Problem:
Since `foldclose` is async, it must wait for the request to return before actually executing, at which point the original window may no longer be valid.
Solution:
Check whether the window is valid before actually performing `foldclose`.
Problem:
The LSP client incorrectly checks for server capabilities when determining
support for self-mapped methods (e.g., 'shutdown'), which do not have
corresponding capabilities in the server's response. This leads to false
negatives when checking if such methods are supported.
This was handled correctly for dynamic registrations, but not for static.
Methods such as 'shutdown', do not have a related server capability and should
be assumed to be supported.
Solution:
Update the `supports_method` logic to always return true for self-mapped
methods.
Problem:
LSP spec allows response message to have a null request-id.
This may happen when for example client sends unparseable request.
https://github.com/microsoft/language-server-protocol/issues/196
Solution:
Guard the server response branches against id=vim.NIL (json null),
and handle error responses with null id by logging a warning
and dispatching on error.
Problem:
CI (ubuntu asan, ubuntu tsan, windows) reports `uv_loop_close()
hang?` from the two new null-id response tests. The leaked
handle is the server-side accepted TCP socket created inside
`server:listen` callback. The tests closed only the listener
but not the accepted socket, so libuv could not finish shutting
down the loop and each test session took ~2s extra to exit.
Solution:
Hoist the accepted socket to the outer `exec_lua` scope and
close it at teardown before closing the listener. The close
runs synchronously inside `exec_lua`, so the loop has time to
dispose the handle before the session exits.
* test(lsp): close accepted socket on read-loop exit/error
Match the precedent in the handler test ("handler can return
false as response") and the shared `_create_tcp_server` helper
in `test/functional/plugin/lsp/testutil.lua`: close the
accepted socket from inside the `create_read_loop` exit/error
callbacks. The teardown close added in the previous commit
remains as belt-and-suspenders, so the socket is disposed
whether the server goes away first or the client does.
Problem:
Use vim.lsp.util.apply_text_edits to re-apply the same textedit causes
an incorrect edit, because apply_text_edits silently modifies the
parameter.
Solution:
- Avoid changing `text_edit._index`.
- Document this fun feature.
Helped-by: Riley Bruins <ribru17@hotmail.com>
Helped-by: Yi Ming <ofseed@foxmail.com>
Co-authored-by: Justin M. Keyes <justinkz@gmail.com>
Problem:
The fromRanges field of the result of callHierarchy/outgoingCalls is
documented as being relative to the caller. Using
vim.lsp.buf.outgoing_calls() opened the qflist with an entry with the
callee's filename, but the caller's line number.
Solution:
Open the qflist with the callers file (the bufnr from the request),
rather than the callees (the uri from the resulting CallHierarchyItem)
Problem:
Cursor-position `vim.lsp.buf.code_action()` requests include all diagnostics on the current line, so unrelated same-line diagnostics affect the returned actions.
Solution:
Filter same-line diagnostics to the cursor position for cursor-position requests.
benchmark: https://gist.github.com/ofseed/6224529d77c016c36f7ab2f977059848
local rounds = tonumber(arg[1]) or 1000
local count = tonumber(arg[2]) or 1000
-- Load the table.clear function.
local clear = require("table.clear")
local function fill(t, n)
for i = 1, n do
t[i] = i
end
end
local function bench_reassign(n_rounds, n_items)
local t = {}
local start = os.clock()
for _ = 1, n_rounds do
t = {}
collectgarbage("collect")
fill(t, n_items)
end
return os.clock() - start
end
local function bench_reassign_no_gc(n_rounds, n_items)
local t = {}
local start = os.clock()
for _ = 1, n_rounds do
t = {}
fill(t, n_items)
end
return os.clock() - start
end
local function bench_clear(n_rounds, n_items)
local t = {}
local start = os.clock()
for _ = 1, n_rounds do
clear(t)
fill(t, n_items)
end
return os.clock() - start
end
-- Warm up LuaJIT before the real benchmark.
do
local t = {}
for _ = 1, 2000 do
clear(t)
fill(t, count)
end
end
collectgarbage("collect")
local reassign_time = bench_reassign(rounds, count)
collectgarbage("collect")
local reassign_no_gc_time = bench_reassign_no_gc(rounds, count)
collectgarbage("collect")
local clear_time = bench_clear(rounds, count)
print(string.format("rounds=%d count=%d", rounds, count))
print(string.format("t = {} + GC : %.6f s", reassign_time))
print(string.format("t = {} : %.6f s", reassign_no_gc_time))
print(string.format("table.clear : %.6f s", clear_time))
print(string.format("vs + GC : %.2fx", reassign_time / clear_time))
print(string.format("vs no GC : %.2fx", reassign_no_gc_time / clear_time))
benchmark result:
rounds=1000 count=1000
t = {} + GC : 0.022469 s
t = {} : 0.002570 s
table.clear : 0.000387 s
vs + GC : 58.06x
vs no GC : 6.64x
`count` is how many items the table has, and `round` is how many rounds we fill
the table, clear, and then refill it. `table = {}` is clear the table by
resigning a new empty one, because this script does not run persistently like
nvim so GC is not triggered, so I added another extreme control group that
manually triggers GC.
Problem:
Codelens virtual lines remain on stale rows after an external file
change and buffer reload.
Solution:
Clear codelens extmarks and cached row/version state in `on_reload`
before requesting fresh code lenses.
Problem: completionItem/resolve response's `detail` field is silently
dropped. Only `documentation` is shown in the popup.
Solution: Prepend `detail` as a fenced code block before `documentation`
in the info popup, skipping if documentation already contains it.
Problem:
After on_refresh() sends a textDocument/codeLens request, the buffer may
be deleted before the response arrives. The response callback then tries
to redraw that deleted buffer and raises Invalid buffer id error.
Solution:
Check buffer validity before redrawing.
AI-assisted: Codex
Co-authored-by: Yi Ming <ofseed@foxmail.com>
* fix(api): allow silencing "Too many highlight groups" error
Problem: Using Lua's `vim.api.nvim_set_hl(0, 'New', {...})` can fail if
there are too many existing highlight groups. However, this error can
not be silenced with `pcall`.
Solution: Make it possible to silence in `nvim_set_hl` and
`nvim_get_hl_id_by_name`.
* fix(lsp): limit number of groups created by `document_color()`
Problem: A file can contain many string colors that would be highlighted
by an LSP server. If this number crosses 19999 (maximum number of
allowed highlight groups), there are general issues with creating
other highlight groups, which can break functionality outside of
`vim.lsp.document_color`.
Solution: Limit number of highlight groups that are created by
`vim.lsp.document_color` to 10000 (half of allowed maximum).
This is not a 100% solution (since there can exist more than 10000
other highlight groups), but explicitly checking number of groups is
slow and 10000 should (hopefully) be enough for most use cases.
Problem:
Difficult for us to provide default handlers for functions like
`vim.lsp.buf.definition`. When users wanted to fine-tune the default behavior,
they don't know how.
Solution:
- Document an example providing `on_list` boilerplate to make it easier for
users to modify and override.
- Also, considering that the parameters of the previous
`on_list`(`vim.lsp.ListOpts.OnList`) are compatible with the parameters of
`setqflist`, remove that custom type in favor of passing
`vim.fn.setqflist.what`.
Problem:
Naming conventions are not automatically checked.
Solution:
Add a check to the doc generator. Eventually we should extract this
somehow, but that will require refactoring the doc generator...
Note: this also checks non-public functions, basically anything that
passes through `gen_eval_files.lua` and `gen_vimdoc.lua`. And that's
a good thing.
Problem
Some variables use the wrong type (ClientToServer instead of ServerToClient)
and some use vaguer types that could be more strict.
Solution
Use the correct types.
Problem:
The window opened by `vim.lsp.util.open_floating_preview()`
allows its buffer to be switched. Presumably that only happens
by accident and is disorienting.
Solution:
Set 'winfixbuf' in the open_floating_preview() window.
continues d0af4cd909.
This commit renames positional parameters. This is only "cosmetic", but
is intended to make it extra clear which name is preferred, since people
often copy existing code despite the guidelines in `:help dev-naming`.
Problem:
To support `collapsedText`, which allows the LSP server to determine the
content of the foldtext, we provided `vim.lsp.foldtext()`. However, such
content does not have highlighting.
Solution
Treat the filetype of `collapsedText` as the filetype of the corresponding
buffer and use tree-sitter to highlight it.
Problem: vim.lsp.util.show_document insert mode is unable
to set the cursor after the target character position if the target character
is at end of line.
Solution: Move cursor after the target character (in append position)
in this case.
Problem: _get_and_set_name edits the name for the whole group,
thus only one client per group gets the didOpen message.
Solution: move the logic to _changetracking and loop over every
client per group.
Problem: hover/signature callback lacked consistency checks, so slow LSP servers
could open a float after the cursor had already moved away.
Solution: guard the callback with buf validity, buf version, and cursor
position checks before opening the float. Also fix table capacity calculation.
Problem:
When a new textDocument/codeLens response arrives with unresolved lenses,
on_win clears the existing codelens row before codeLens/resolve
completes. This causes the displayed codelens text to flicker while
typing.
Solution:
Keep the current virtual lines if any of the refreshed lenses are still
unresolved. Clear the old virtual lines only when the line no longer has
lenses or all its lenses are resolved.
A trade-off is that the user may temporarily see outdated codelenses.
However, that's preferable to spamming updates on every refresh.
AI-assisted: Codex
Problem:
`vim.lsp.buf.definition`/`vim.lsp.buf.declaration` use the same underlying code
via `get_locations`, whereas `vim.lsp.buf.reference` does not. This is because
`buf.reference` does not perform a jump when there is only one item.
Solution:
In #38510, I simplified the jump logic using `:cfirst`, so they can now share
code more easily. Additionally, this PR enables `buf.definition` to display the
corresponding qflist name.
Problem:
`:help dev-name-common` states that "buf" should be used instead of
"buffer" but there are cases where buffer is mentioned in the lua API.
Solution:
- Rename occurrences of "buffer" to "buf" for consistency with the
documentation.
- Support (but deprecate) "buffer" for backwards compatibility.
Co-authored-by: Justin M. Keyes <justinkz@gmail.com>
Problem:
Since 2f6d1d3c88, `apply_text_edits`
unconditionally sets `buflisted=true`, causing spurious BufDelete events
if plugins restore the original 'buflisted' state on unlisted buffers:
65ef6cec1c/src/nvim/option.c (L2159-L2169)
Solution:
- Don't set 'buflisted' in `apply_text_edits`. Set it more narrowly, in
`apply_workspace_edit` where the semantics requires affected buffers
to be visible to the user.
- Also skip setting 'buflisted' if it would not be changed, to avoid
redundant `OptionSet` events.
Problem:
The current LSP diagnostic implementation can't differ between a pull
diagnostic with no identifier and a set of diagnostics provided via push
diagnostics.
"Anonymous pull providers" are expected by the protocol https://microsoft.github.io/language-server-protocol/specifications/lsp/3.17/specification/#diagnosticOptions
, depending on how the capability was registered:
- Dynamic registrations have an identifier.
- Static registrations will not.
Solution:
Restore the `is_pull` argument removed in
https://github.com/neovim/neovim/pull/37938, keeping the identifier of
pull diagnostic collections.
Problem:
`:checkhealth vim.lsp` validates configured filetypes against
`getcompletion('', 'filetype')`. This only reflects runtime support
files.
This causes false warnings in `:checkhealth vim.lsp` for configured
filetypes that are known to the Lua filetype registry, including
values added with `vim.filetype.add()` and built-in registry-only
filetypes.
Solution:
Build the healthcheck's known-filetype set from both
`getcompletion('', 'filetype')` and `vim.filetype.inspect()`.
Problem: `buf` is optional even though its needed to perform conversions
and the ordering of `(buf, row, col)` is not consistent.
Solution: make `buf` mandatory on `vim.range` and `vim.pos` and enforce
the `buf, row, col` ordering
Problem:
LSP error responses implicitly rely on a custom `__tostring` function
(`vim.lsp.rpc.format_rpc_error`) for formatting. This causes errors that are not
created via `vim.lsp.rpc.error` to behave inconsistently with those that are.
Furthermore, we usually use `log.error` to print these errors, which uses
`vim.inspect` under the hood, so the custom `__tostring` provides little
benefit.
This increases the difficulty of refactoring the code, as it tightly couples RPC
error handling with the LSP.
Solution:
Convert every potential `__tostring` call to an explicit one. Since we don't
describe this behavior in the documentation, this should not be a breaking
change.
Problem:
LSP jump operations such as `buf.definition`/`buf.type_definition` do
not follow the 'switchbuf' option. Instead their behavior is controlled
by `vim.lsp.LocationOpts.reuse_win`. When `reuse_win=true`, the effect
is very similar to `set switchbuf=useopen`.
Note that functions like `buf.definition` open the quickfix
window when there are multiple results, and jumping between quickfix
entries already follows 'switchbuf', so unifying the behavior is more
intuitive.
Solution:
Follow the 'switchbuf' option and drop `reuse_win`.
We can achieve this behavior by using :cfirst when the quickfix list has
only one item, rather than customizing the jump logic as before.
Problem:
The snippet preview is not being highlighted by treesitter for
completion items from servers which don't support
`completionItem/resolve` (like gopls). This was broken by #38428.
Solution:
Call `update_popup_window` after updating the completion item with the
snippet preview.
I've added assertions to the `selecting an item triggers
completionItem/resolve + (snippet) preview` test case which covers the
snippet preview being shown since no tests failed when I removed the
`nvim__complete_set` call which actually populates the preview on this
codepath.
Problem:
Using nested `vim.Pos` objects to represent each `vim.Range` object
requires 3 tables for each `vim.Range`, which may be undesirable in
performance critical code. Using key-value tables performs worse than
using array-like tables (lists).
Solution:
Use array-like indices for the internal fields of both `vim.Pos` and
`vim.Range` objects. Use a metatable to allow users to access them like
if they were key-value tables.
---
Problem:
The `vim.Pos` conversion interface for `extmark` indexing does not take
into account the difference in how a position on top of a newline is
represented in `vim.Pos` and `extmark`.
- `vim.Pos`: for a newline at the end of row `n`, `row` takes the value
`n + 1` and `col` takes the value `0`.
- `extmark`: for a newline at the end of for `n`, `row` takes the value
`n` and `col` takes the value `#row_text`.
Solution:
Handle this in the `extmark` interface.
---
Problem:
Not all `to_xxx` interfaces have wrapping objects like `to_lsp`.
Solution:
Return unwrapped values in `to_xxx` interfaces where it makes sense.
Accept unwrapped values in "from" interfaces where it makes sense.
---
Problem:
`start` and `end` positions have different semantics, so they can't be
compared. `vim.Range` relies on comparing the `end` and `start` of two
ranges to decide which one is greater, which doesn't work as expected
because this of the different semantics.
For example, for the ranges:
local a = {
start = { row = 0, col = 22, },
end_ = { row = 0, col = 24, },
}
local b = {
start = { row = 0, col = 17, },
end_ = { row = 0, col = 22, },
}
in this code:
local foo, bar = "foo", "bar"
-- |---||-|
-- b a
The range `b` is smaller than the range `a`, but the current
implementation compares `b._end` (`col = 22`) and `a.start` (`col = 22`)
and concludes that, since `b.col` is not smaller than `a.col`, `b`
should be greater than `a`.
Solution:
- Use a `to_inclusive_pos` to normalize end positions inside of
`vim.Range` whenever a comparison between a start and an end position
is necessary.
