GeezerVison

I've been experimenting with a custom algorithm as per the check list item above. This version eliminates all syntax highlight checks (and therefore no longer requires the various hacks to make those checks work) plus it is approx the same speed as my first rewrite attempt and uses about the same amount of code.

However the core benefits of this version is that it works when syntax highlighting is switched off, and eliminates the performance bottle neck of syntax highlight checks. This means that it will be possible to write a version of the algorithm in Vim9 script, to get another significant performance improvement! (Maintaining both a legacy Vim script and Vim9 script versions should be feasible due to how much simpler this is than the original implementation.)

TL;DR: I'll push a new version to this branch soon which will be:

1. Simpler with no hacks,
2. Can be become even faster (via Vim9 script),
3. Works even when syntax highlighting is switched off.

Once this branch is feature complete, I'll share benchmarks comparing it to the original as is on master.

Performance benchmarks as of current change (+ unpushed Vim9 version). This is measuring the time to reindent a whole file on a Macbook Pro 2021, so expect much more noticable performance gains on slower hardware. 🚀

(I'll share new benchmarks (and the full reports generated by Vim) once this branch is feature complete, although I don't expect it to change much.)

Even with the Vim9 implementation in the code too (not pushed yet, still figuring out how best to integrate it in a maintainable way), we can expect the file size of indent/clojure.vim to be ~150 LOC less than the original!

Regarding Neovim, until Vim9 script support is added, the "New (legacy)" code will be used. If anyone would like to volunteer to write a Lua implementation of this code for Neovim please open an issue and we can discuss how to integrate it.

FYI, the "uniform" indent_style isn't quite following Tonsky's Better Clojure formatting

Multi-line lists that start with a symbol are always indented with two spaces,
Other multi-line lists, vectors, maps and sets are aligned with the first element (1 or 2 spaces).

Multi-line lists that start with a non-symbol are currently indented with 2 instead of 1 space.

E.g.

(ns namespace
  (:require
   [clojure.set :as s]))

The final line is indented 1 space because :require is keyword and not a symbol.

if indent_style ==# 'uniform' | return base_indent + 1 | endif

Not sure the best way way for this to check if the first list item is a symbol or not. 🤔

I wrote this for #21 before I realized this rewrite was being done. It's a pattern that will match symbols, adapted from the tree-sitter grammar. (Though it may be easier to just special case non-symbols as first list item.)

local suffixonly = ":#'0-9\\/" -- not allowed as first character
local forbidden = {
  [[\[\]\n\r\t \\]], -- brackets, whitespace, slashes
  '(){}"@~^;`,', -- parens, braces, special characters
  suffixonly,
}
local prefix = string.format("[^%s]", table.concat(forbidden))
local suffix = string.format([[\(%s\|[%s]\)*]], prefix, suffixonly)
local symbol = prefix .. suffix

[^\[\]\n\r\t \\(){}"@~^;`,:#'0-9\/]\([^\[\]\n\r\t \\(){}"@~^;`,:#'0-9\/]\|[:#'0-9\/]\)*

Thanks for raising this @garrett-hopper. I'll add it to my list and try to fix it soon.

(Yet more mildly irritating rules in the common Clojure indentation styles 😆 😭)

Apologies to anyone waiting patiently for this to be merged. I haven't managed to find much time to dedicate to this and each time I come back to it, it requires a lot of context reloading as the standard Clojure indentation styles are shockingly all very complicated (contrary to Clojure itself) and to implement them efficiently in Vim script is forever painful.

Currently what exists in this branch is IMO already far superior to what is currently on the master branch, so my current plan is:

1. fix a few of the more major outstanding issues,
2. document the clojure_indent_rules option with warnings that it may change in the future,
3. merge, 🥳
4. open issues for the remaining known problems that I don't consider to be dealbreakers,
5. gather more feedback on other issues with wider use,
6. fix issues critical to the core algorithm,
7. later complete Create Vim 9 script variant of the new indentation algorithm #35 and Port performance improvements to Neovim/Lua #32 once the core algorithm is perfected to avoid needing to implement the same fixes 3 times.

I hope to get round to the first 4 soon, but can't make any promises on when I will do so.

Pretty exciting! Thanks for shepherding this! I've been using this branch exclusively since you opened the PR and I'm quite pleased with it.

Bug report: defining protocol methods in defrecord under "standard" still indents the body to match the arg vector:

(defrecord Example [a b c]
  MyProto
  (method1 [this d e f]
    (+ a b c d e f)))

becomes

(defrecord Example [a b c]
  MyProto
  (method1 [this d e f]
           (+ a b c d e f)))

It looks like there's tests covering this, so I'm not sure what's wrong.