fix pty output make main process oom

[theanarkh]

Fixes #234393

The local pty process communicates directly with the render process, but the main process save the output of pty process to buffer and there is no consumer.

Running

node -e "setInterval(() => console.log('x'.repeat(100)), 10)"

in the terminal and take a JS heap snapshot of main process.

take JS heap snapshot again after some time.

[vs-code-engineering]

📬 CODENOTIFY

The following users are being notified based on files changed in this PR:

@deepak1556

Matched files:

• src/vs/code/electron-main/app.ts

[theanarkh]

@microsoft-github-policy-service agree

[Copilot]

Pull request overview

This PR fixes a memory leak (OOM) in the main Electron process caused by unbounded buffering of PTY process output. The local PTY process communicates directly with the renderer process via MessagePort, but the main process was also buffering all onProcessData events even though they were never consumed, leading to memory exhaustion.

Key Changes:

• Added a skip option to ProxyChannel.fromService to selectively exclude events from registration
• Applied the skip option to exclude onProcessData events from the main process PTY service channel

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
src/vs/base/parts/ipc/common/ipc.ts	Added skip option to ICreateServiceChannelOptions interface to allow selective event registration filtering
src/vs/code/electron-main/app.ts	Applied skip option to exclude onProcessData from the PTY host channel registration in main process

[Tyriar]

lgtm thanks!

FYI @joaomoreno @alexdima if you have thoughts on ipc/ parts

[bpasero]

I am reading this comment and fail to understand what actually happens when I use this option. If I understand correctly, a "lazy" event will not buffer any data until a first listener is added. But then I would update this comment to actually explain this better. And should be called differently?

[theanarkh]

Yes. I have updated the comment, please help review again .

[bpasero]

Whats the rule here to decide if an event should be lazy? Does this apply to other proxy channels?

[theanarkh]

Thanks for reviewing ! An event should be lazy if it is triggered automatically but may not have subscribers; otherwise, the event data will be cached in the buffer (Event.buffer), causing a memory leak. I only find that ptyHostService.onProcessData should be lazy currently.

[Tyriar]

@theanarkh I think what @bpasero is likely getting at is this should get fixed generically to prevent future leaks like this in the future. onProcessData is used, but conditionally, just because an event isn't listened to shouldn't mean the data should hang around forever. Any thoughts on how we could solve this generically for all events without hardcoding this particular event or compromising performance?