Open djames-bloom opened 2 weeks ago
As mentioned, this issue happens when a sub-directory contains .git
paths
e.g. the following structure triggers the issue
.
├── dummy.go
├── main.go
└── subdir
└── .git
3 directories, 2 files
Sorry, not able to replicate this one on linux. Attempt:
: ; find /tmp/treedebug -ls
1588 0 drwxr-xr-x 3 alex users 100 Nov 10 09:18 /tmp/treedebug
1589 0 -rw-r--r-- 1 alex users 0 Nov 10 09:18 /tmp/treedebug/dummy.go
1590 0 -rw-r--r-- 1 alex users 0 Nov 10 09:18 /tmp/treedebug/main.go
1603 0 drwxr-xr-x 3 alex users 60 Nov 10 09:18 /tmp/treedebug/subdir
1604 0 drwxr-xr-x 2 alex users 40 Nov 10 09:18 /tmp/treedebug/subdir/.git
Collected a log, normalised and compared to yours:
: ; uudiff 2997.log actual.log
--- 2997.log 2024-11-10 09:23:53.799669092 +1100
+++ actual.log 2024-11-10 09:33:13.543302474 +1100
@@ -1,32 +1,36 @@
[profile] START core init /tmp/treedebug
[watcher] Watcher:create '/tmp/treedebug' nil
[watcher] Event:create '/tmp/treedebug'
[watcher] Event:start '/tmp/treedebug'
[profile] START git toplevel git_dir /tmp/treedebug
[git] git -C /tmp/treedebug rev-parse --show-toplevel --absolute-git-dir
-fatal: not a git repository (or any of the parent directories): .git
+fatal: not a git repository (or any parent up to mount point /)
+Stopping at filesystem boundary (GIT_DISCOVERY_ACROSS_FILESYSTEM not set).
[profile] END git toplevel git_dir /tmp/treedebug ms
[profile] START explore /tmp/treedebug
[profile] START populate_children /tmp/treedebug/dummy.go
[profile] END populate_children /tmp/treedebug/dummy.go ms
[profile] START populate_children /tmp/treedebug/main.go
[profile] END populate_children /tmp/treedebug/main.go ms
[profile] START populate_children /tmp/treedebug/subdir
[watcher] Watcher:create '/tmp/treedebug/subdir' nil
[watcher] Event:create '/tmp/treedebug/subdir'
[watcher] Event:start '/tmp/treedebug/subdir'
[profile] END populate_children /tmp/treedebug/subdir ms
[profile] END explore /tmp/treedebug ms
[profile] END core init /tmp/treedebug ms
+[profile] START view open
+[profile] END view open ms
[profile] START draw
[profile] END draw ms
-[profile] START reload /tmp/treedebug
-[profile] END reload /tmp/treedebug ms
[profile] START git toplevel git_dir /tmp/treedebug/subdir
[git] git -C /tmp/treedebug/subdir rev-parse --show-toplevel --absolute-git-dir
-fatal: not a git repository (or any of the parent directories): .git
+fatal: not a git repository (or any parent up to mount point /)
+Stopping at filesystem boundary (GIT_DISCOVERY_ACROSS_FILESYSTEM not set).
[profile] END git toplevel git_dir /tmp/treedebug/subdir ms
-[profile] START reload /tmp/treedebug/subdir
-[profile] END reload /tmp/treedebug/subdir ms
+[profile] START explore /tmp/treedebug/subdir
+[profile] START populate_children /tmp/treedebug/subdir/.git
+[profile] END populate_children /tmp/treedebug/subdir/.git ms
+[profile] END explore /tmp/treedebug/subdir ms
[profile] START draw
[profile] END draw ms
That +Stopping at filesystem boundary (GIT_DISCOVERY_ACROSS_FILESYSTEM not set).
might be interesting, or may just be a difference between git implementations. Have you got anything interesting or unusual in your ~/.gitconfig
?
I'd be grateful if you can perform some further tests:
/tmp
filesystem_watchers.enable = false
git.enable = false
Rolling back to minor versions:
cd /path/to/nvim-tree.lua
git pull
git checkout v1.7.0
When you're finished testing:
git checkout master
Hey Alex, thanks for looking into this
RE gitconfig, it doesn't look like theres anything in there that would be causing this kind of effect, but in case, you can find a copy here.
I'll update this comment once I've had a chance to run the additional tests.
Edit: testing the above results in the following, in a path in my home directory, using the cleanroom config
Disabling the above changes and testing with previous versions
v1.7.0
- slowv1.6.0
- slowv1.5.0
- slowv1.4.0
- slowfilesystem_watchers.enable = false - Appears to resolve - log
git toplevel takes 25ms, for me (linux) it's 5ms, however that's not a great issue. You might want to try git fsmonitor daemon
git.enable = false - No change - log
Matches my times.
It seem that the issue is the filesystem watchers, specifically closing them. Are you running any virus scanners or similar file system monitoring daemons?
This might be a vim problem. Please try this:
mkdir "${HOME}/fs_event_test"
cd "${HOME}/fs_event_test"
touch {1000..9999}
nvim
:source /tmp/foo.lua
Creation should be near-instant. Time exiting, also should be instant.
/tmp/foo.lua:
local fail
local FS_EVENT_FLAGS = {
stat = false,
recursive = false,
}
local function create_fs_event(path)
-- print("create_fs_event: " .. path)
local fs_event
fs_event, _, fail = vim.loop.new_fs_event()
if not fs_event then
print(path .. " " .. fail)
os.exit(1)
end
local rc
rc, _, fail = fs_event:start(path, FS_EVENT_FLAGS, function()
print("callback " .. path)
end)
if rc ~= 0 then
print(path .. " " .. fail)
os.exit(2)
end
end
for i = 1000, 9999, 1 do
create_fs_event(tostring(i))
end
print("created all events successfully")
Description
After upgrading to 0.10.*, closing nvim causes a long pause, up to 10 seconds.
This seems to be similar to #2438, but it is replicable with zero changes made to the cleanroom config, and is not related to
node_modules
or any similar large directory.This is replicable using the absolute barebones cleanroom config without any changes made.
There appears to be no relevant info in nvim-tree.log as all logged lines occurs before the attempted close.
This behaviour has two interesting cases
.git
directory, the bug occurs, it does not only apply to the root open path.git
directory simply exists without being an actual git repoNeovim version
Operating system and version
MacOS 14.1
Windows variant
No response
nvim-tree version
c763948
Clean room replication
Steps to reproduce
Expected behavior
Standard close behavior.
Actual behavior
A hang/delay of up to 10 seconds before nvim closes.