Does the index that tabnine creates ever get stored to disk? If so in which directory?

Also does it place files anywhere other than ~/Library/Preferences/TabNine?

I have tested bingo language server instead of go-langserver.
But it does not work with TabNine.
https://github.com/saibing/bingo

@zxqfl Can you test it?

My configuration.

[language.go]
command = "bingo"
args = ["-mode", "stdio"]
install = [["go", "get", "-u", "github.com/saibing/bingo"]]

• It works with vim-lsp
• "go-langserver" works with TabNine
• bingo needs Go 1.11+

From TommyX12/company-tabnine#5.

How is this supposed to work?

~/TabNine/1.0.7/x86_64-unknown-linux-gnu $ chmod +x TabNine
~/TabNine/1.0.7/x86_64-unknown-linux-gnu $ ./TabNine
~/TabNine/1.0.7/x86_64-unknown-linux-gnu $

OS info:

 - system: `"x86_64-linux"`
 - host os: `Linux 4.14.87, NixOS, 18.03.133389.b551f89e256 (Impala)`
 - multi-user?: `yes`
 - sandbox: `no`
 - version: `nix-env (Nix) 2.0.4`
 - channels(matt): `"nixos-18.03.133236.8edf566ee40, unstable-19.03pre152634.218ce4de508"`
 - channels(root): `"nixos-18.03.133389.b551f89e256, unstable-19.03pre162390.86344b3a52b"`
 - nixpkgs: `/nix/var/nix/profiles/per-user/root/channels/nixos/nixpkgs/nixpkgs`

I use my text editor for both editing code and taking notes in .txt files. I'd love a feature where TabNine suggestions could be turned off when editing certain file extensions e.g. .txt or in specific files, e.g. notes.txt.

P.S. Thank you for the amazing product. Bought it right away.

root
|
|-- subproj1
|
|-- subproj2
    |
    |-- subsubproj1
    |
    |-- subsubproj2

Currently TabNine will stop at the first .git it finds, but it should continue to try to find a .tabnine_root.

The Vim plugin is based on YouCompleteMe, which I can't use because it conflicts with my other autocompleting plugin (deoplete). I would happily write my own plugin (deoplete source) but the main thing preventing me from doing that is no central source for binaries; instead they're included in the git repo (which also means you have to download every version of every binary for every platform when you install the plugin.)

Could there be some documentation provided for update.tabnine.com?

I'm using TabNine 1.0.9 in SublimeText 3.1.1 on OS X 10.13.6.

Much of the time, the TabNine process shows that it's using 100-120% CPU. The rest of the time it bounces between 0% and 40% CPU. I'm not sure what is causing this; I thought it might be related to backup activity causing a lot of filesystem events, but now I'm not sure.

Assuming this is not expected behavior, I'm happy to help troubleshoot this.

I should add that I really like TabNine's functionality, so I've been putting up with the high CPU except when running on battery.

While working on rustc, I noticed TabNine and rls regularly consuming 100% CPU each for quite a while. I usually only notice it when I switch to a different virtual desktop when I build rustc, so it's not only while actually working in vim, but also(?) a while after I stop changing things.

Additionally, the memory usage for TabNine is huge. Currently top shows:

23316 doener 20 0 9964048 4,9g 0 S 0,0 31,2 74:41.38 TabNine

The process was started about 86 hours ago.

I have enabled semantic completion in deoplete + deoplete-tabnine environment.
But it seems does not generate candidates in python files.
semantic completion is already enabled and pyls is executable.

Can you test it?

I would like to update the binary via command only rather than auto-updating as described in the TabNine FAQ. I don't want any external requests to be made except by command.

I recently got these two messages telling me that TabNine wants to access my contacts and calendar.
That sounds a bit creepy if you ask me, could you explain to me why TabNine would want to do that?

Golang server is a bit slow, I want to try bingo. Could you support this option in config file? Thanks!

I've tried several completion backends and all result in the same popup:

I have tried JavaScript with the default settings, with flow pre-installed and removing the install command, removing the install command and using /usr/local/bin/flow as an absolute path for command. I've tried hie in a stack repo, in both subdir files and Setup.hs at the project root. I've also run hie and flow alone and they run fine if started manually.

I'd like to debug this, but without stdout/stderr logged somewhere useful I don't know how to debug this short of trying to ptrace the binary, but don't have time right this moment.

Edit: This is on MacOS, within MacVim using the YCM fork vim plugin.

Eg. no need to wait 50ms every time user types . if the backend always takes at least 200ms

A lot of users = profit

TabNine only inserts a completion for the first cursor when using multiple cursors in VSCode, not for all cursors as expected.

First of all, just wanted to say that I'm loving the product, keep up the good work! The moment you released semantic completion I instantly bought a license!

It would be nice to get semantic completion for CSS so that I don't have to manually trigger my editor's IntelliSense to get suggestions.

This is my configuration.

[language.typescript]
command = "typescript-language-server"
args = ["--stdio"]
install = [["npm", "install", "-g", "typescript-language-server"]]

If you build it, I buy it. :)

Hi! I'm a maintainer of Reason. Currently, tabnine seems to use ocaml-language-server. For Reason, we now have a new language server more suited to it here.

cc @jaredly

Thanks!

Sometimes language servers use lots of resources upon first startup

I'm using TabNine daily at work and would like to support TabNine's continued development/maintenance. I like your current business model of one-time payment, but it'd be nice for those who wish to continue to support you beyond the one-time payment to have the option of doing so.

Thank you.

Hi there! First I would like to thank you for your wonderful auto-completer. I've been trying it out for the last couple of days and it works well. Is there any plan to release a roadmap or backlog for the development of this plugin?

w3 feed validator reports following:

This feed does not validate.

line 1, column 14: XML parsing error: <unknown>:1:14: not well-formed (invalid token) [help]

<script async src="https://www.googletagmanager.com/gtag/js?id=UA-125818377- ...
              ^
In addition, interoperability with the widest range of feed readers could be improved by implementing the following recommendation.

Feeds should not be served with the "text/html" media type [help]

https://validator.w3.org/feed/check.cgi?url=https%3A%2F%2Ftabnine.com%2Fblog

When writing JavaScript I frequently get completion offers like someMethod('. While this may be useful to someone who's editor does not manage pairs of delimiter characters for them, using this in NeoVim equipped with plugins to manage these is a PITA (as this suggestion always is above someMethod). Is there a way to disable this or configure some characters to be ignored?

IMHO, automated management of quote/parentheses pairs in editors and IDEs is so ubiquitous these days that this behaviour should just be removed completely or at least it should require an explicit option turned on to behave like this. YMMV.

First of all: TabNine is awesome, immediately purchased it after using it for 5 minutes.
I have one small problem though: Every time I restart Sublime (3180) I have to enter my license again because TabNine seemingly forgets that it is already activated.

Sublime Console output:

reloading plugin TabNine.TabNine
reloading settings Packages/TabNine/TabNine.sublime-settings
TabNine: starting version 0.4.7
Exception while interacting with TabNine subprocess: [Errno 32] Broken pipe
TabNine: starting version 0.11.2

Displayed warning:

With kind regards,
Philip Trauner

This is needed by Emacs client

I'm using TabNine latest, nvim latest, with the deoplete plugin. macOS latest as well. I have semantic completion enabled with solargraph (also latest) as the backend (stdio). My TabNine.toml, .solargraph.yml, and .vimrc files are located here: https://github.com/jaydorsey/dotfiles

I see an issue where the TabNine results show an extra/duplicate letter as in the screenshot below:

If I press enter/return it enters the text as present (with the extra character).

Stopping & restarting nvim makes the problem go away, but not for very long.

I don't remember it doing this when I first enabled semantic completion, but I might be mistaken. It seems to have manifested over the last update or two. I've seen it happen with one other user but they have nearly the same setup I have.

Things I've tried that didn't work:

• Changing the num_server_instances (1 - 4) in TabNine.toml
• Changing the synchronous_timeout_ms (100-500) in TabNine.toml
• Updating/clearing/downloading the solargraph-cores
• Deleting my TabNine.toml file (in hopes that semantic completion would be disabled)
• Trying a nearly empty vimrc file (also ran PlugClean & restarted nvim after emptying) (below)
• Trying a nearly empty vimrc, and an empty TabNine.toml file

My mostly empty vimrc file:

set nocompatible  " We don't want vi compatibility.
set termguicolors " Enable true color support
set t_Co=256

let g:python3_host_prog = '/usr/local/bin/python3'
let g:python_host_prog = '/usr/local/bin/python'

call plug#begin()

Plug 'Shougo/deoplete.nvim', { 'do': ':UpdateRemotePlugins' }
Plug 'tbodt/deoplete-tabnine', { 'do': './install.sh' }
let g:deoplete#enable_at_startup = 1
let g:deoplete#ignore_sources = get(g:,'deoplete#ignore_sources',{})
" Disable tags file as a source
let g:deoplete#ignore_sources.ruby = ['tag']

" Use <tab> key to step into the selections
inoremap <expr><tab>  pumvisible() ? "\<c-n>" : "\<tab>"
call plug#end()

This does appear to affect things like trying to type TabNine::restart, even when I don't select rrestart as an autocompletion (I can't call restart by typing)

Other things:

• I can open up a new nvim instance, aside from one that's been impacted by this, and everything works fine
• Restarting nvim completely makes the issue go away for a short period of time (minutes)
• I would try disabling semantic completion, but I don't see a TabNine::disable_sem command?
• It never does this duplication with any more than the first 2 letters from what I can tell

My next step is to try to revert back to an older version of TabNine to see if I could isolate it vs solargraph but I wanted to see if anyone else was having similiar issues or could recommend a setting I haven't looked at.

LSP semantic completion has type information.
I think it should be displayed in tabNine client.

I want to support the feature in deoplete.

Even the Premium version of TabNine is advertised with a limit of "Up to 15 MB indexed".

I understand that for most smaller projects that's sufficient. But looking at the sources of Blender or the Linux kernel as bigger, but still reasonable projects it is evident that this restriction is somewhat too small.

This makes me think that there are some hidden performance problems when running on such large code bases. For example, will the code completion persist any indexing it does to disk?

Since the free version supports only 200 KB of files I can hardly do any testing. Do you have any solid benchmarks to show where the completion runs on larger projects?

Ideally, user-configurable

here in this gif when i try to write exists after exi tabnine just inserts e and wont let me delete it. this is really annoying also for some reason it inserted o after exist in the gif .
Is there any way to disable autofil in tabnine and allow only to fill using or ?

It is like languages.yml.

I have tested some LSP clients and they work well.
I think it should be default.

For many projects, .gitignore is not enough, we also want to explicitly ignore certain folders of autogenerated code while keeping them in version control.

For example, in our project we have:

client/
    js/
        vendor/
             ...       // we want to ignore everything in here, but keep it in our VCS

server/
    build/
        ...            // we also want to ignore all autogenerated code in here, while keeping it in our VCS
    src/
        ...            // but we do want to autocomplete any files from here

Suggested Fix:
Allow adding a .tabnineignore file, just like .gitignore, .dockerignore, .npmignore, etc.
This can be excluded from VCS or added, it's up to the user.

Might be useful/appropriate to give a shoutout to https://github.com/github/linguist etc.

Moving tbodt/deoplete-tabnine#18 to the appropriate place.

The log appears to be out of control:

~/.config/nvim/plugged/deoplete-tabnine master*
❯ ll tabnine.log
-rw-r--r--  1   941M Jan 31 13:27 tabnine.log

It seems to be mostly polluted by the following message:

~/.config/nvim/plugged/deoplete-tabnine master* 27s
❯ tail tabnine.log
[2019-01-31][13:27:12][server::pipeline::walk][ERROR] Walk error: No such file or directory (os error 2)
[2019-01-31][13:27:12][server::pipeline::walk][ERROR] Walk error: No such file or directory (os error 2)
[2019-01-31][13:27:12][server::pipeline::walk][ERROR] Walk error: No such file or directory (os error 2)
[2019-01-31][13:27:12][server::pipeline::walk][ERROR] Walk error: No such file or directory (os error 2)
[2019-01-31][13:27:12][server::pipeline::walk][ERROR] Walk error: No such file or directory (os error 2)
[2019-01-31][13:27:12][server::pipeline::walk][ERROR] Walk error: No such file or directory (os error 2)
[2019-01-31][13:27:12][server::pipeline::walk][ERROR] Walk error: No such file or directory (os error 2)
[2019-01-31][13:27:12][server::pipeline::walk][ERROR] Walk error: No such file or directory (os error 2)
[2019-01-31][13:27:12][server::pipeline::walk][ERROR] Walk error: No such file or directory (os error 2)
[2019-01-31][13:27:12][server::pipeline::walk][ERROR] Walk error: No such file or directory (os error 2)

I'm not sure if it's related to the CPU usage problems, but TabNine has been spiking to 200-300% cpu util without any apparent need (in fact often vim is actually suspended through C-Z).

Potentially related to #24?

Maybe you can add plugin for CudaText too? Python API.

TabNine can launch a language server for semantic completion. But language servers do much more than mere completion - they provide linting, code navigation, documentation, code actions etc.

If I launch a language server I want access to all these things, and many LSP clients provide them, but if TabNine is the one launching the server I can only use it for completion - and nothing else.

Is it possible to have TabNine simply delegate messages from the client to the language server and from the language server to the client? If it gets completion requests/responses it does its magic, but everything else it should just let pass through. That way I can just set my LSP client to launch TabNine and get completion from TabNine+language server and everything else from the language server.

It would be great if there's TabNine binary for FreeBSD platform. I mainly work on macOS and on FreeBSD, and I think it would be great if I could use TabNine on both platforms.

Thanks!

A lot of users = profit

The extension should be able to determine the most common name convention used in the workspace and prioritize suggesting those. Currently it only seems to prioritize snake_case, even though all of my files use camelCase.

There's a blacklist feature in autocomplete-plus for example. I'd like to turn suggestions off for text files like .txt, .md, etc

I have .tabnineignore set up but it seems like though it's not indexing those files it's still providing suggestions

"xml":"Enabled" in tabnine_config.json

Deoplete appears to be supported but NCM2 is popular as well.