Problem: When preparing the parameters for a code actions LSP request,
the code set `context.diagnostics` when processing the first LSP client,
and then reused those `context.diagnostics` for subsequent LSP clients.
This meant that the second and next LSP clients got diagnostics that
did not originate from them, and they did not get the diagnostics that
they sent.
Solution: Avoid setting `context.diagnostics` (which is referenced by
all clients). Instead, set `params.context.diagnostics` directly, which
is specific to a single client.
Fixes#30001
Caused by #29501
(cherry picked from commit 7031949be0)
Problem:
If there are errors in the last line of a buffer, something like `Gdk` or
`G2k3J` will produce an error (at least with `lua_ls`):
Error executing vim.schedule lua callback:
.../neovim/share/nvim/runtime/lua/vim/diagnostic.lua:1446: Invalid 'line': out of range
Solution:
Only set extmarks if the target buffer line still exists
(cherry picked from commit e5e81262af)
`buf.code_action` always included diagnostics on a given line from all
clients. Servers should only receive diagnostics they published, and in
the exact same format they sent it.
Should fix https://github.com/neovim/neovim/issues/29500
(cherry picked from commit 720b309c78)
Problem:
When there are multiple inlay hints present at the same position, they
should be rendered in the order they are received in the response from
LSP as per the LSP spec. Currently, this is not respected.
Solution:
Gather all hints for a given position, and then set it in a single
extmark call instead of multiple set_extmark calls. This leads to fewer
extmark calls and correct inlay hints being rendered.
**Problem:** A query file for something like `html_tags` will not be
given html node completion
**Solution:** Check for parser aliases before offering completions
Co-authored-by: Lewis Russell <me@lewisr.dev>
(cherry picked from commit 05dcda8f9b)
Problem:
Error when calling vim.treesitter.start() and vim.treesitter.stop() in
init.lua.
Solution:
Ensure syntaxset augroup exists after loading synload.vim.
(cherry picked from commit d413038b4f)
Problem: Treesitter highlighter clears the already populated highlight
state when performing spell checking while drawing a
smoothscrolled topline.
Solution: Save and restore the highlight state in the highlighter's
_on_spell_nav callback.
(cherry picked from commit da4e8dc5b0)
Problem:
When 'ignorecase' is set, the default keymap Q and Q would exit visual
mode.
This issue was raised in #28287 and a fix was applied in #28289.
However, `==` operator is subject to user `ignorecase` setting.
Solution:
Switching to `==#` operator would guarantee case sensitive comparison
between visual mode and linewise visual mode.
Co-authored-by: Kuanju Chen <kuanju.chen@mksinst.com>
(cherry picked from commit 61aabe0730)
fix(tui): move $COLORTERM check to _defaults.lua (#29197)
We currently check $COLORTERM in the TUI process to determine if the
terminal supports 24 bit color (truecolor). If $COLORTERM is "truecolor"
or "24bit" then we automatically assume that the terminal supports
truecolor, but if $COLORTERM is set to any other value we still query
the terminal.
The `rgb` flag of the UI struct is a boolean which only indicates
whether the UI supports truecolor, but does not have a 3rd state that we
can use to represent "we don't know if the UI supports truecolor". We
currently use `rgb=false` to represent this "we don't know" state, and
we use XTGETTCAP and DECRQSS queries to determine at runtime if the
terminal supports truecolor. However, if $COLORTERM is set to a value
besides "truecolor" or "24bit" (e.g. "256" or "16) that is a clear
indication that the terminal _does not_ support truecolor, so it is
incorrect to treat `rgb=false` as "we don't know" in that case.
Instead, in the TUI process we only check for the terminfo capabilities.
This must be done in the TUI process because we do not have access to
this information in the core Neovim process when `_defaults.lua` runs.
If the TUI cannot determine truecolor support from terminfo alone, we
set `rgb=false` to indicate "we don't know if the terminal supports
truecolor yet, keep checking". When we get to `_defaults.lua`, we can
then check $COLORTERM and only query the terminal if it is unset.
This means that users can explicitly opt out of truecolor determination
by setting `COLORTERM=256` (or similar) in their environment.
(cherry picked from commit d7651b27d5)
Co-authored-by: Gregory Anders <8965202+gpanders@users.noreply.github.com>
LspDetach is now triggered by the main on_detach callback that is added
when an LSP client is attached to a buffer. The semantic_tokens module
already includes a LspDetach handler that does the right thing. When the
LspDetach trigger was added to the main LSP on_detach, it created a race
condition in semantic tokens when a buffer was deleted that would
trigger both its own on_detach and the LspDetach handlers. If the former
came last, an error was thrown trying to delete a non-existent augroup
(destroy() was being called twice).
(cherry picked from commit 43581011e4)
Co-authored-by: jdrouhard <john@drouhard.dev>
Problem: When an lsp client is stopped, the client will
only clear the diagnostics for the attached buffers but
not the unattached buffers.
Solution: Reset the diagnostics for the whole namespace rather than
for only the attached buffers.
(cherry picked from commit 025c874415)
Co-authored-by: crwebb85 <51029315+crwebb85@users.noreply.github.com>
Problem: if on_lines is called before the LSP is initialized, the buffer
is detached.
Solution: check for uninitialized clients before detaching.
(cherry picked from commit 292365fa1b)
Co-authored-by: Ilia Choly <ilia.choly@gmail.com>
fix(lsp): handle nil root_dir in health check (#29007)
The root directory could show up as something like:
Root directory: ~/path/to/cwd/v:null
Despite being `nil`
(cherry picked from commit f03b1622ad)
Co-authored-by: Mathias Fußenegger <mfussenegger@users.noreply.github.com>
feat(lsp): update LSP healthcheck format (#28980)
This is mostly an aesthetic change, although there are a few new pieces
of information included. Originally I wanted to investigate including
server capabilities in the healthcheck, but until we have the ability to
fold/unfold text in health checks that would be too much information.
(cherry picked from commit 5d26934c7c)
Co-authored-by: Gregory Anders <8965202+gpanders@users.noreply.github.com>
The new default SwapExists autocommand displays warning text (W325) but
does not use the WarningMsg highlight group as other warnings do. Use
the WARN log level when displaying this warning.
(cherry picked from commit e71713ba2b)
fix(comment): fall back to using trimmed comment markers (#28938)
Problem: Currently comment detection, addition, and removal are done
by matching 'commentstring' exactly. This has the downside when users
want to add comment markers with space (like with `-- %s`
commentstring) but also be able to uncomment lines that do not contain
space (like `--aaa`).
Solution: Use the following approach:
- Line is commented if it matches 'commentstring' with trimmed parts.
- Adding comment is 100% relying on 'commentstring' parts (as is now).
- Removing comment is first trying exact 'commentstring' parts with
fallback on trying its trimmed parts.
(cherry picked from commit 0a2218f965)
Co-authored-by: Evgeni Chasnovski <evgeni.chasnovski@gmail.com>
Problem: The changetracking state can de-sync when reloading a buffer
with more than one LSP client attached.
Solution: Fully detach all clients from the buffer to force buf_state to
be re-created.
(cherry picked from commit 879d17ea8d)
Co-authored-by: Ilia Choly <ilia.choly@gmail.com>
Problem: layout i.e. whitespace that is part of codelenses is currently
displayed as weird symbols and large amounts of spaces
Solution: replace all consecutive whitespace symbols with a single space
character when trying to display codelenses as virtual text
(cherry picked from commit d9a2acdab3)
Co-authored-by: Mango The Fourth <40720523+MangoIV@users.noreply.github.com>
Problem:
vim.fs.normalize() normalizes too much vim.loader and is slow.
Solution:
Make it faster by doing less. This reduces the times spent in
vim.fs.normalize in vim.loader from ~13ms -> 1-2ms.
Numbers from a relative benchmark:
- Skipping `vim.validate()`: 285ms -> 230ms
- Skipping `path_resolve_dot()`: 285ms -> 60ms
- Skipping `double_slash`: 60ms -> 35ms
