&shell suddenly set to sh instead of bash?

[user202729]

Description

While using vimtex, I notice once that &shell is suddenly set to sh. My default setting is bash.

Steps to reproduce

Currently not reproducible, but I suspect it's some sort of race condition in
autoload/vimtex/jobs/vim.vim → vim_unix_run -- maybe if two files attempt to call the function independently, it will break...?

VimtexInfo

Custom fork, actually. Might not be latest version, and I haven't tried reproducing it in latest version.

[lervag]

Strange. I believe this is relevant:

vimtex/autoload/vimtex/jobs/vim.vim

Lines 152 to 185 in 8dad59a

	let &shell = s:shell
	set shellcmdflag& shellquote& shellredir&
	silent! call system(a:cmd)
	let [ &shell,
	\ &shellcmdflag,
	\ &shellquote,
	\ &shellredir] = s:saveshell
	endfunction
	" }}}1
	function! s:vim_unix_capture(cmd) abort " {{{1
	let s:saveshell = [
	\ &shell,
	\ &shellcmdflag,
	\ &shellquote,
	\ &shellredir,
	\]
	let &shell = s:shell
	set shellcmdflag& shellquote& shellredir&
	silent! let l:output = systemlist(a:cmd)
	let [ &shell,
	\ &shellcmdflag,
	\ &shellquote,
	\ &shellredir] = s:saveshell
	return v:shell_error == 127 ? ['command not found'] : l:output
	endfunction
	let s:shell = executable('sh')
	\ ? 'sh'
	\ : (executable('/usr/bin/sh')
	\ ? '/usr/bin/sh' : '/bin/sh')

But it's hard to consider this without being able to reproduce any problems...

[user202729]

Anyway the reason why I suspect that is because the time I see that happen, I run :9verb set shell and see that that function is the culprit.

Is there any possibility the function can be called "asynchronously"? (i.e. is there some threading in vimscript?) Because if it's possible, it's certainly possible for two interleaving calls of the function to incorrectly restore the values:

thread A: backup saveshell = &shell (= bash)
thread A: set &shell = sh
thread B: backup saveshell = &shell (= sh)
etc.

then the original value (bash) is lost.

---

Edit: looks like there's a few calls of timer_start and job_start in the code. I'll see if any of them can cause concurrent execution.

Edit2: With some testing, looks like the call to system() or systemlist() are uninterruptable (even against job_start callback):

function H(channel, msg)
	echom "Callback " .. a:msg
endfunction
function I(job, exit_status)
	echom "Exit"
endfunction

call job_start(["bash", "-c", "sleep 0.5; echo a"], {"out_cb": function('H'), "out_io": "pipe", "exit_cb": function("I")})

echom "Hello"
call systemlist("sleep 2")
echom "Done"

so I'm not sure if it's possible to trigger a race condition this way.

Too bad I don't write down exactly which line causes the problem in the function, but if I recalled correctly it's the latter restoration line (i.e. s:saveshell) -- but either way if the culprit were a former line (i.e. it if the system() call is interrupted and the rest of the function is not executed? Is this possible?) then subsequent call will set s:saveshell to sh anyway.

[lervag]

Is there any possibility the function can be called "asynchronously"?

No, I don't think there is. Vimscript does not allow asynchronous stuff.

… i.e. it if the system() call is interrupted and the rest of the function is not executed? Is this possible?

Good question. It may be possible, but it would be very unlikely!