This reverts commit ab8371a26c.
Need to think of a different solution, which may require adding new
flags to nvim_ui_attach() (e.g. passing stdout or stderr fd).
Problem: If Nvim server fails to --listen and prints error before the
TUI enters alternate screen, the error isn't visible.
Solution: Forward server stderr using client side stderr handler instead
of having the server inherit client stderr file descriptor.
This does mean that `stderr_isatty` will be `false` in the server, but
that value doesn't matter in embedded mode.
Always pass `stdin_fd` to embedded server to avoid a hang when reading
from stdin when it's a TTY (not sure why one wants to do that, perhaps
by mistake), because if `stdin_fd` isn't passed, the server will try to
use stderr as stdin.
Example test failure on CI:
FAILED test/functional/terminal/tui_spec.lua @ 41: TUI exit status 1 and error message with server --listen error #34365
test/functional/terminal\tui_spec.lua:55: Failed to match any screen lines.
Expected (anywhere): "nvim%.exe: Failed to %-%-listen: address already in use:"
Actual:
|{114:nvim.exe -h"} |
| |
|[Process exited 1]^ |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
|{5:-- TERMINAL --} |
Snapshot:
screen:expect([[
{114:nvim.exe -h"} |
|
[Process exited 1]^ |
|*13
{5:-- TERMINAL --} |
]])
stack traceback:
test\functional\ui\screen.lua:909: in function '_wait'
test\functional\ui\screen.lua:537: in function 'expect'
test/functional/terminal\tui_spec.lua:55: in function <test/functional/terminal\tui_spec.lua:41>
In this case, it appears that the client entered alternate screen in the
middle of the server's print_mainerr().
fix: allocate hidden console for detached server
Starting the server with UV_PROCESS_DETACHED results in DETACHED_PROCESS, leaving the child without a console. Without a console:
CONIN$ / CONOUT$ cannot resolve, causing channel_from_stdio to fail.
ConPTY cannot attach, breaking :terminal.
This patch allocates a hidden console via AllocConsole() when the server has none, restoring working stdio and enabling ConPTY.
Also updates os_set_cloexec to clear HANDLE_FLAG_INHERIT on the RPC pipe
handles, matching the Unix F_DUPFD_CLOEXEC behavior.
Problem:
Automatic background detection sets the background option too late,
which loads colorschemes twice and causes problems when the user's
terminal background doesn't match the default (#32109, #36211, #36416).
Solution:
Use a DA1 query to determine whether the TTY supports OSC 11. Wait for
background detection and setting to complete before processing user
config.
Note: To preserve the existing behavior as much as possible, this triggers
OptionSet manually on VimEnter (since it won't trigger automatically if
we set bg during startup). However, I'm unsure if this behavior is
truly desired given that the documentation says OptionSet is triggered
"After setting an option (except during |startup|)."
Also fixes flickering issue #28667. To check for flickering:
nvim --clean --cmd "set termguicolors" --cmd "echo \"foo\"" --cmd "sleep 10"
On master, this gives me a black screen for 10 seconds, but on this
branch, the background is dark or light depending on the terminal
background (since the option is now set during startup rather than after
VimEnter).
Problem: When nvim is started with "-s -" (read from stdin), ":restart"
drops the "-" argument but keeps "-s", leaving an orphaned "-s" in the
restarted server's argv, causing a crash.
Solution: Drop "-s" and its scriptfile argument when copying v:argv for
the restarted server. Like "-- [files…]", the scriptfile is a one-shot
startup input that should not be replayed on restart
When using feed_data() with hidden buffer, terminal refresh may arrive
during may_restore_curbuf(), causing "last cursor" positions to change.
Get the two "last cursor" positions in the same RPC call.
Adding input doesn't help here, as the important part here is processing
main_loop.events, not running the main loop. Instead, use a non-fast API
request to ensure that previously queued events are processed.
This is a better way to prevent parallel tests from interfering with
each other, as there are many ways files can be created and deleted in
tests, so enforcing different file names is hard.
Using $TMPDIR can also work in most cases, but 'backipskip' etc. have
special defaults for $TMPDIR.
Symlink runtime/, src/, test/ and README.md to Xtest_xdg dir to make
tests more convenient (and symlinking test/ is required for busted).
Also, use README.md instead of test/README.md in the Ex mode inccommand
test, as test/README.md no longer contains 'N' char.
Problem:
When MsgArea highlight is changed, the next message may flash
and disappear because msg_start() renders with stale highlight attributes.
msg_puts_len() uses HL_ATTR(HLF_MSG) to render message text, which
happens before update_screen() calls highlight_changed().
So the message is rendered with outdated attrs.
Solution:
Call highlight_changed() in msg_start().
When the terminal process is suspended, putting cursor at bottom-left
hints that pressing a key will change the suspended state. However, when
returning to Normal mode, the user is more likely to interact with the
actual terminal output (e.g. copying it), so it's better to put cursor
at the old position which should be closer to the output.
Also, using is_focused() to check for mode is confusing. Just check
`State & MODE_TERMINAL` instead.
Problem:
- `:restart <cmd>` prepends `-c <cmd>` before the original `-c` args (if
any). So the original `-c` args may "override" it, which is
surprising.
- Confusing logic: `v:argv` is partially prepared in `ex_docmd.c`, and
then later `ui.c` skips other parts of it.
Current behavior is nonsense, for example this sequence:
:restart echo "Hello"
:restart +qall echo "Hello" | echo "World"
results in this v:argv:
[
'nvim'
'-c'
'echo "Hello" | echo "World"'
'--embed'
'-c'
'echo "Hello"'
...
]
Whereas after this commit, v:argv is:
[
'nvim'
'--embed'
...
'-c'
'echo "Hello" | echo "World"'
]
Solution:
- Append `-c <cmd>` at the _end_ of `v:argv`, not the start.
- Use a dummy placeholder `+:::` to mark where the "restart command"
appears in `v:argv`.
- Do all `v:argv` preparation in `ex_docmd.c`. This simplifies `ui.c`.
- Drop `-- [files…]` from `v:argv` since it is probably more annoying
than useful. (Users can use sessions to restore files on restart.)
Problem: Terminal buffers are not refreshed when processing keys that
trigger partial mappings.
Solution: Process due terminal refreshes before redrawing.
Problem: Terminal doesn't detect if the PTY process is suspended or
offer a convenient way for the user to resume the process.
Solution: Detect suspended PTY process on SIGCHLD and show virtual text
"[Process suspended]" at the bottom-left. Resume the process
when the user presses a key.
Problem:
restart hangs when nvim was started with stdin input, "-" marker stays
in v:argv, causing the restarted instance to block reading from stdin.
Solution:
filter out the "-" argument when rebuilding v:argv during restart.
Stdin content is ephemeral and shouldn't be re-read after restart.
Problem: Terminal scrollback may be wrong when increasing height after
outputting lines with full scrollback.
Solution: Ensure enough number of scrollback lines have been deleted.
Problem: Changing terminal height immediately after outputting lines
may lead to wrong scrollback.
Solution: Insert pending scrollback lines before the old window height.
Problem: Wrong scrollback when passing a buffer with many lines to
nvim_open_term().
Solution: Delete all buffer lines before opening the terminal.
I tried to use buf_clear(), but it crashes inside deleted_lines_mark(),
so I'll just use deleted_lines_buf() for now. The behavior of marks can
be decided later.
Problem: heap use-after-free if OptionSet autocommands from setting a terminal's
&buftype wipe the buffer.
Solution: set b_locked during OptionSet for &buftype.
Problem: Terminal doesn't handle ED 3 (clear scrollback) properly.
Solution: Add vterm callback for sb_clear().
Also fix another problem that scrollback lines may be duplicated when
pushing to scrollback immediately after reducing window height, as can
be seen in the changes to test/functional/terminal/window_spec.lua.
Problem: TermResponse deferred due to blocked autocommands lacks "data" payload.
Also, it may not fire if a new v:termresponse reuses the same string address.
Solution: add it. Use the value of v:termresponse for "data.sequence". Replace
pointer comparisons with a flag.
The removal of "old_termresponse" comparisons is required to pass the test on
the CI, or locally for me when compiled in RelWithDebInfo.
Problem: if buf_free_all autocommands open a terminal, it will remain open after
the buffer is freed.
Solution: close terminals again later, this time while blocking autocommands.
Did consider terminal_open checking stuff like b_locked_split instead, but
that's set during BufHidden, etc., which doesn't mean the buffer's being wiped.
Problem: close_buffer autocmds may switch buffers at the last moment when
closing a window, causing terminal_check_size to prefer the size of a closed
window, or TabClosed to set an old <abuf>.
Solution: use the actual last buffer, similar to what TabClosed did before.
NOTE: If buffer was unloaded/deleted (not wiped), then TabClosed's <abuf> may
not use it. (w_buffer = NULL) Maybe odd, but it's how it worked before anyhow.
Relies on close_buffer reliably setting w_buffer to NULL if freed, otherwise
buf_valid is better. Only concern I see is if the window wasn't in the window
list after closing the buffer (close_buffer won't set it to NULL then), but then
win_close{_othertab} should've returned earlier.
Problem: terminal's size may not update after one of its windows close.
Solution: call terminal_check_size after closing a window.
Disable test for Windows, as for some reason it only shows a few lines...
Problem: windows may scroll horizontally upon resize using the old terminal
size, which may be unnecessary and cause the content to be partially out-of-view.
Solution: reset the horizontal scroll after resizing.
Problem:
Recombining composing chars in terminal doesn't work at right edge.
Solution:
Check for the case where printing the previous char didn't advance the
cursor. Reset at_phantom when returning to combine_pos.
Problem: 0x08 is treated as Backspace instead of Ctrl-H on Windows.
Solution: Treat 0x7f as Backspace if VT input is enabled, so that 0x08
is treated as Ctrl-H.
Ref: https://github.com/microsoft/terminal/issues/4949
Problem:
On Windows, writing to a pipe doesn't work if the pipe isn't connected
yet. This causes an RPC request to a session newly created by connect()
to hang, as it's waiting for a response to a request that never reaches
the server.
Solution:
Wait for uv.pipe_connect() callback to be called when using connect().
Problem: Wrong terminal scrollback if BufFile* autocommand drains PTY
output but doesn't process the pending refresh.
Solution: Refresh scrollback before refreshing screen in terminal_open()
if scrollback has been allocated.
Problem: Terminal loses output if a BufFilePre or BufFilePost autocmd
polls for events.
Solution: Rename the buffer after allocating the terminal instance. Also
fix buffer getting wrong name if BufFilePre uses NameBuff.
Usually, terminal_close() calls refresh_terminal(), which allocates the
scrollback buffer, and term_may_alloc_scrollback() in terminal_open()
won't dereference the buffer. However, refresh_terminal() is not called
during Nvim exit, in which case a heap-use-after-free may happen if
TermOpen wipes buffer. Check for non-NULL buf_handle to avoid that.
