latest update to nvim-cmp causes stack overflow crash #12
Comments
|
I can't replicate it. |
|
I have this in my # Disable completion from the language server
options(languageserver.server_capabilities =
list(completionProvider = FALSE,
completionItemResolve = FALSE)) |
|
I have the same. I will investigate. |
|
I have found that the overflow occurs in ---Copy table
---@generic T
---@param tbl T
---@return T
misc.copy = function(tbl)
if type(tbl) ~= 'table' then
return tbl
end
if islist(tbl) then
local copy = {}
for i, value in ipairs(tbl) do
copy[i] = misc.copy(value)
end
return copy
end
local copy = {}
for key, value in pairs(tbl) do
copy[key] = misc.copy(value)
end
return copy
endThis loops infinitely with this table: I am not sure what is going on. All other lsp (python, etc.) still work fine. |
|
More investigation. At some point nvim-cmp receive this from cmp_r |
|
I found the problematic commit: using this fixes the issue for now "hrsh7th/nvim-cmp",
commit = "ae644feb7b67bf1ce4260c231d1d4300b19c6f30", |
|
for k, v in pairs(completion_item) do
self.completion_item[k] = v or self.completion_item[k]
endI have no familiarity with Anyway, I can't replicate the bug on my side. |
|
I fixed two bugs in the C code in the last commit. After starting R and opening the Object Browser, the bugs could be replicated with the following code: ddd <- data.frame(aa = 1, bb = 2)
rm(ddd)
x <- 1Bug 1: The Object Browser wasn't updated after Bug 2: Invalid string received by Maybe the second bug was causing the stack overflow crash in |
|
ah nice, i am actively trying to debug this for the past hour. Have you created a branch? |
|
No luck, still having this when I hit enter on a completion selection I am trying to make a reproducible example with this https://github.com/hrsh7th/nvim-cmp/blob/main/utils/vimrc.vim |
|
I am able to reproduce with this if has('vim_starting')
set encoding=utf-8
endif
scriptencoding utf-8
if &compatible
set nocompatible
endif
let s:plug_dir = expand('/tmp/plugged/vim-plug')
if !filereadable(s:plug_dir .. '/plug.vim')
execute printf('!curl -fLo %s/autoload/plug.vim --create-dirs https://raw.githubusercontent.com/junegunn/vim-plug/master/plug.vim', s:plug_dir)
endif
execute 'set runtimepath+=' . s:plug_dir
call plug#begin(s:plug_dir)
" Plugin list
Plug 'hrsh7th/nvim-cmp'
Plug 'hrsh7th/cmp-buffer'
Plug 'hrsh7th/cmp-nvim-lsp'
Plug 'hrsh7th/vim-vsnip'
Plug 'neovim/nvim-lspconfig'
Plug 'R-nvim/cmp-r'
Plug 'R-nvim/R.nvim'
Plug 'nvim-treesitter/nvim-treesitter', {'do': ':TSUpdate'}
call plug#end()
PlugInstall | quit
" Setup nvim-cmp
lua << EOF
local cmp = require("cmp")
cmp.setup {
snippet = {
expand = function(args)
vim.fn["vsnip#anonymous"](args.body)
end,
},
mapping = cmp.mapping.preset.insert({
["<C-n>"] = cmp.mapping.select_next_item(),
["<C-p>"] = cmp.mapping.select_prev_item(),
["<C-d>"] = cmp.mapping.scroll_docs(-4),
["<C-f>"] = cmp.mapping.scroll_docs(4),
["<C-Space>"] = cmp.mapping.complete({}),
["<CR>"] = cmp.mapping.confirm({
behavior = cmp.ConfirmBehavior.insert,
select = false,
}),
["<Tab>"] = cmp.mapping(function(fallback)
if cmp.visible() then
cmp.select_next_item()
-- elseif luasnip.expand_or_locally_jumpable() then
-- luasnip.expand_or_jump()
else
fallback()
end
end, { "i", "s" }),
["<S-Tab>"] = cmp.mapping(function(fallback)
if cmp.visible() then
cmp.select_prev_item()
-- elseif luasnip.locally_jumpable(-1) then
-- luasnip.jump(-1)
else
fallback()
end
end, { "i", "s" }),
}),
sources = cmp.config.sources({
{ name = "nvim_lsp" },
{ name = "buffer" },
{ name = "cmp_r" },
})
}
-- Configure cmp-r
require("cmp_r").setup({})
-- require("R.nvim").setup({})
-- Set up LSP with cmp capabilities
local capabilities = require('cmp_nvim_lsp').default_capabilities()
EOF
" Setup Treesitter
lua << EOF
require'nvim-treesitter.configs'.setup {
ensure_installed = { "markdown", "markdown_inline", "r", "rnoweb", "yaml", "vim"},
highlight = {
enable = true, -- Enable syntax highlighting
},
}
EOF
Then open a file like this |
|
Still no bug on my side with your |
|
No luck either. With my minimal example, do you get any completion at all? And no crash? |
|
Yes, the completion works normally, and there is no crash. |
|
Very strange. This is starting nvim with nothing else. Maybe my nvim version? I will explore this. |
|
I tried both stable and nightly and same issue. |
|
A similar bug was already reported, but not reproducible hrsh7th/nvim-cmp#1347 |
|
I am updating lua, just in case, I am out of ideas. |
|
As far as I know, bugs such "stack overflow" or "buffer overflow" should only happen in languages with low-level access to memory management, such as C or C++. Other programming languages (Java, R, Python, Lua, etc...) protect their users from these bugs (they don't give users the power to create them). Can you replicate the bug on a different machine? |
Yes. Both on at home and at job. Both on linux. |
Makes sense. The issue is that the table received here #12 (comment) seems to contain indefinite number of entries and the copy function eventually crashes. |
|
More info. The error originate from the call to local completion_item = misc.copy(e.completion_item)If I add a print statement print("confirm", vim.inspect(e.completion_item))
local completion_item = misc.copy(e.completion_item)I am getting this output (when I select {
cls = "(",
documentation = {
kind = "markdown",
value = 'function stats::lm\n\n**Fitting Linear Models**\n\n`lm` is used to fit linear models, including multivariate o
nes. It can be used to carry out regression, single stratum analysis of variance and\nanalysis of covariance (although `aov
` may provide a more convenient interface for these).\n\n---\n```r\nlm(formula, data, subset, weights, na.action, method =
"qr", model = TRUE, x = FALSE, y = FALSE, qr = TRUE, singular.ok = TRUE, contrasts = NULL,\n offset, ...)\n```\n'
},
env = "stats",
items = { <table 1> },
kind = 3,
label = "lm",
sortText = "9",
textEdit = {
newText = "lm",
range = {
["end"] = {
character = 1,
line = 0
},
start = {
character = 0,
line = 0
}
}
}
} This seems ok, the received table is only of length 1. misc.copy = function(tbl)
if type(tbl) ~= "table" then
return tbl
end
if islist(tbl) then
local copy = {}
-- print the length of the table
print("Reveived table in misc.copy\n" .. vim.inspect(tbl))
for i, value in ipairs(tbl) do
copy[i] = misc.copy(value)
end
return copy
end
local copy = {}
for key, value in pairs(tbl) do
copy[key] = misc.copy(value)
end
return copy
endIf I print the content of the reveived Reveived table in misc.copy
<1>{ {
cls = "(",
documentation = {
kind = "markdown",
value = 'function stats::lm\n\n**Fitting Linear Models**\n\n`lm` is used to fit linear models, including multivariate
ones. It can be used to carry out regression, single stratum analysis of variance and\nanalysis of covariance (although `a
ov` may provide a more convenient interface for these).\n\n---\n```r\nlm(formula, data, subset, weights, na.action, method
= "qr", model = TRUE, x = FALSE, y = FALSE, qr = TRUE, singular.ok = TRUE, contrasts = NULL,\n offset, ...)\n```\n'
},
env = "stats",
items = <table 1>,
kind = 3,
label = "lm",
sortText = "9",
textEdit = {
newText = "lm",
range = {
["end"] = {
character = 2,
line = 0
},
start = {
character = 0,
line = 0
}
}
}
} }
Reveived table in misc.copy
<1>{ {
cls = "(",
documentation = {
kind = "markdown",
value = 'function stats::lm\n\n**Fitting Linear Models**\n\n`lm` is used to fit linear models, including multivariate
ones. It can be used to carry out regression, single stratum analysis of variance and\nanalysis of covariance (although `a
ov` may provide a more convenient interface for these).\n\n---\n```r\nlm(formula, data, subset, weights, na.action, method
= "qr", model = TRUE, x = FALSE, y = FALSE, qr = TRUE, singular.ok = TRUE, contrasts = NULL,\n offset, ...)\n```\n'
},
env = "stats",
items = <table 1>,
kind = 3,
label = "lm",
sortText = "9",
textEdit = {
newText = "lm",
range = {
["end"] = {
character = 2,
line = 0
},
start = {
character = 0,
line = 0
....
...
... |
|
I can't imagine the infinite loop coming from the sorting = {
comparators = {
cmp.config.compare.order,
},
}, |
|
I replaced my config with yours and still the same (but now the order is different). Added 2 print statements here and I never receive the "Done copy" message. feedkeys.call("", "n", function()
local ctx = context.new()
print("confirm", vim.inspect(e.completion_item))
local completion_item = misc.copy(e.completion_item)
print("Done copy")
if not completion_item.textEdit then
completion_item.textEdit = {}
local insertText = completion_item.insertText
if misc.empty(insertText) then
insertText = nil
end
completion_item.textEdit.newText = insertText or completion_item.word or completion_item.label
end
local behavior = option.behavior or config.get().confirmation.default_behavior
if behavior == types.cmp.ConfirmBehavior.Replace then
completion_item.textEdit.range = e.replace_range
else
completion_item.textEdit.range = e.insert_range
end
local diff_before = math.max(0, e.context.cursor.col - (completion_item.textEdit.range.start.character + 1))
local diff_after = math.max(0, (completion_item.textEdit.range["end"].character + 1) - e.context.cursor.col)
local new_text = completion_item.textEdit.newText
completion_item.textEdit.range.start.line = ctx.cursor.line
completion_item.textEdit.range.start.character = (ctx.cursor.col - 1) - diff_before
completion_item.textEdit.range["end"].line = ctx.cursor.line
completion_item.textEdit.range["end"].character = (ctx.cursor.col - 1) + diff_after
if api.is_insert_mode() then
if false then
--To use complex expansion debug.
vim.print({ -- luacheck: ignore
item = e.completion_item,
diff_before = diff_before,
diff_after = diff_after,
new_text = new_text,
text_edit_new_text = completion_item.textEdit.newText,
range_start = completion_item.textEdit.range.start.character,
range_end = completion_item.textEdit.range["end"].character,
original_range_start = e.completion_item.textEdit.range.start.character,
original_range_end = e.completion_item.textEdit.range["end"].character,
cursor_line = ctx.cursor_line,
cursor_col0 = ctx.cursor.col - 1,
})
end
local is_snippet = completion_item.insertTextFormat == types.lsp.InsertTextFormat.Snippet
if is_snippet then
completion_item.textEdit.newText = ""
end
vim.lsp.util.apply_text_edits({ completion_item.textEdit }, ctx.bufnr, "utf-8")
local texts = vim.split(completion_item.textEdit.newText, "\n")
vim.api.nvim_win_set_cursor(0, {
completion_item.textEdit.range.start.line + #texts,
(#texts == 1 and (completion_item.textEdit.range.start.character + #texts[1]) or #texts[#texts]),
})
if is_snippet then
config.get().snippet.expand({
body = new_text,
insert_text_mode = completion_item.insertTextMode,
})
end
else
local keys = {}
table.insert(
keys,
keymap.backspace(
ctx.cursor_line:sub(completion_item.textEdit.range.start.character + 1, ctx.cursor.col - 1)
)
)
table.insert(
keys,
keymap.delete(ctx.cursor_line:sub(ctx.cursor.col, completion_item.textEdit.range["end"].character))
)
table.insert(keys, new_text)
feedkeys.call(table.concat(keys, ""), "in")
end
end)Ccing @yioneko and @hrsh7th just in case they have some ideas. |
|
Maybe this was caused by the bug fixed in the last R.nvim commit... |
|
No luck. Maybe I can close this if it is only me with this issue. |
|
I have added that to my init.lua vim.cmd("set verbosefile=~/.config/nvim/nvimlog")
vim.cmd("set verbose=3")I am getting this information on crash: |
|
We can't the issue before it's fixed. |
Not yet. I made my own for eventually open an issue on |
|
Same issue with the minimal config.
|
|
@PMassicotte you are not alone. I have the same bug and I use your workaround for now. |
|
You can try making these changes: diff --git a/lua/cmp_r/init.lua b/lua/cmp_r/init.lua
index 9670b90..8b8a770 100644
--- a/lua/cmp_r/init.lua
+++ b/lua/cmp_r/init.lua
@@ -531,6 +531,7 @@ end
---the data to rnvimserver which calls back this function.
---@param txt string The text almost ready to be displayed.
source.resolve_cb = function(txt)
+ vim.fn.writefile({ "", vim.inspect(txt) }, "/tmp/cmp_r_resolve_log", "a")
local s = fix_doc(txt)
if last_compl_item.def then
s = last_compl_item.label .. fix_doc(last_compl_item.def) .. "\n---\n" .. s
@@ -546,6 +547,7 @@ end
---@param compl table The completion data.
source.complete_cb = function(cid, compl)
if cid ~= compl_id then return nil end
+ vim.fn.writefile({ "", vim.inspect(compl) }, "/tmp/cmp_r_complete_log", "a")
local resp = {}
for _, v in pairs(compl) doThen, look at the log files and try to find something strange. To follow the log files as more lines are appended: tail -f /tmp/cmp_r_resolve_log |
|
When I try to complete with selecting |
|
After replacing {
{ cls = "(", env = "nvimcom", label = "etags2ctags" },
{ cls = "(", env = "nvimcom", label = "nvim.interlace.rmd" },
{ cls = "(", env = "nvimcom", label = "nvim.interlace.rnoweb" },
--- 2552 lines omitted
{ cls = "(", env = "base", label = "xtfrm.POSIXlt" },
{ cls = "(", env = "base", label = "xzfile" },
{ cls = "(", env = "base", label = "zapsmall" }
}No visible problem. After replacing function stats::lm
**Fitting Linear Models**
`lm` is used to fit linear models, including multivariate ones. It can be used to carry out regression, single stratum analysis of variance and
analysis of covariance (although `aov` may provide a more convenient interface for these).
---
```r
lm(formula, data, subset, weights, na.action, method = "qr", model = TRUE, x = FALSE, y = FALSE, qr = TRUE, singular.ok = TRUE, contrasts = NULL,
offset, ...)
```It's just a valid Markdown text. The |
|
In the Lua table from
My guess is that In the past, BSD users have fixed bugs in the C code of But if repeated labels were the problem, there should be no stack overflow when completing Conclusion: "I know that I know nothing". |
|
That seems to be a not so easy issue to fix. Thank you for looking at it. I will continue to investigate and report back if I find a solution. |
|
IMO, this issue should be fixed with this. |
|
Thanks you @hrsh7th. The last commit solved the issue. |
Thank you very much @hrsh7th , this solved it. Thank for the great plugin. |
I am getting a stack overflow when I am trying to use completion since the latest update of nvim-cmp (https://github.com/hrsh7th/nvim-cmp/commits/main/).
I am not sure how to debug that. Any ideas?
The text was updated successfully, but these errors were encountered: