Behaviour of watch_file on FreeBSD is different

[iblislin]

Revise.jl/src/Revise.jl

Lines 188 to 200 in 5754f39

	function watch_files_via_dir(dirname)
	watch_file(dirname) # this will block until there is a modification
	latestfiles = String[]
	wf = watched_files[dirname]
	for file in wf.trackedfiles
	path = joinpath(dirname, file)
	if mtime(path) + 1 >= floor(wf.timestamp) # OSX rounds mtime up, see #22
	push!(latestfiles, path)
	end
	end
	updatetime!(wf)
	latestfiles
	end

On FreeBSD modifying a existed file which located at dir will not being caught by watch_dir(dir).

[timholy]

I don't have access to a FreeBSD system, so I might need some help with this one. I did a little reading. Can you check kern.maxfiles? If it's set to a low value then supporting FreeBSD is probably hopeless for anyone who doesn't have root access to their machine to change that setting.

[iblislin]

Here is mine

└─[iblis@gungnir]% uname -a
FreeBSD gungnir.nba.nctu.edu.tw 11.1-RELEASE FreeBSD 11.1-RELEASE #0 r321309: Fri Jul 21 02:08:28 UTC 2017     root@releng2.nyi.freebsd.org:/usr/obj/usr/src/sys/GENERIC  amd64
┌─[~]
| [-- INSERT --]
└─[iblis@gungnir]% sysctl kern.maxfiles
kern.maxfiles: 261113
┌─[~]
| [-- INSERT --]
└─[iblis@gungnir]% sysctl kern.maxfilesperproc
kern.maxfilesperproc: 234999

Different FreeBSD version

└─[iblis@abeing]% uname -a
FreeBSD abeing 12.0-CURRENT FreeBSD 12.0-CURRENT #0 r323335: Sun Sep 17 00:56:35 CST 2017     root@abeing:/usr/obj/usr/src/sys/GENERIC  amd64     
┌─[~/git/TimeSeries]
| [Venv(py36)] [ master] [-- INSERT --]
└─[iblis@abeing]% sysctl kern.maxfiles                                                                                                            
kern.maxfiles: 522683                                                                                                                             
┌─[~/git/TimeSeries]
| [Venv(py36)] [ master] [-- INSERT --]
└─[iblis@abeing]% sysctl kern.maxfilesperproc                                                                                                     
kern.maxfilesperproc: 470412

[timholy]

It will be close, but perhaps possible:

~/.julia/v0.6$ find . -type f | wc -l
197127

[iblislin]

But the number 197127 is the worst case, right?

[iblislin]

└─[iblis@abeing]% find . -type f | wc -l                                 
   67527                            
┌─[~/.julia/v0.6]                   
| [Venv(py36)] [-- INSERT --]       
└─[iblis@abeing]% find . -type f -name '*.jl' | wc -l                                                                                             
    2951

[iblislin]

Here are VM images if you need: https://download.freebsd.org/ftp/releases/VM-IMAGES/

[timholy]

But the number 197127 is the worst case, right?

Correct, but I think it applies to all processes on your system. See #26 (comment).

[iblislin]

Any clue about implementation?
I can try to implement if you draw the picture of pseudo code.

[iblislin]

The days without workspace and reload make me anxious.

[timholy]

Sorry, I'm very busy on Julia Base with the push for 0.7-alpha, and didn't notice these pings. Having to first install a VM will also mean this won't be an easy fix for me to make.

This may not be too hard to fix because we used to watch individual files: see #22. You should be able to modify the watching code so that it checks for FreeBSD and if so uses a strategy that's more similar to the original one. I wouldn't try to do some kind of git revert though, because all kinds of stuff changed in between. Just check out the changes in #22 and then see if you can modify the current code accordingly.

I think the code is pretty well commented, but let me know if you have questions.