fix(provider): terminate job before closing terminal to prevent orphaned processes by pedropombeiro · Pull Request #168 · nickjvandyke/opencode.nvim

pedropombeiro · 2026-02-09T22:11:08Z

Summary

When Neovim exits, the opencode --port process stays orphaned in memory instead of being terminated.

Root Cause

During VimLeavePre, snacks.terminal has already cleaned up its internal state, which invalidates the Neovim job. This means:

jobstop(job_id) returns true but doesn't actually terminate the process
jobpid(job_id) returns nil - Neovim no longer knows which PID the job maps to

Solution

Capture the PID (not just job ID) when the terminal opens, and use shell kill to terminate it directly during VimLeavePre. This bypasses Neovim's job tracking entirely.

Changes

snacks provider: Track _pid at toggle/start time using jobpid(terminal_job_id)
snacks provider: Use vim.fn.system('kill -TERM ' .. pid) in stop() for reliable termination
plugin/provider.lua: Change autocmd from VimLeave to VimLeavePre for earlier cleanup

Testing

Start opencode via :Opencode
Exit Neovim (:qa)
Verify no orphaned processes: ps aux | grep "opencode --port" should return nothing

nickjvandyke · 2026-02-09T22:27:14Z

Thanks for taking a crack at this!

Is it possible to stop the job and all its children too? The child process (opencode) seems to stick around for me. Not sure if this is specific to fish.

pedropombeiro · 2026-02-09T23:37:32Z

Thanks for taking a crack at this!

Is it possible to stop the job and all its children too? The child process (opencode) seems to stick around for me. Not sure if this is specific to fish.

Can you try again with the latest version?

Track the terminal's PID at toggle/start time and use shell 'kill' to terminate it during VimLeavePre. This is more reliable than jobstop() because snacks.terminal cleanup can invalidate the job before our stop() is called.

nickjvandyke · 2026-02-10T17:55:19Z

I think the root issue is in opencode. Opened anomalyco/opencode#13001. Will see what happens there first 🙂

nickjvandyke

Thanks for this clever solution! Just a couple questions 😁

lua/opencode/provider/terminal.lua

lua/opencode/provider/snacks.lua

… Windows support Move PID capture logic into an on_buf callback in the constructor so it fires automatically for both toggle() and open(), eliminating duplicated code. Wrap the user's existing on_buf callback to preserve custom behavior (e.g., keymap application). Guard the shell 'kill -TERM' call with a Unix platform check and fall back to vim.uv.kill() on non-Unix systems for cross-platform compatibility.

…l provider Extract process kill and PID capture logic into opencode.provider.util, shared by both snacks and terminal providers. This eliminates duplicated code and ensures consistent behavior. Key fix: use os.execute() instead of vim.fn.system() for process group kill during VimLeavePre, and skip jobstop when PID kill succeeds to prevent SIGHUP from causing the process to respawn. Apply the same PID-based termination to the terminal provider, which suffered from the same orphaned process issue as the snacks provider.

Add diagnostic disable/enable annotations around private field access in the on_buf closure, which is part of the constructor but lua-ls treats as an external context.

…haned processes Add PID-based process group kill to the tmux provider, matching the approach used by the snacks and terminal providers. Capture the pane PID via tmux display-message and kill the process group before tmux kill-pane. This is a workaround for anomalyco/opencode#13001 (opencode does not handle SIGHUP gracefully). Document the upstream issue in provider.util so the workaround is easy to identify and remove later.

nickjvandyke

This is looking great, thank you 😀 just a couple more questions, then I think we are set! Thanks for all the digging you did - turns out there are a lot of quirks here 😅

lua/opencode/provider/util.lua

lua/opencode/provider/snacks.lua

…ture Remove job_id parameter from util.kill() and the jobstop fallback, since jobstop sends SIGHUP which causes the process to respawn — making it counterproductive. Only PID-based process group kill is effective. Add comments explaining why PID must be captured eagerly at startup: terminal_job_id is no longer available by the time VimLeavePre/stop() runs.

TermOpen fires exactly when the terminal job starts, rather than relying on a constant delay that may be too early or too late.

pedropombeiro · 2026-02-20T16:30:14Z

Thanks @nickjvandyke! I've addressed your comments 🏓

nickjvandyke · 2026-02-20T17:05:38Z

Thank you, looks great!

I may just check out the branch and move code around a bit according to the long-term provider API I have in mind if that's alright. Specifically Provider:get_pid() (which is really neat that we get that in this PR for a few of them!) Then we're good to go 😀

pedropombeiro · 2026-02-20T20:49:07Z

Sure, feel free to make any changes. Looking forward to having this solved 👍

nickjvandyke · 2026-02-20T21:15:06Z

Interestingly turns out we do need to vim.fn.jobstop in Terminal:stop when Neovim isn't exiting. I guess it attempts to restart the process or something if it's killed while the job is still running? Not necessary for snacks.terminal - I guess it does that internally in its win:close() or such.

Thank you for the quality contribution! 😁

nickjvandyke mentioned this pull request Feb 9, 2026

bug: provider's opencode process orphans when stopping #118

Open

3 tasks

pedropombeiro force-pushed the fix/terminate-job-on-stop branch from 9a5837e to d41da26 Compare February 9, 2026 23:02

pedropombeiro closed this Feb 9, 2026

pedropombeiro reopened this Feb 9, 2026

pedropombeiro force-pushed the fix/terminate-job-on-stop branch from d41da26 to f56624a Compare February 9, 2026 23:34

pedropombeiro force-pushed the fix/terminate-job-on-stop branch from f56624a to 733c590 Compare February 9, 2026 23:40

fix(snacks): terminate opencode process on VimLeavePre

ca9251c

Track the terminal's PID at toggle/start time and use shell 'kill' to terminate it during VimLeavePre. This is more reliable than jobstop() because snacks.terminal cleanup can invalidate the job before our stop() is called.

pedropombeiro force-pushed the fix/terminate-job-on-stop branch from 733c590 to ca9251c Compare February 9, 2026 23:41

nickjvandyke force-pushed the main branch from 1811541 to 65ce845 Compare February 15, 2026 17:26

nickjvandyke reviewed Feb 19, 2026

View reviewed changes

lua/opencode/provider/terminal.lua Outdated Show resolved Hide resolved

lua/opencode/provider/snacks.lua Outdated Show resolved Hide resolved

lua/opencode/provider/snacks.lua Outdated Show resolved Hide resolved

pedropombeiro and others added 5 commits February 19, 2026 22:56

fix(snacks): suppress lua-language-server invisible field warnings

70ea107

Add diagnostic disable/enable annotations around private field access in the on_buf closure, which is part of the constructor but lua-ls treats as an external context.

Merge branch 'main' into fix/terminate-job-on-stop

a702a58

nickjvandyke reviewed Feb 20, 2026

View reviewed changes

lua/opencode/provider/util.lua Outdated Show resolved Hide resolved

lua/opencode/provider/snacks.lua Outdated Show resolved Hide resolved

lua/opencode/provider/snacks.lua Outdated Show resolved Hide resolved

pedropombeiro added 2 commits February 20, 2026 17:18

refactor(snacks): use TermOpen autocmd for more reliable PID capture

b3a2d68

TermOpen fires exactly when the terminal job starts, rather than relying on a constant delay that may be too early or too late.

pedropombeiro force-pushed the fix/terminate-job-on-stop branch from 1a1fdf3 to b3a2d68 Compare February 20, 2026 16:27

nickjvandyke added 3 commits February 20, 2026 13:38

not sure this boolean check is right?

1fe20fc

localize getting pid to providers

dd20eac

comments

6b927b6

nickjvandyke added 2 commits February 20, 2026 13:57

use kill for tmux for consistency

681afb3

nit

3ee215d

nickjvandyke added 2 commits February 20, 2026 14:06

argh

9bcd256

properly stop terminal process when not exiting

bce4875

nickjvandyke merged commit 125c7dc into nickjvandyke:main Feb 20, 2026
2 checks passed

github-actions bot mentioned this pull request Feb 20, 2026

chore(main): release 0.4.0 #177

Open

Uh oh!

Comments

Conversation

pedropombeiro commented Feb 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Root Cause

Solution

Changes

Testing

Uh oh!

nickjvandyke commented Feb 9, 2026

Uh oh!

pedropombeiro commented Feb 9, 2026

Uh oh!

nickjvandyke commented Feb 10, 2026

Uh oh!

nickjvandyke left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

nickjvandyke left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

pedropombeiro commented Feb 20, 2026

Uh oh!

nickjvandyke commented Feb 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pedropombeiro commented Feb 20, 2026

Uh oh!

nickjvandyke commented Feb 20, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

pedropombeiro commented Feb 9, 2026 •

edited

Loading

nickjvandyke commented Feb 20, 2026 •

edited

Loading