fix(lsp): define autocmds for capabilities in new(), not on_attach()
Problem: Defining autocmds in on_attach() caused issues when multiple
clients provide the same capability for a buffer. Each attaching client
would "replace" the previously defined one since they are all identical.
Then the first one to detach clears them out and any remaining attached
clients would no longer trigger the autocmd for the capability. Further,
the semantic tokens module itself didn't quite work with multiple
clients since any LspNotify (from any client) would send a token request
to all attached clients with no differentiation and the debounce timer
was shared across all clients.
Solution: Always define the buffer-local autocmds in the capability's
`new()` function, and don't mess with the autocmds in on_attach or
on_detach. The capability framework itself will clear the autocmds when
the last client detaches. Also, refactor a bit of the semantic tokens
module so that the various methods take a specific client_id to perform
the work on, and split out timers so each client has its own.
Problem:
Legacy cmdwin set 'buflisted', but new one doesn't.
https://github.com/neovim/neovim/issues/40431#issuecomment-4811593353
'buflisted' useful for:
- "bufferline" style tablines, they usually show only listed buffers.
- some automatic actions can be conditioned on whether a buffer is
listed; 'buflisted' signals that the buffer is "important enough".
Solution:
Set 'buflisted'.
Problem:
After ctrl-f from the cmdline, the last 2 lines of cmdwin are redundant.
Solution:
In `open_cmdwin`, clear the live cmdline so that unwinding it (via
Ctrl_C) does not add it to history.
Problem:
Diagnostic tracking used a separate bufstates table and manual
LspDetach/LspNotify autocmd management via _enable()/_refresh(),
duplicating the lifecycle logic already provided by the Capability
framework. This caused inconsistencies in how client attach/detach and
buffer teardown were handled compared to other LSP features.
Solution:
Replace the ad-hoc bufstate tracking and _enable/_refresh pattern in
vim.lsp.diagnostic with a proper Diagnostics subclass of Capability.
This cleans up a few random places in the main lsp module and client
module that were poking the diagnostics. It also fixes some pre-existing
bugs and inconsistencies that were discovered:
- Refresh diagnostics immediately on attach instead of lazily by the
first didOpen/didChange notification
- Fix Capability.active lookup in M.enable() to key by it_bufnr instead
of the filter bufnr
- Set lsp defaults before calling the _text_document_did_open_handler in
Client:on_attach() so defaults are there before any lsp notification
occurs
- Log (and return early) on any error from a diagnostic request result
instead of only returning early for server cancelled errors
Problem:
cmd given as string[] always starts using Nvim's CWD, which is arbitrary.
Solution:
If cmd_cwd is not given, use root_dir as CWD.
BREAKING CHANGE: LSP commands given as string arrays now use `root_dir` as
the process working directory when `cmd_cwd` is unset.
Co-authored-by: Justin M. Keyes <justinkz@gmail.com>
Problem:
`:edit <dir>` and `nvim <dir>` currently rely on netrw to show local directory
contents.
Solution:
- Provide `filetype=directory`.
- Introduce dir.lua, a small plugin that provides directory listing, opening
items, parent navigation, and refresh.
- `netrw` remains available for `:Explore`, remote paths, archives, and file
operations. To continue
Co-authored-by: Justin M. Keyes <justinkz@gmail.com>
Problem:
Executing :messages while in cmdwin fails:
Error in "msg_history_show" UI event handler (ns=nvim.ui2):
Lua: …/_core/ui2/messages.lua:699: Invalid 'height': expected positive Integer
stack traceback:
[C]: in function 'nvim_win_set_config'
…/_core/ui2/messages.lua:699: in function 'set_pos'
…/_core/ui2/messages.lua:329: in function 'set_target_pos'
…/_core/ui2/messages.lua:389: in function 'show_msg'
…/_core/ui2/messages.lua:553: in function 'handler'
…/_core/ui2.lua:161: in function 'ui_callback'
…/_core/ui2.lua:210: in function <…/_core/ui2.lua:202>
The bug: when `texth.all` is small (e.g. 0 from a hidden pager whose new
content isn't laid out yet), or when the available height after
subtracting the cmdwin is small, `math.min(min, …)` can yield 0, and
`nvim_win_set_config` rejects `height=0`.
Solution:
Floor at 1.
Problem:
cmdwin (the `:q` cmdline buffer) has various limitations which require
special-casing all over the codebase.
Besides complicating the code, it also breaks async plugins if they try
to create buffers/windows after some work is done, if the user happens
to open cmdwin at the wrong the moment:
Lua callback: …/guh.nvim/lua/guh/util.lua:531:
E11: Invalid in command-line window; <CR> executes, CTRL-C quits
stack traceback:
[C]: in function 'nvim_buf_delete'
…/guh.nvim/lua/guh/util.lua:531: in function <…/guh.nvim/lua/guh/util.lua:526>
Solution:
Just say no to "inception". Reimplement cmdwin as a normal buffer+window.
All of the cmdwin contortions (in both core, and innocent plugins) exist
literally only to support "inception": recursive
cmdwin-in-cmdline-things, like `<c-r>=`, `/`, search-during-substitute,
`:input()`, etc. So we just won't support that (though I have
a potential plan for that later, which I call "modal parking lot").
The benefit is that plugins, and core, no longer have to care about
cmdwin.
BONUS:
- mouse-drag on vertical separators works (it only worked for
horizontal/statusline before)
- inccommand-in-cmdwin now works correctly, for free (thus don't need
#40077).
POTENTIAL FOLLOWUPS
- Drop `CHECK_CMDWIN` ("E11: Invalid in command-line window"), allow chaos.
- Unify `BUFLOCK_OK` / `LOCK_OK` ?
DESIGN:
- Eliminate lots of C globals, `EX_CMDWIN`, etc.
- `text_locked()` no longer reports true for cmdwin.
- cmdwin = a normal window with 'winfixbuf', 'bufhidden=wipe',
'buftype=nofile'. Invariants come from those options rather than
special cases throughout the codebase.
- `nv_record` for q:/q//q? calls Lua
`nlua_call_vimfn("vim._core.cmdwin", …)`. No `K_CMDWIN`
/ cmdline-reader detour.
- `cedit_key` (`c_CTRL-F`) schedules a deferred event that calls
`vim._core.cmdwin.open(type, content, pos)` and returns `Ctrl_C` so
the in-flight cmdline cancels. Reader state is not serialized; instead
the captured `(type, line, col)` is replayed via
`nvim_feedkeys(type..line.."<CR>", "nt", …)` after user confirms.
- On confirm/cancel: `<CR>` / `<C-C>` calls into Lua which closes the
window and re-feeds the cmdline.
BREAKING CHANGES:
- Expression-register cmdline (`<C-R>=` from insert-mode) no longer
supports cmdwin. Same applies to `input()` / `inputlist()` (already
covered by `text_locked`).
- Usage of cmdwin in macros/mappings will probably break (assuming they
ever worked).
Problem: The table `history` in treesitter/_select.lua stores references
to previously selected TSNodes, but a TSNode needs to keep its
whole TSTree alive in memory, which may be kept alive even
after closing the buffer to which this history corresponds.
Solution: Store just the bits of information from the TSNode we need to
go back in selection history (the range and ID), without
referencing the TSNode itself.
Problem: The function visual_select() in treesitter/_select.lua is
executing `:normal! v<Esc>` to ensure that `gv` later goes
back to character Visual mode, but doing so when mode() is
already `v` first switches back to the Normal mode, then
executes <Esc>, which causes a beep.
Solution: Check if the mode() is already `v` and avoid any switching in
that case.
Some servers register `workspace/didChangeWatchedFiles` watchers for URI
schemes that cannot be watched locally. Skipping the unsupported glob
and keep the rest of the registration batch active.
Problem: LspNotify autocmds were not being triggered for didChange
requests when being used during undo/redo (and possibly other) actions.
autocmds are blocked when calling on_lines() callbacks while doing the
undo/redo action.
Solution: Defer firing the autocmd until after the action is complete.
This is closer to what existed before, but now there's a check in the
deferred function to only fire the autocmd if the client is still
active and the buffer is still attached, if applicable.
Problem:
Terminal probes sent with `nvim_ui_send()` can reach more than one stdout TTY
UI. Probes with a known TTY UI owner should not accept `TermResponse`s from
unrelated UI channels.
Solution:
Thread the existing `chan` filter through owned terminal probes. This covers
startup/attach background detection, fallback truecolor detection, and
`vim.tty.query()` forwarding opts to `vim.tty.request()`.
Note: Not every terminal escape path is updated here. I only passed `chan` when
the caller already knows which TTY UI owns the probe. For example, this does
not include:
- (Followup) OSC 52 (system clipboard) detection. It needs to capture the
`UIEnter` channel. Adding `{ chan = ... }` only to the nested query would be
half a fix.
- OSC 52 _paste_ is left global because the provider callback does not have
a UI channel (paste is invoked without a ui channel/tty ui object).
Problem: Attach-time terminal probes cannot distinguish responses from
different attached UIs.
Solution: Identify the UI by RPC channel id in `TermResponse` and make
`vim.tty.request()` filter responses by channel.
Problem: Clearing the screen doesn't clear the "last" virtual text.
Dupe counter virtual text is not increased beyond 1.
Solution: Clear "last" virtual text when clearing the screen.
Restore assignment lost in a previous commit.
Problem: Terminal background and truecolor detection runs only at startup,
gated on a UI being attached. A headless server has no UI then, so
`'background'` and `'termguicolors'` are never detected and remote
UIs ignore the terminal's theme.
Solution: Also (re)detect on UIEnter. The most recently attached terminal
wins; an explicit user value is preserved.
Problem: Writing a message with a large number of newlines
(:echo "foo\n"->repeat(1000000)) takes longer than it has to
(since c973c7ae).
Solution: Accumulate newlines in a single API call when possible.
Problem: LspNotify never passed a buffer when executing the autocmds, so
buffer-local LspNotify autocmd subscriptions didn't have the correct buf
in the event metadata. It was also wrapped in a schedule() so the actual
autocmd was delayed until after the event loop.
This could result in the wrong buffer receiving the notification if
multiple LspNotify autocmds with buffer filters were added. Only the
"latest" one would actually receive non-buffer-filtered autocmds, not
the matching one. It also caused listeners to receive the notification
"out of sync" with when the notification is actually sent. If a buffer
is being deleted (which fires a textDocument/didClose notification), the
notification is scheduled and fired after the buffer is already gone.
Solution: For LSP notifications that pertain to a particular buffer, set
it when executing the LspNotify autocmds so the callback functions that
are filtered on that buffer will get the correct notifications and the
metadata buf field will be correct. Additionally, there is no need to
wrap the LspNotify callback in vim.schedule when it can be called inline
when the notification to the rpc server is fired.
This is tested by removing now-unnecessary autocmds from semantic tokens
(InsertEnter and BufWinEnter should no longer be necessary now that
requests are fired by LspNotify). Without this fix, simply modifying a
buffer doesn't actually trigger LspNotify correctly, and the test for
that fails.
Problem: When multiline semantic token support was introduced, the loop
that finds the end line for a particular token didn't sanitize the token
length sent back by the LSP server. If the server returned an overflowed
length (near uint32 max), neovim would burn cpu and loop for an
extremely long time while trying to find the "end line" represented by
the massively large token, causing neovim to seemingly hang.
Solution: Stop looping once the calculated end_line reaches the actual
last line of the buffer.
Fixes#36257
Problem:
- Current 'statuscolumn' click label caveat is restrictive.
- v:virtnum is not unique to a line if it has both above and
below virtual lines.
- 'statuscolumn' click handler may expect v:virt/lnum to be set.
Solution:
- Store per-row click definitions for the statuscolumn in a
(nested line/virt number) map.
- Implement strategy that gives each 'statuscolumn' row a unique
v:virtnum.
- Set v:virt/lnum when determining which line is clicked.
Problem: linked_editing_range was still doing most of the capability
boilerplate itself.
Solution: refactor it to make use of the common Capability framework for
handling enabling, disabling, etc.
Problem:
:Man does not syntax highlight codeblocks (injected language).
Analysis:
init_pager() swapped pcall(parse_ref, ref) return values order, so
vim.b.man_sect was set to the manpage name instead of the section
number, so C syntax highlighting did not load.
Solution:
Swap the calls.
Today there is a constraint that these arguments to the enable filter be
mutually exclusive, but I do not know why such a constraint exists (it
is perfectly reasonable to want to enable a capabilility for just one
buffer and just one client).
Problem:
an empty `{ isIncomplete = true, items = {} }` ends completion
instead of requerying.
Solution:
keep isIncomplete on empty lists and retrigger on keypress while incomplete.
Reset on <C-e> so it doesn't immediately re-query.
Problem: a font set via nvim_set_hl is lost or corrupted with font-only groups,
attribute combining, and update=true, and is dropped on any attr-table rebuild.
nvim__inspect_cell also frees the font name while a returned dict still borrows it.
Solution: register font-only groups, carry font through hl_combine_attr, inherit
it on update, and persist sg_font so a rebuild can restore it. Keep interned font
names across a rebuild instead of clearing them. Add the missing font field to
get_hl_info.
Co-authored-by: Ryan Patterson <cgamesplay@cgamesplay.com>
Problem:
'autoread' error if file goes missing.
E211: File "runtime/lua/nvim/autoread.lua" no longer available
vim.schedule callback: /usr/local/share/nvim/runtime/lua/nvim/autoread.lua:146:
Vim:E211: File "test/functional/options/autoread_spec.lua" no longer available
stack traceback:
[C]: in function 'checktime'
/usr/local/share/nvim/runtime/lua/nvim/autoread.lua:146:
in function </usr/local/share/nvim/runtime/lua/nvim/autoread.lua:138>
Solution:
Ignore E211 thrown by `:checktime`.
Problem: vim.diagnostic.Opts.VirtualLines.current_line and
vim.diagnostic.Opts.VirtualText.current_line cause redraw
on every CursorMoved event that makes the text jump
uncomfortably with rapid cursor movement.
Solution: `current_line` state is applied only on CursorHold,
while still being cleared on CursorMoved making it so that
until the cursor has stopped on a line, its diagnostic
looks like cursor is not on the line preventing flicker.
Problem:
This test is flaky since fff9897ce3 :
FAILED ...Xtest_xdg_terminal/test/functional/terminal/tui_spec.lua @ 4359:
TUI bg color does not trigger OptionSet from automatic background processing
Expected values to be equal.
Expected:
{ true, 0 }
Actual:
{ true, 1 }
stack traceback:
...Xtest_xdg_terminal/test/functional/terminal/tui_spec.lua:4380:
in function <...Xtest_xdg_terminal/test/functional/terminal/tui_spec.lua:4359>
`apply_optionset_autocmd_now` does not emit `OptionSet` during startup:
void apply_optionset_autocmd_now(...) {
// Don't do this while starting up, failure or recursively.
if (starting || errmsg != NULL || *get_vim_var_str(VV_OPTION_TYPE) != NUL) {
return;
}
...
}
but if OSC 11 response arrives AFTER VimEnter, then
`nvim_set_option_value('background',…)` call triggers `OptionSet`.
Solution:
Use `:noautocmd` when setting 'background'.
Per fff9897ce3, OSC11 is not intended to trigger OptionSet.
fix https://github.com/neovim/neovim/issues/40235
Problem: The refactor to use Capability left around some cruft and
semi-broken configuration for debounce.
Solution: Clean up now-unnecessary helper methods and simplify
deprecated ones to pass through to the non-deprecated ones. `debounce`
now defaults to 200 for all buffers but is overridable via the
deprecated start() method, which continues to take the max value
specified for any client attached to the buffer.
If we wish to expose changing the debounce in a non-deprecated way, we
will need to consider a "configuration" function, or even a bespoke
method to set the debounce time on the main metaclass (or provide
options to override for a particular buffer). General configuration of
specific LSP features is an as-of-yet unsolved problem.
Problem: filetype: tf files sometimes incorrectly recognized
(Christian Robinson)
Solution: Add support for g;filetype_tf variable to override detection,
re-write the filetype detection loop
closes: vim/vim#20510
Supported by AI
522a39a489
Co-authored-by: Christian Brabandt <cb@256bit.org>
Problem:
Any random ftplugin or other autocmd, can throw an error when
`:checktime` reloads a buffer. This causes a trace which makes it look
like an issue with `autoread.lua`.
vim.schedule callback: …/runtime/lua/nvim/autoread.lua:146:
FileType Autocommands for "*"..function <SNR>1_LoadFTPlugin[20] ..script
…/runtime/ftplugin/help.lua: Vim(runtime):E5113: Lua chunk:
…/runtime/lua/vim/treesitter.lua:216: Index out of bounds
stack traceback:
[C]: in function 'nvim_buf_get_text'
…/runtime/lua/vim/treesitter.lua:216: in function 'get_node_text'
…/runtime/lua/vim/treesitter/query.lua:558: in function 'handler'
…/runtime/lua/vim/treesitter/query.lua:843: in function '_match_predicates'
…/runtime/lua/vim/treesitter/query.lua:1082: in function '(for generator)'
…/runtime/ftplugin/help.lua:91: in function 'runnables'
…/runtime/ftplugin/help.lua:124: in main chunk
[C]: in function 'checktime'
…/runtime/lua/nvim/autoread.lua:146: in function <…/runtime/lua/nvim/autoread.lua:138>
Solution:
Use pcall() and surface the error via nvim_echo.
Problem:
Old 'autoread' only did `:checktime` on focus-change and shell (":!")
commands, and only for non-hidden buffers. Since 'autoread' is now
driven by OS filewatcher events, buffers are updated much more eagerly.
This should be surfaced to the user somehow, either via a carefully
placed notification, or a minimal UI indicator.
A "notification" would be noisy, unless it is conditional on specific
circumstances (e.g. when "many" buffers are updated).
Solution:
Use the existing 'busy' buffer-local option as a subtle hint about
activity.
Problem: The semantic token module is using its own debounce timer for
the buffer on_lines event. If its internal debounce is shorter than the
changetracking module's debounce, it's possible for semantic token
requests to fire for changed buffers before the textDocument/didChange
notification is sent to the server.
Solution: Trigger semantic token requests from the LspNotify autocmd
when the method is the didChange or didOpen notifications, which
enforces a strict happens-before relationship for the sync change
notification followed by a semantic token request.
Note: There is still an internal debounce mechanism in the semantic
token module to handle other debouncing needs specific to its
functionality, such as debouncing server refresh notifications and
handling WinScrolled events when using range requests.
Problem: Keys valid in CTRL-X mode are never mapped while insert
completion is active, so <C-N> and <C-P> cannot be remapped
for completion started by complete().
Solution: Do not disable mappings in CTRL_X_EVAL mode. In this mode a
mapping cannot interfere with selecting the completion
method, which is what the no-mapping rule exists for.
related: vim/vim#6440
related: vim/vim#16880
closes: vim/vim#20489076585e6ad
Co-authored-by: Thomas M Kehrenberg <tmke8@posteo.net>
Problem: Running `vim.filetype.match()` when current working directory
was removed from disk throws a `vim.fs.abspath` assertion error.
However, the matching might still be possible without trying to match
against full path (like if it is a known extension).
Solution: Safely compute absolute path and ignore it if it errors.
