package-local-nicknames support

[remexre]

Right now, I don't get type hints or completion when a symbol uses a p-l-n name.

[kovisoft]

Could you please give me an example code snippet? I'd like to check how it is handled in Emacs/Slime, what is the communication between Emacs and the Swank server.

[remexre]

(uiop:define-package :example
  (:use :cl)
  (:local-nicknames (#:a :alexandria)))
(in-package :example)

(a:define-constant foo :foo)

This does need a new enough UIOP; iiuc, 3.3.4 for SBCL, 3.3.5 for other impls.

[kovisoft]

Thank you for the example. I played with it a little bit in emacs+slime, and it turned out that emacs+slime is sending swank:completions to the swank server, while vim+slimv is sending either swank:fuzzy-completions or swank:simple-completions (depending on the value of g:slimv_simple_compl). This might be a change in slime, because as I recall, older versions used swank:simple-completions as well, that's why I implemented it the same way in slimv.

I'm going to check how to add swank:completions to slimv. Maybe it will be a new option. I'll come back to you when I have a solution.

[stag-enterprises]

Is there any progress update on this?

[kovisoft]

Sorry for the big delay. I have fixed it in commit a7a4715
The default behavior is now to use swank:completions that handles local-nicknamed packages.

[stag-enterprises]

@kovisoft Thank you so much! Appreciate your work. This can probably be closed now.

[stag-enterprises]

Update: It looks like while autocomplete for local-nicknames works, there is no argument hinting. When typing the name of a function under a local-nicknames package, the little text shown on the bottom of the screen with the argument list does not show.

[kovisoft]

It seems that emacs uses swank:autodoc for getting the argument list, and this one handles local-nicknames as well. Therefore in commit bbdb7a4 I replaced operator-arglist with autodoc for .lisp files. Unfortunately autodoc is not supported for all file types (e.g. it does not work in swank-mit-scheme.scm).

[stag-enterprises]

ty! for sbcl it looks like everything works fine.

[Jach]

I generally prefer the fuzzy matching behavior so it's been somewhat frustrating that I've had to choose between fuzzy matching or having package-local nickname support. I have a couple keybinds to toggle between the two but it's usually left in fuzzy mode.

I'm now experimenting with a change on a local branch to add nickname support for the fuzzy matching case. This adds some Common Lisp specific code into the Python code which seems contrary to the project's overall design but I'm not sure how else to do it besides modifying the underlying swank call somehow. (Normally for local hacks of this sort I'd put a supporting function in my ~/swank.lisp file to get auto loaded.) If desired I can make a pull request after adding some further gating such as a new g:slimv flag so it's not default behavior, and indirectly getting the UIOP functions to prevent errors if ASDF/UIOP isn't loaded for some reason.

With the context of

(defpackage #:test-pkg
  (:use #:cl)
  (:local-nicknames (#:b #:bt)
                    (#:a #:alexandria)))
(in-package :test-pkg)

If I have g:slimv_completions = 'swank' (or empty) and type (bt:mt<tab> then I get nothing, as there's no fuzzy matching.
If I type (bt:make-<tab> then I get bt:make-conidtion-variable, bt:make-lock, bt:make-recursive-lock, bt:make-semaphore, and bt:make-thread. I can also type (b:make-<tab> and get the same matches, except now the b: is preserved so it's b:make-condition-variable etc. Typing (c:make-<tab> also results in nothing.

Setting g:slimv_completions = 'fuzzy' and before my changes, (bt:mt<tab> results in bt:make-thread, bt2:make-thread, bt:make-condition-variable, bt:start-multiprocessing, ... However (b:mt<tab> results in bt:make-thread, bt2:make-thread, base64:make-decode-table, babel:make-external-format, serapeum/bundle:map-tree, ... This is consistent with the fuzzy match looking at packages that have b in their name, as well as doing a fuzzy match in each for the mt part. The annoying behavior is that for the match I want (the first one), it will replace the local nickname prefix with whatever the match has.

After my changes, typing (bt:mt<tab> results in b:make-thread, bt2:make-thread, b:make-condition-variable, ... In other words my explicitly typed bt gets replaced to the local nickname of b, but the bt part was still used in the initial filter. Same thing if I type alexandria:<tab>, the suggestion list becomes a:, alexandria-2:, a:alist-hash-table, a:alist-plist, and so on for the rest of alexandria's symbols. If I type (b:mt then I get back b:make-thread, bt2:make-thread, base64:make-decode-table, ... In other words the nickname b is still being fuzzy matched for packages but if a package has a local nickname of b then it'll be preserved. And finally of course (c:mt<tab> results in cl:machine-type, cl:make-hash-table, etc.