refactor: update impatient

[kylo252]

Description

• update impatient: merge the the latest upstream version. it still needs to be loaded before any plugins, so it will have to be managed manually either way
• feat: use an independent cache directory: overridden vim.fn.stdpath to use LUNARVIM_x_DIR instead, since a lot of plugins call this function interally

resolves #1256
fixes #2490

How to test

• added new tests for vim.fn.stdpath:
  

  LunarVim/tests/specs/bootstrap_spec.lua

  Lines 13 to 22 in 211fa20

  	a.it("should be able to use lunarvim directories using vim.fn", function()
  	assert.equal(lvim_runtime_path, vim.fn.stdpath "data")
  	assert.equal(lvim_config_path, vim.fn.stdpath "config")
  	assert.equal(lvim_cache_path, vim.fn.stdpath "cache")
  	end)
  	a.it("should be to retrieve default neovim directories", function()
  	local xdg_config = os.getenv "XDG_CONFIG_HOME"
  	assert.equal(join_paths(xdg_config, "nvim"), vim.call("stdpath", "config"))
  	end)

• check refactor: update impatient #2477 (comment)

Notes

For future references:

• it's easy to restore the original behavior of vim.fn.stdpath, by either removing some entries from https://github.com/kylo252/LunarVim/blob/978820b8edc56a12fd0c34664d6adabbe4e51734/lua/lvim/bootstrap.lua#L73-L83, or overriding the function, yet again, in config.lua

[abzcoding]

tested with a fuck ton of plugins

1. without this PR -> 240 to 260 ms
2. with this PR(uncommented) -> 270 to 300 ms
3. with this PR(setup commented) -> 320 to 600 ms

[kylo252]

1. without this PR -> 240 to 260 ms
2. with this PR(uncommented) -> 270 to 300 ms

Are the cache files growing in size with this PR? I noticed that it might be more gradual with this new upstream implemention.

[abzcoding]

Are the cache files growing in size with this PR? I noticed that it might be more gradual with this new upstream implemention.

I'll try it a bit more and report back

[abzcoding]

wtpr -> without this PR
pruc -> with this pr
prc -> with this pr and setup commented

State	Cache Size	Startup Time
wtpr: first	388008	1696
wtpr: second	2545147	266
wtpr: nth	3107775	235
pruc: first	28080+381451	408
pruc: second	231239+6153682	287
pruc: nth	231005+6154390	274
prc: first	0	575
prc: second	0	365
prc: nth	0	363

i'm on mac and this is my nvim version

NVIM v0.8.0-dev
Build type: Release
LuaJIT 2.1.0-beta3

---

update: here are the results for with this pr after a couple of hours

cache size

106103  lvim_luacache_modpaths
3446049 lvim_luacache_chunks

startuptime

257.299  000.002: --- NVIM STARTED ---
261.131  000.002: --- NVIM STARTED ---
263.070  000.002: --- NVIM STARTED ---
243.790  000.002: --- NVIM STARTED ---
259.039  000.002: --- NVIM STARTED ---
259.509  000.002: --- NVIM STARTED ---

[kylo252]

thanks for all the effort doing all those tests! so far the results seem to indicate a 10% loss in "reported" startup time, but we still get a better profiler, some other neat tricks with overloading loadfile, and hopefully get rid of all the remaining errors generated by gitsigns.

btw, have you made some script to automate this? otherwise I can try to adapt this one https://github.com/echasnovski/mini.nvim/tree/main/benchmarks/starter

[abzcoding]

hopefully get rid of all the remaining errors generated by gitsigns.

🤞

btw, have you made some script to automate this? otherwise I can try to adapt this one https://github.com/echasnovski/mini.nvim/tree/main/benchmarks/starter

no it was a manual process 😅

[kylo252]

Here are my results with my daily config, using hyperfine with 50 rounds

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
lvim	76.2 ± 1.0	74.0	78.7	1.11 ± 0.02
lv2	73.5 ± 1.1	71.5	76.1	1.07 ± 0.02
lv_no_cache	105.2 ± 1.4	102.3	109.1	1.54 ± 0.03
nvim	68.5 ± 1.1	65.7	71.1	1.00

the lv2 (e.g. ~/.local/bin/lv2) command is basically another instance of lvim with this branch

lv2

#!/usr/bin/env bash

export LUNARVIM_RUNTIME_DIR="/tmp/lv2/lunarvim"
export LUNARVIM_CONFIG_DIR="/tmp/lv2/config"
export LUNARVIM_CACHE_DIR="/tmp/lv2/cache"
export LUNARVIM_BASE_DIR="$LUNARVIM_RUNTIME_DIR/lvim"

# function install() {
#   local url="https://raw.githubusercontent.com/lunarvim/lunarvim/master/utils/installer/install.sh"
#   export LV_REMOTE="kylo252/lunarvim.git"
#   export LV_BRANCH="upstream-impatient"
#   bash <(curl -s "$url")
#   # NOTE: remember to use the same config
# }

# install

function lvim() {
  nvim -u "$LUNARVIM_BASE_DIR/init.lua" "$@"
}

lvim "$@"

the lv_no_cache (e.g. ~/.local/bin/lv_no_cache) command is basically another instance of lvim with this branch but the setup function turned off with en environment variable

diff --git a/lua/lvim/bootstrap.lua b/lua/lvim/bootstrap.lua
index d409e73..d0608f6 100644
--- a/lua/lvim/bootstrap.lua
+++ b/lua/lvim/bootstrap.lua
@@ -93,7 +93,7 @@ function M:init(base_dir)
   end
 
   -- FIXME: currently unreliable in unit-tests
-  if not in_headless then
+  if not in_headless and not os.getenv "DISABLE_IMPATIENT" then
     _G.PLENARY_DEBUG = false
     require("lvim.impatient").setup {
       chunks = { path = join_paths(get_cache_dir(), "lvim_luacache_chunks") },

benchmark.sh

#!/usr/bin/env bash
set -eo pipefail

results_dir="nv-benchmarks"

# Number of rounds to perform benchmark
n_rounds=50

# Path to output .csv file with startup times per round
csv_file="startup-times.csv"

# Path to output file with summary table
summary_json="startup-summary.json"
summary_file="startup-summary.md"

tmp_bench_file="$(mktemp)"

nvim_bins="lvim,lv2,nvim"

function benchmark {
  rm -rf "$results_dir"
  mkdir -p "$results_dir"

  pushd "$results_dir"

  local after="tail -n 1 $tmp_bench_file | cut -d ' ' -f1 >> $csv_file"
  local cmd="{nvim_bin} +'lua vim.schedule(function() vim.cmd [[qall]] end)' --startuptime $tmp_bench_file && $after"
  local setup="rm -f $csv_file && echo {nvim_bin} > $csv_file"
  # shellcheck disable=2155
  local cleanup="mv $csv_file {nvim_bin}_$csv_file"
  hyperfine --runs "$n_rounds" \
    --setup "$setup" \
    --export-json "$summary_json" \
    --export-markdown "$summary_file" \
    --cleanup "$cleanup" \
    --warmup 3 \
    --command-name "{nvim_bin}" \
    --parameter-list nvim_bin "$nvim_bins" "$cmd"

  popd
}

benchmark

[kylo252]

@abzcoding, here are more tests using your own config on an M1 machine

the lv3 (e.g. ~/.local/bin/lv3) command is basically another instance of lvim with this branch impatient-v3, which incorporates a new workaround for handling the cache dir, see kylo252@8ccca67, which seems to give the best results out of all three implementations!

lv2

#!/usr/bin/env bash

export LUNARVIM_RUNTIME_DIR="/tmp/lv3/lunarvim"
export LUNARVIM_CONFIG_DIR="/tmp/lv3/config"
export LUNARVIM_CACHE_DIR="/tmp/lv3/cache"
export LUNARVIM_BASE_DIR="$LUNARVIM_RUNTIME_DIR/lvim"

function install() {
  local
  url="https://raw.githubusercontent.com/lunarvim/lunarvim/master/utils/installer/install.sh"
  export LV_REMOTE="kylo252/lunarvim.git"
  export LV_BRANCH="impatient-v3"
  bash <(curl -s "$url")

  # NOTE: remember to use the same config
  rm -rf $LUNARVIM_CONFIG_DIR
  gh repo clone abzcoding/lvim $LUNARVIM_CONFIG_DIR

  mkdir -p $LUNARVIM_CACHE_DIR
}

# install

function lvim() {
  nvim -u "$LUNARVIM_BASE_DIR/init.lua" "$@"
}

lvim "$@"

[abzcoding]

fantastic, let's merge the third implementation then 👍

[kylo252]

fantastic, let's merge the third implementation then 👍

should I move the vim.fn.stdpath change into a separate PR? wdyt?

[abzcoding]

fantastic, let's merge the third implementation then 👍

should I move the vim.fn.stdpath change into a separate PR? wdyt?

let's keep it in the same PR, it is gonna be one big PR, but at least all changes related to impatient would be in the same place

[abzcoding]

not sure why, but it seems like it doesn't recognize the installed lang servers

[kylo252]

let's keep it in the same PR, it is gonna be one big PR, but at least all changes related to impatient would be in the same place

I think the CI errors are related to how tmp files are handled on macos, could you try the branch and run the tests locally?

[kylo252]

not sure why, but it seems like it doesn't recognize the installed lang servers

It's because it's now using an independent vim.fn.stdpath, you need to install them again or move them manually. We could also the lsp-install directory to match the one used before ~/.local/share/nvim/lsp_servers

[abzcoding]

I think the CI errors are related to how tmp files are handled on macos, could you try the branch and run the tests locally?

this env var is not available on mac machines
local xdg_config = os.getenv "XDG_CONFIG_HOME"

[abzcoding]

not sure why, but it seems like it doesn't recognize the installed lang servers

It's because it's now using an independent vim.fn.stdpath, you need to install them again or move them manually. We could also the lsp-install directory to match the one used before ~/.local/share/nvim/lsp_servers

we should probably do that, since the language server binaries are exactly the same for nvim and lvim

[abzcoding]

LGTM 🚀