[quickbeard]

electron-builder --mac --universal did not produce a working app. Three separate problems, all found by building and then inspecting the packaged result.

1. The terminal backend was pinned to the build host's arch

electron.vite.config.ts rewrote @lydell/node-pty to @lydell/node-pty-${process.platform}-${process.arch}. That is the build host's arch, resolved once — and electron-vite build runs before electron-builder, emitting a single bundle for whatever is packaged afterwards. On Apple Silicon it baked darwin-arm64 in, so an --x64 or --universal build shipped an Electron shell that starts and then cannot load its terminal backend. The failure reads as an app bug, not a build one.

The wrapper's own entry already does the same process.platform/process.arch lookup at run time, so the import is externalised and left to dispatch.

2. …but the wrapper was never shipped

That change alone would have been worse than the bug. @lydell/node-pty was a devDependency, so electron-builder never packed it — the old narrowing worked only because it pointed at the platform packages, which are optionalDependencies.

$ asar list app.asar | grep lydell
/node_modules/@lydell/node-pty-darwin-arm64        ← present
/node_modules/@lydell/node-pty-darwin-x64          ← present
                                                   ← wrapper absent

So require("@lydell/node-pty") could not resolve. Worth noting the bundled server (packages/opencode/dist/node/node.js) already required the bare wrapper, so packaged builds were carrying this break before any of my changes. It moves to dependencies.

3. @electron/universal refused to merge

Both halves of a universal build carry all six platform packages, so eight Mach-O files end up byte-identical across x64 and arm64 — node-pty's pty.node and its extension-less spawn-helper, plus @parcel/watcher and msgpackr-extract. @electron/universal treats that as a mistake and aborts. They are prebuilt binaries chosen at run time, not per-arch build output, so mac.x64ArchFiles now covers the unpacked node_modules.

I enumerated all eight by diffing the two temp app trees rather than discovering them one failed build at a time — my first pattern only covered *.node and missed spawn-helper.

Verification

On a packaged universal app, copied outside the repo and launched:

main:      x86_64 arm64
framework: x86_64 arm64
helper:    x86_64 arm64
pty.node loaded from: /Contents/Resources/app.asar.unpacked/node_modules/
                      @lydell/node-pty-darwin-arm64/prebuilds/darwin-arm64/pty.node

Running it from inside the repo is not a valid test: Node's module walk-up finds codev-code/node_modules and masks a missing wrapper. That is exactly how problem 2 nearly shipped — the app looked fine until I moved it to /tmp.

Intel itself is unverified — I have no x64 hardware here. What is verified is that the x64 prebuild ships and that the selection is made at run time from process.arch.

bun typecheck clean; bun lint 4741 + 1 error against main's 4746 + 1.

Not included

No CHANGELOG entry: release.yml builds no desktop artifact, so no shipped build changes behaviour for a user today. Worth adding if desktop builds ever start shipping — at which point problem 2 would have made the terminal dead on arrival.

🤖 Generated with Claude Code