7.0 Release#

I am pleased to announce lsp-mode 7.0!

Here are the most important features/news in 7.0 release

lsp-mode/the team/the ecosystem#

lsp-mode team and the activity in the repo has vastly increased. lsp-mode is no longer one man project but it is a product of Emacs LSP team and emacs community. This has allowed us to implement some very time consuming changes/refactoring(more about them bellow) which an year ago were unthinkable due to the lack manpower. emacs-lsp repo has become a nice place to start contributing to open source or learn elisp and now there are 244 contributors in the main repo. Even more some of the current maintainers had zero elisp knowledge before doing their first lsp-mode contributions.

We started moving out some of the server specific packages(e. g. lsp-dart maintained by ericdallo and lsp-metals maintained by kurnevsky, lsp-python-ms by seagle0128, etc) in order to provide focused support by a an expert in the language at hand. Considering where we were few months ago the level of integrated experience provided by those packages has increased and often it is comparable with backed by paid full time developers editors like VScode. Here it is a gif illustrating that: lsp-dart. LSP in its nature should be language agnostic but in order to be comparable with VScode lsp-mode should support language server extensions as well which is much more work than the work on protocol itself and requires a constant monitoring of the server repo since there is no contract and the server team might change the contract whenever they want. On the bright side lsp-mode has attracted several server side developers or lsp-mode team members(e. g. mpanarin) has become server side contributors and often we are ahead of changes or we directly address issues in the server when they arise. We are still looking for maintainer(s) for JS/TS suite(typescript-language-server, eslint, angular) though.

Major features#

3.16 spec support#

We provided full support for 3.14 features one year after the version has been released. Now, we have support for all major features of 3.16 protocol even before it has been released (except for result streaming which apparently hasn't been implemented by any language server).

New website#

It is created ericdallo - here it is the website. Check it out and eventually provide feedback.

Semantic highlights#

This is part of 3.16 spec. It is implemented by sebastiansturm and it can be enabled by setting lsp-enable-semantic-highlighting. We have dropped the Theia protocol support for semantic highlighting since it is not going to make to the official spec. ATM it is supported by Rust Analyzer, Clangd(from master branch) and lua language server.

org-mode integration(preview)#

org-mode mode is the Emacs flagman and this integration brings up the literate programming to a new level. The integration is pretty raw but usable - you basically can run lsp-mode language intelligence over org-mode source block directly in the org-mode document. We have achieved that by creating virtual buffer abstraction which allows us to fake the server that the client has actually opened the real file.

Check gif and the docs.

I think that the ability to implement such features is why Emacs is editor in a league of its own. And implementing such features is one of the signs that we are slowly moving from the phase catching up into getting ahead (or maybe I am overly optimistic).

Language Server Protocol bindings, upcoming changes#

We generated json schema from the protocol and from that schema we generated dash destructoring, getters, setters, validators, constructors, indication for optional/required fields, etc. which allow us to perform compile time validation of the protocol usage. This was huge change because we had to walk through almost all methods and replace explicit access to the elisp data structure with protocol usage. At the same time, emacs-lsp organization has more than 20k lines of code. This was coordinated effort and a lot of individuals contributed to that. The benefits are that we now can switch the underlying data structure from hashmap to plists which will yield better performance and also that the overall quality of the codebase has increased. Example

Before:

(defun lsp--position-to-point (params)
  "Convert Position object in PARAMS to a point."
  (lsp--line-character-to-point (gethash "line" params)
                                (gethash "character" params)))

After:

(lsp-defun lsp--position-to-point ((&Position :line :character))
  "Convert `Position' object in PARAMS to a point."
  (lsp--line-character-to-point line character))

In this example :line and :character are validated as fields of Position structure. Also, we do some handy conversions, if the field is optional it will end up with ? like :foo?. To reduce boiler plate code we convert camelCase to :camel-case when we are destructoring it.

Performance optimizations(flycheck, code completion)#

We have rewritten the flycheck integration - the new integration provides the same UE as the old one but it is much more effective in reporting the errors to flycheck and lsp checker is much closer to the traditional flycheck checkers which are initiated/controlled from Emacs. Completion integration code is now much simpler/faster thanks to kiennq.

What's next#

Upcoming breaking changes
Dropping Emacs 25 support. This will allow us utilize Emacs multi-threading to improve lsp-mode responsibility
Switching from hashtables to plists - once this change goes in users must recompile all lsp-mode extensions.
Drop company-lsp support(not ported to use lsp-protocol.el)
Implementing lsp-mode configuration wizard to improve beginners experience - check this issue for more details. The overall goal is let user pick settings for all features that are opinionated (e. g. lsp-ui) and improve the discoverability of lsp-mode features/extensions/settings.
We will try to switch to monthly/weakly release cycles and to different person for writing the announcements. We might eventually encourage melpa-stable usage.

Last update: April 28, 2024