>
> The changelog for Emacs 27.1 says "For ElDoc support, you should set
> 'eldoc-documentation-function' instead of calling 'eldoc-message'
> directly".
>
> At least one of my packages calls eldoc-message directly, and I'm not
> sure how to get rid of it.

I'm quite interested in a solution to this one, too.

This thread where Dmitry is suggesting something to this effect might
be relevant:

https://github.com/slime/slime/issues/400#issuecomment-329616259

João

Re: Calling eldoc-message directly

[Stefan Monnier]

> In that package, I use a server process to get information about the current
> buffer.  I can use it to asynchronously retrieve documentation about
> identifiers, among other things.  This pattern seems pretty common (I think
> that's how tide, elpy, cider, and multiple other modes work).

Indeed async use is currently not supported.

Not sure how to best add support for sync-processing.
Ideally, we'd pass a "continuation" to the eldoc-documentation-function,
but that would break existing functions which wouldn't expect that extra arg.

We could pass that extra arg via a dynamically-scoped variable like
`eldoc-continuation-function`.

An alternative would be to allow eldoc-documentation-function to return
a function, which is then called with the continuation.

So you'd do something like

    (defun my-eldoc-documentation-function (orig-fun)
      (if (not (my-determine-if-we-should-be-in-charge))
          (funcall orig-fun)
        (lambda (k)
          (let ((p (start-process ...)))
            (set-process-sentinel p
              (lambda (_p status)
                (funcall k (my-get-doc-string))))))))
    [...]
      (add-function :around (local 'eldoc-documentation-function)
                    #'my-eldoc-documentation-function)


-- Stefan

Re: Calling eldoc-message directly

[Clément Pit-Claudel]

On 2018-05-09 22:01, Stefan Monnier wrote:
>> In that package, I use a server process to get information about the current
>> buffer.  I can use it to asynchronously retrieve documentation about
>> identifiers, among other things.  This pattern seems pretty common (I think
>> that's how tide, elpy, cider, and multiple other modes work).
>
> Indeed async use is currently not supported.
>
> Not sure how to best add support for sync-processing.
> Ideally, we'd pass a "continuation" to the eldoc-documentation-function,
> but that would break existing functions which wouldn't expect that extra arg.
>
> We could pass that extra arg via a dynamically-scoped variable like
> `eldoc-continuation-function`.
>
> An alternative would be to allow eldoc-documentation-function to return
> a function, which is then called with the continuation.

That would be nice.  This model works pretty well in company-mode.