Unexpectedly leaving insert mode when job writes to hidden buffer using prompt buffers (regression by patch 8.2.1654) #7035

puremourning · 2020-09-28T17:40:28Z

Describe the bug

After patch 8.2.1654, insert mode is exited unexpectedly using prompt-buffers when a job writes to a hidden buffer.

I used the repro case to bisect this to 8.2.1654 patch.

To Reproduce
Detailed steps to reproduce the behavior:

test.vim

set buftype=prompt

call prompt_setprompt( bufnr(), 'Enter some text: ' )

call job_start( [ 'uuencode', '/dev/urandom', '/dev/stdout' ], #{
      \ out_io: 'buffer',
      \ out_name: ''
      \ } )

vim -Nu NONE -S test.vim
i
type something

You will be bumped unexpectedly out of insert mode into normal mode as you're typing.

Expected behavior

Remain in insert mode until pressing <CR> in the prompt buffer.

Environment (please complete the following information):

Vim version 8.6.1654 and later
OS: Linux RHEL 7, macOS
Terminal: tmux, PuTTY, iTerm

The text was updated successfully, but these errors were encountered:

brammool · 2020-09-28T19:18:01Z

I'm surprised patch 8.6.1654 would cause this, and only now gets reported.
Are you sure it's not related to 8.2.1762 ?

puremourning · 2020-09-28T20:30:55Z

I can check but I bisected and it indicated the above commit. I can confirm that with 8.2.1762 we no longer have the pum left open, but it still exits insert mode unexpectedly. With 8.1.1654, you end up with both - exit insert mode unexpectedly and the pum is left open.

tonymec · 2020-09-28T20:44:50Z

Does it happen also in gvim?
What are the answers to

:set timeout? timeoutlen? ttimeout? ttimeoutlen?

(including the question marks)?

Best regards,
Tony.

puremourning · 2020-09-28T21:22:31Z

Does it happen also in gvim?

Will check motif GUI at work tomorrow, sorry don't have a gvim here right now.

For the record, it does repro in MacVim (GUI) with:

ben@BenMBP vim-clean % gvim --version
VIM - Vi IMproved 8.2 (2019 Dec 12, compiled Sep 28 2020 22:02:11)
macOS version
Included patches: 1-1768
Compiled by Homebrew
Huge version with MacVim GUI.  Features included (+) or not (-):

What are the answers to :set timeout? timeoutlen? ttimeout? ttimeoutlen?

timeout
timeoutlen=1000
ttimeout
ttimeoutlen=100

vim-ml · 2020-09-28T23:32:40Z

On Mon, Sep 28, 2020 at 11:23 PM Ben Jackson ***@***.***> wrote: Does it happen also in gvim? Will check motif GUI at work tomorrow, sorry don't have a gvim here right now. For the record, it *does* repro in MacVim with: ***@***.*** vim-clean % gvim --version VIM - Vi IMproved 8.2 (2019 Dec 12, compiled Sep 28 2020 22:02:11) macOS version Included patches: 1-1768 Compiled by Homebrew Huge version with MacVim GUI. Features included (+) or not (-): What are the answers to :set timeout? timeoutlen? ttimeout? ttimeoutlen? [image: Screenshot 2020-09-28 at 22 21 56] <https://user-images.githubusercontent.com/10584846/94487504-09dde480-01d9-11eb-8c34-ec4fec370429.png> - timeout - timeoutlen=1000 - ttimeout - ttimeoutlen=100 Hm, with the above settings, Vim ought to be able to tell the difference

between an Esc key hit at the keyboard and a ^[ sent as part of a keycode, so I suppose it isn't that. Also, if the problem happens in the MacVim GUI I would expect it to happen also in an X11 or W32/64 GUI but I'm less sure about that. Best regards, Tony.

puremourning · 2020-09-29T07:00:40Z

Yes Tony I’m fairly sure that this is not a mapping issue. It’s as if something is calling stopinsert. I have a couple of theories of how to minimise the repro case. Will try after work hopefully.

Re the point about being not reported before, it seems that it’s only prompt buffers that are affected, which are not frequently used.

puremourning · 2020-09-29T13:10:00Z

OK, this is easy to repro. Here's a trivial case.

set buftype=prompt

call prompt_setprompt( bufnr(), 'Enter some text: ' )

call job_start( [ 'uuencode', '/dev/urandom', '/dev/stdout' ], #{
      \ out_io: 'buffer',
      \ out_name: ''
      \ } )

vim -Nu NONE -S test.vim
enter insert mode i and type some characters - you're bumped out of insert mode

puremourning · 2020-09-29T13:23:04Z

Here's a test that repros it:


function! Test_Prompt_While_Writing_To_Hidden_Buffer()
  call CanTestPromptBuffer()
  let scriptName = 'XpromptscriptHiddenBuf'
  let script =<< trim END
    set buftype=prompt
    call prompt_setprompt( bufnr(), 'cmd:' )
    let job = job_start( [ 'uuencode', '/dev/urandom', '/dev/stdout' ], #{
      \ out_io: 'buffer',
      \ out_name: '' } )
    startinsert
  END
  exec script->writefile( scriptName )

  let buf = RunVimInTerminal('-S ' . scriptName, {})
  call WaitForAssert({-> assert_equal('cmd:', term_getline(buf, 1))})

  call term_sendkeys( buf, 'test' )
  call WaitForAssert( {-> assert_equal('cmd:test', term_getline(buf, 1)) } )

  call StopVimInTerminal(buf)
  call delete(scriptName)
endfunc

|| command line..script /Users/ben/Development/vim/src/testdir/runtest.vim[459]..function RunTheTest[39]..Test_Prompt_While_Writing_To_Hidden_Buffer[17]..WaitForAssert[2]..6_WaitForCommon[11]..3 line 1: Expected 'cmd:test' but got 'cmd:t'

puremourning · 2020-09-29T20:45:41Z

Been debugging this. I'm thinking that it's related to this:

    static void
leaving_window(win_T *win)
{
    // Only matters for a prompt window.
    if (!bt_prompt(win->w_buffer))
	return;

    // When leaving a prompt window stop Insert mode and perhaps restart
    // it when entering that window again.
    win->w_buffer->b_prompt_insert = restart_edit;
    if (restart_edit != 0 && mode_displayed)
	clear_cmdline = TRUE;		// unshow mode later
    restart_edit = NUL;

    // When leaving the window (or closing the window) was done from a
    // callback we need to break out of the Insert mode loop and restart Insert
    // mode when entering the window again.
    if (State & INSERT)
    {
	stop_insert_mode = TRUE;
	if (win->w_buffer->b_prompt_insert == NUL)
	    win->w_buffer->b_prompt_insert = 'A';
    }
}

    static void
entering_window(win_T *win)
{
    // Only matters for a prompt window.
    if (!bt_prompt(win->w_buffer))
	return;

    // When switching to a prompt buffer that was in Insert mode, don't stop
    // Insert mode, it may have been set in leaving_window().
    if (win->w_buffer->b_prompt_insert != NUL)
	stop_insert_mode = FALSE;

    // When entering the prompt window restart Insert mode if we were in Insert
    // mode when we left it.
    restart_edit = win->w_buffer->b_prompt_insert;
}

When using aucmd_prepwin we actually leave the window in order to write to the hidden buffer, and we're calling leaving_window(). . But it seems like we're not calling entering_window() for the prompt buffer when reverting this temporary window.

For certain when we do call entering_window(), bt_prompt(win->w_buffer) is false. I'ma summing that's actually entering the temporary autocmd window.

Are we missing a call to win_enter in aucmd_restbuf() ? That seems intrusive, but I haven't fully groked this code yet.

brammool · 2020-09-29T20:57:54Z

I can't reproduce because "uuencode: Command not found". Perhaps something more common (and not keeping the CPU busy)?

brammool · 2020-09-29T21:01:13Z

Oh, and there is no aucmd_prepwin(), you probably mean aucmd_prepbuf(). But it doesn't call leaving_window() directly.

puremourning · 2020-09-29T21:03:42Z

Just need a job to write to a hidden buffer while in insert mode.

This repro's it too

set buftype=prompt

call prompt_setprompt( bufnr(), 'Enter some text: ' )

augroup TEST
  au InsertEnter call job_start( [ 'ls' ], #{
      \ out_io: 'buffer',
      \ out_name: ''
      \ } )
augroup END

this seems to resolve it:

ben@BenMBP vim-clean % git diff
diff --git a/src/autocmd.c b/src/autocmd.c
index d5c61ca6a..1789d0477 100644
--- a/src/autocmd.c
+++ b/src/autocmd.c
@@ -1530,6 +1530,7 @@ win_found:
 
        restore_snapshot(SNAP_AUCMD_IDX, FALSE);
        (void)win_comp_pos();   // recompute window positions
+       win_enter(aco->save_curwin, TRUE);
        unblock_autocmds();
 
        if (win_valid(aco->save_curwin))
ben@BenMBP vim-clean %

puremourning · 2020-09-29T21:06:01Z

Oh, and there is no aucmd_prepwin(), you probably mean aucmd_prepbuf(). But it doesn't call leaving_window() directly.

Yeah I meant aucmd_prepbuf() - this calls win_split_ins(), which calls win_enter_ext(wp, FALSE, FALSE, TRUE, TRUE, TRUE); ... and that calls leaving_window()

puremourning · 2020-09-29T21:38:07Z

OK here's a test which repros it too; had to introduce some sleeps though :/

function! Test_Prompt_While_Writing_To_Hidden_Buffer()
  call CanTestPromptBuffer()
  let scriptName = 'XpromptscriptHiddenBuf'
  let script =<< trim END
    set buftype=prompt
    call prompt_setprompt( bufnr(), 'cmd:' )
    let job = job_start( [ '/bin/sh', '-c', 'while true; do echo line; sleep 0.2; done' ], #{
      \ out_io: 'buffer',
      \ out_name: '' } )
    startinsert
  END
  exec script->writefile( scriptName )

  let buf = RunVimInTerminal('-S ' . scriptName, {})
  call WaitForAssert({-> assert_equal('cmd:', term_getline(buf, 1))})

  call term_sendkeys( buf, 'test' )
  call WaitForAssert( {-> assert_equal('cmd:test', term_getline(buf, 1)) } )
  sleep 10m
  call term_sendkeys( buf, 'test' )
  call WaitForAssert( {-> assert_equal('cmd:testtest', term_getline(buf, 1)) } )
  sleep 10m
  call term_sendkeys( buf, 'test' )
  call WaitForAssert( {-> assert_equal('cmd:testtesttest', term_getline(buf, 1)) } )

  call StopVimInTerminal(buf)
  call delete(scriptName)
endfunc

tjdevries · 2020-09-30T00:38:15Z

I was wondering if this patch (that I just made myself when I first encountered the bug) is useful:

neovim/neovim#12977

I have no idea if this is the "right" solution or not, just figured I'd show what I was thinking. I just expected :stopinsert to just completely shutdown everything that could be happening in insert mode.

happy-dude · 2020-10-01T02:55:50Z

Just so no one is confused about @puremourning 's test script, WaitForAssert is defined in https://github.com/puremourning/vimspector/blob/4ed915828285c067b7f4a35fecc7e0ed56fc9be8/tests/lib/plugin/shared.vim.

I believe you are calling helper functions from the vimspector plugin, puremourning?

EDIT: if you can upload your full script (and dependencies), it could help us run your test case 👍
As it is now, CanTestPromptBuffer, RunVimInTerminal and other functions/vars are undefined.

puremourning · 2020-10-01T06:47:53Z

@happy-dude my test runs in Vim’s test suite. Paste it into test_promt_buffer.vim

Those functions are defined in vim’s shared.vim and friends.

brammool · 2020-10-01T17:56:41Z

Thanks for figuring this out and writing a test. I'll clean it up a bit.

puremourning · 2020-10-01T20:38:06Z

Thanks Bram. That's working nice now.

Problem: Writing to prompt buffer interferes with insert mode. Solution: Use win_enter() instead of just setting "curwin". (Ben Jackson, closes vim/vim#7035) vim/vim@4537bcc Patch v8.2.1783 has fixes.

Problem: Writing to prompt buffer interferes with insert mode. Solution: Use win_enter() instead of just setting "curwin". (Ben Jackson, closes vim/vim#7035) vim/vim@4537bcc

Problem: Writing to prompt buffer interferes with insert mode. Solution: Use win_enter() instead of just setting "curwin". (Ben Jackson, closes vim/vim#7035) vim/vim@4537bcc Vim test will be skipped, so add a Lua test. The problem boils down to the use of aucmd_restbuf in a callback, so just test that (via nvim_buf_set_lines).

happy-dude mentioned this issue Sep 28, 2020

fix: #12976 neovim/neovim#12977

Closed

puremourning changed the title ~~Unexpectedly leaving insert mode when completion is triggered in prompt buffer (regression by patch 8.2.1654)~~ Unexpectedly leaving insert mode when job writes to hidden buffer using prompt buffers (regression by patch 8.2.1654) Sep 29, 2020

brammool closed this as completed in 4537bcc Oct 1, 2020

happy-dude mentioned this issue Oct 1, 2020

:stopinsert does not clean up insert mode properly. neovim/neovim#12976

Closed

happy-dude mentioned this issue Oct 2, 2020

vim-patch:8.2.1762: when a timer uses :stopinsert completion isn't stopped neovim/neovim#13025

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Unexpectedly leaving insert mode when job writes to hidden buffer using prompt buffers (regression by patch 8.2.1654) #7035

Unexpectedly leaving insert mode when job writes to hidden buffer using prompt buffers (regression by patch 8.2.1654) #7035

puremourning commented Sep 28, 2020 •

edited

brammool commented Sep 28, 2020

puremourning commented Sep 28, 2020

tonymec commented Sep 28, 2020

puremourning commented Sep 28, 2020 •

edited

vim-ml commented Sep 28, 2020 via email

puremourning commented Sep 29, 2020

puremourning commented Sep 29, 2020

puremourning commented Sep 29, 2020

puremourning commented Sep 29, 2020

brammool commented Sep 29, 2020

brammool commented Sep 29, 2020

puremourning commented Sep 29, 2020

puremourning commented Sep 29, 2020

puremourning commented Sep 29, 2020 •

edited

tjdevries commented Sep 30, 2020

happy-dude commented Oct 1, 2020 •

edited

puremourning commented Oct 1, 2020

brammool commented Oct 1, 2020

puremourning commented Oct 1, 2020

Unexpectedly leaving insert mode when job writes to hidden buffer using prompt buffers (regression by patch 8.2.1654) #7035

Unexpectedly leaving insert mode when job writes to hidden buffer using prompt buffers (regression by patch 8.2.1654) #7035

Comments

puremourning commented Sep 28, 2020 • edited

brammool commented Sep 28, 2020

puremourning commented Sep 28, 2020

tonymec commented Sep 28, 2020

puremourning commented Sep 28, 2020 • edited

vim-ml commented Sep 28, 2020 via email

puremourning commented Sep 29, 2020

puremourning commented Sep 29, 2020

puremourning commented Sep 29, 2020

puremourning commented Sep 29, 2020

brammool commented Sep 29, 2020

brammool commented Sep 29, 2020

puremourning commented Sep 29, 2020

puremourning commented Sep 29, 2020

puremourning commented Sep 29, 2020 • edited

tjdevries commented Sep 30, 2020

happy-dude commented Oct 1, 2020 • edited

puremourning commented Oct 1, 2020

brammool commented Oct 1, 2020

puremourning commented Oct 1, 2020

puremourning commented Sep 28, 2020 •

edited

puremourning commented Sep 28, 2020 •

edited

puremourning commented Sep 29, 2020 •

edited

happy-dude commented Oct 1, 2020 •

edited