Server process exited with signal SIGTRAP

[jar-presta-onepoint]

Hi,

On VScode version 1.101.0, me and other people on my project are seeing the following error in ESlint's output log:
Server process exited with signal SIGTRAP

I am on Ubuntu 24.10, and another person that had the same issue is on MacOS.

Rolling back to a previous version of VScode helped resolve the issue for at least one person.

We believe that there might be a load balancing issue with Electron or node. The upgrades made to both node and electron in VScode's 1.101 might have something to do with this.

[jar-presta-onepoint]

I should also point out that our project is a (huge) monorepo that uses Yarn for package management.

We're also using Yarn's zero-installs caching strategy

[Wroud]

Same here, and also using yarn with pnp (without zero-installs)
Should we open an issue in Yarn's repo?
https://github.com/yarnpkg/berry/issues

[jar-presta-onepoint]

I have created a ticket on yarnpkg/berry#6822

[jar-presta-onepoint]

FYI, increasing ESlint's memory budget, as proposed in this comment does not solve the issue

[arcanis]

Something seems broken in the Electron or VSCode build. The following setting workarounds the issue; note it's the exact same version as the one VSCode is supposed to ship:

{
  "eslint.runtime": "/Users/mael.nison/.volta/tools/image/node/22.15.1/bin/node",
}

Unfortunately I don't know how to bisect efficiently to figure out which Electron / VSCode version introduced the regression.

[arcanis]

It can be reproduced more easily by cloning the yarnpkg/berry repository then running the following command; it works with Node, but crashes with the Electron process:

$ ELECTRON_RUN_AS_NODE=1 \
  "/Applications/Visual Studio Code.app/Contents/Frameworks/Code Helper (Plugin).app/Contents/MacOS/Code Helper (Plugin)" \
  --require $PWD/.pnp.cjs \
  --experimental-loader file://$PWD/.pnp.loader.mjs \
  ./.yarn/sdks/eslint/lib/unsupported-api.js

[1]    8936 trace trap  ELECTRON_RUN_AS_NODE=1  --require /Users/mael.nison/berry/.pnp.cjs

[jar-presta-onepoint]

I confirm that setting a hard path to my node executable (version 22.16.0 in my case) indeed allows the ESlint server to work properly.

Do you think it's fair to say that this is more of a Yarn problem than an ESlint problem then ?

Really appreciate your help by the way, thank you 👍

[arcanis]

I suspect it's an Electron issue - they apply patches on top of Node.js, and one of them probably accidentally causes an assertion to trigger under some specific conditions (Node.js has been refactoring the resolver code lately, I imagine it'd explain why it suddenly started to break).

I didn't manage to figure out which assertion it is yet due to OSX preventing to debug releases processes; at the very least we should open an issue on the Electron bug tracker and ask for their opinion.

[jar-presta-onepoint]

Well I sure hope I asked in the right place, 'cause at this point I regret not having paid more attention to my system programming classes

[arcanis]

To note, if you're on Ubuntu and it repros, you might be able to use gdb to get more information than I could (especially the stacktrace triggering the trap). I tried on WSL but it didn't repro there.

[dbaeumer]

@deepak1556 does this ring any bell. The ESLint server only crashes with the SIGTRAP signal when running on our latest?

[jar-presta-onepoint]

To note, if you're on Ubuntu and it repros, you might be able to use gdb to get more information than I could (especially the stacktrace triggering the trap). I tried on WSL but it didn't repro there.

I am struggling to find VScode's builtin electron binary/executable, and I'm not sure if this is what I ought to look for to analyse the stacktrace (Does it have the debug flags on ? Should I just git clone Electron and compile it with gcc's debug flag ?).
Could you supply instructions as to what I or someone else needs to do to analyse the stacktrace ?