test: implement test isolation for e2e tests

.gitignore

@@ -91,11 +91,15 @@ typings/
 # Electron-Forge
 out/
 
-bin/
+bin/*/thv
+bin/*/README.md
+bin/*/LICENSE
 
 # Playwright
 /test-results/
 /playwright-report/
 /blob-report/
 /playwright/.cache/
 /test-videos/
+
+

bin/Dockerfile

@@ -0,0 +1,19 @@
+FROM docker:dind
+
+RUN apk add --no-cache dbus dbus-x11 gnome-keyring libsecret
+
+RUN setcap -r /usr/bin/gnome-keyring-daemon 2>/dev/null || true
+
+ENV XDG_CURRENT_DESKTOP=GNOME \
+    XDG_SESSION_DESKTOP=gnome \
+    DESKTOP_SESSION=gnome \
+    # a writable runtime dir (glib falls back to /tmp if this is unset)
+    XDG_RUNTIME_DIR=/tmp/xdg-runtime
+ENV DOCKER_HOST=unix:///var/run/docker.sock
+
+COPY --chmod=755 linux-x64/thv        /usr/local/bin/thv
+COPY ./ephemeral/entrypoint.sh /usr/local/bin/
+RUN chmod +x /usr/local/bin/entrypoint.sh
+
+ENTRYPOINT ["entrypoint.sh"]
+CMD []

bin/ephemeral/entrypoint.sh

@@ -0,0 +1,21 @@
+#!/usr/bin/env sh
+set -eu
+
+export CI=true
+
+/usr/local/bin/dockerd-entrypoint.sh &
+until docker info >/dev/null 2>&1; do sleep 0.5; done
+echo "🐳 Docker-in-Docker daemon is ready."
+
+mkdir -p /tmp/xdg-runtime/keyring
+chmod 700 /tmp/xdg-runtime /tmp/xdg-runtime/keyring
+
+eval "$(dbus-launch --sh-syntax)"
+export XDG_RUNTIME_DIR=/tmp/xdg-runtime
+
+echo "default-password" | gnome-keyring-daemon --unlock --components=secrets,ssh &
+sleep 2
+
+export GNOME_KEYRING_CONTROL GNOME_KEYRING_PID
+
+exec "$@"

bin/ephemeral/thv.sh

@@ -0,0 +1,19 @@
+#!/usr/bin/env bash
+set -euo pipefail
+set -x
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+ROOT_DIR="$(cd "${SCRIPT_DIR}/.." && pwd)"
+
+(
+  cd "${ROOT_DIR}"
+  docker run --privileged \
+    --cap-drop=SETPCAP \
+    --cap-add=IPC_LOCK \
+    --tmpfs /run \
+    --rm -i \
+    --network host \
+    -v "${ROOT_DIR}:/workspace" \
+    -w /workspace \
+    thv-containerized thv "$@"
+)

main/src/toolhive-manager.ts

@@ -7,23 +7,16 @@ import type { Tray } from 'electron'
 import { updateTrayStatus } from './system-tray'
 import log from './logger'
 import * as Sentry from '@sentry/electron/main'
+import { delay } from '../../utils/delay'
+
+// Use environment variables for binary customization with Windows fallback
+const binName =
+  process.env.BIN_NAME ?? (process.platform === 'win32' ? 'thv.exe' : 'thv')
+const binArch = process.env.BIN_ARCH ?? `${process.platform}-${process.arch}`
 
-const binName = process.platform === 'win32' ? 'thv.exe' : 'thv'
 const binPath = app.isPackaged
-  ? path.join(
-      process.resourcesPath,
-      'bin',
-      `${process.platform}-${process.arch}`,
-      binName
-    )
-  : path.resolve(
-      __dirname,
-      '..',
-      '..',
-      'bin',
-      `${process.platform}-${process.arch}`,
-      binName
-    )
+  ? path.join(process.resourcesPath, 'bin', binArch, binName)
+  : path.resolve(__dirname, '..', '..', 'bin', binArch, binName)
 
 let toolhiveProcess: ReturnType<typeof spawn> | undefined
 let toolhivePort: number | undefined
@@ -75,6 +68,8 @@ export async function startToolhive(tray?: Tray): Promise<void> {
     }
   )
 
+  await delay(4000)
+
   log.info(`[startToolhive] Process spawned with PID: ${toolhiveProcess.pid}`)
 
   if (tray) {

package.json

@@ -10,7 +10,8 @@
   },
   "scripts": {
     "start": "electron-forge start",
-    "e2e": "env CI=true sh -c \"tsc -b --clean && tsc -b && electron-forge package && playwright test\"",
+    "start:ephemeral": "BIN_NAME='thv.sh' BIN_ARCH='ephemeral' electron-forge start",
+    "e2e": "env CI=true sh -c \"tsc -b --clean && tsc -b && electron-forge package && BIN_NAME='thv.sh' BIN_ARCH='ephemeral' playwright test\"",
     "start:mockUpdate": "MOCK_UPDATE_SERVER=true ELECTRON_IS_DEV=0 electron-forge start",
     "package": "tsc -b --clean && tsc -b && electron-forge package",
     "make": "tsc -b --clean && tsc -b && electron-forge make",