From f61efe3fe77c9a517dccb9fd5ff7f16c0660ced4 Mon Sep 17 00:00:00 2001
From: Evgeni Chasnovski <evgeni.chasnovski@gmail.com>
Date: Thu, 18 Jul 2024 18:26:27 +0300
Subject: perf(filetype): implement parent pattern pre-matching (#29660)

Problem: calling `vim.filetype.match()` has performance bottleneck in
  that it has to match a lot of Lua patterns against several versions of
  input file name. This might be the problem if users need to call it
  synchronously a lot of times.

Solution: add "parent pattern pre-matching" which can be used to quickly
  reject several potential pattern matches at (usually rare) cost of
  adding time for one extra Lua pattern match.

  "Parent pattern" is a manually added/tracked grouping of filetype
  patterns which should have two properties:
    - Match at least the same set of strings as its filetype patterns.
      But not too much more.
    - Be fast to match.

  For them to be effective, group should consist from at least three
  filetype patterns.

  Example: for a filetpye pattern ".*/etc/a2ps/.*%.cfg", both "/etc/"
  and "%.cfg" are good parent patterns (prefer the one which can group
  more filetype patterns).

  After this commit, `vim.filetype.match()` on most inputs runs ~3.4
  times faster (while some inputs may see less impact if they match
  many parent patterns).
---
 runtime/doc/dev_vimpatch.txt | 36 ++++++++++++++++++++++++++++++++++++
 1 file changed, 36 insertions(+)

(limited to 'runtime/doc')

diff --git a/runtime/doc/dev_vimpatch.txt b/runtime/doc/dev_vimpatch.txt
index d6e4ced054..6d450424c5 100644
--- a/runtime/doc/dev_vimpatch.txt
+++ b/runtime/doc/dev_vimpatch.txt
@@ -302,4 +302,40 @@ used in new documentation:
 - `{Only when compiled with ...}`: the vast majority of features have been
   made non-optional (see https://github.com/neovim/neovim/wiki/Introduction)
 
+==============================================================================
+FILETYPE DETECTION                 *dev-vimpatch-filetype*
+
+Nvim's filetype detection behavior matches Vim, but is implemented as part of
+|vim.filetype| (see $VIMRUNTIME/lua/vim/filetype.lua).
+
+Pattern matching has several differences:
+- It is done using explicit Lua patterns (without implicit anchoring) instead
+  of Vim regexes: >
+    "*/debian/changelog" -> "/debian/changelog$"
+    "*/bind/db.*" -> "/bind/db%."
+<
+- Filetype patterns are grouped by their parent pattern to improve matching
+  performance. For this to work properly, parent pattern should:
+    - Match at least the same set of strings as filetype patterns inside it.
+      But not too much more.
+    - Be fast to match.
+
+  When adding a new filetype with pattern matching, consider the following:
+  - If there is already a group with appropriate parent pattern, use it.
+  - If there can be a fast and specific enough pattern to group at least
+    3 filetype patterns, add it as a separate grouped entry.
+
+    Good new parent pattern should be:
+      - Fast. Good rule of thumb is that it should be a short explicit string
+        (i.e. no quantifiers or character sets).
+      - Specific. Good rules of thumb (from best to worst):
+          - Full directory name (like "/etc/", "/log/").
+          - Part of a rare enough directory name (like "/conf", "git/").
+          - String reasonably rarely used in real full paths (like "nginx").
+
+  Example:
+  - Filetype pattern: ".*/etc/a2ps/.*%.cfg"
+  - Good parent: "/etc/"; "%.cfg$"
+  - Bad parent: "%." - fast, not specific; "/a2ps/.*%." - slow, specific
+
 vim:tw=78:ts=8:noet:ft=help:norl:
-- 
cgit