mirror of
https://github.com/stablyai/orca.git
synced 2026-10-07 16:02:29 +00:00
Neilandczzczz
0a45ad5ec4
The jcode journal reader, the Claude transcript reader and the Command Code transcript reader each carried their own copy of the same reverse chunked line scan; they now share one tested helper. jcode's managed hook script drops its hand-rolled curl for buildPosixAgentHookPostCommand, which also gains it the raw-JSON transport and the --noproxy guard the bespoke copy was missing. Co-authored-by: czzczz <chanzrz_zbf@foxmail.com>
70 lines
2.9 KiB
TypeScript
70 lines
2.9 KiB
TypeScript
// Why: jcode's TUI client starts its own server and then waits a hardcoded 5s for
|
|
// the socket to answer a ping (`wait_for_server_ready` in
|
|
// crates/jcode-app-core/src/server/socket.rs). Orca gives every pane its own
|
|
// JCODE_RUNTIME_DIR, so every jcode pane is a COLD daemon start — and a cold start
|
|
// on a loaded machine overruns that budget, which is how an orchestration worker
|
|
// died with "Timed out waiting for responsive server socket" before its prompt was
|
|
// ever delivered. Starting the daemon as the PTY spawns gives it the shell's own
|
|
// startup time as head start, so the client finds a live socket instead of racing.
|
|
import { spawnProcess } from '../../shared/child-process/run-process'
|
|
import { getTuiAgentLaunchCommand, TUI_AGENT_CONFIG } from '../../shared/tui-agent-config'
|
|
|
|
export type JcodeDaemonPrewarm = {
|
|
launchAgent?: string
|
|
runtimeDir?: string
|
|
cwd?: string
|
|
env?: Record<string, string>
|
|
platform?: NodeJS.Platform
|
|
}
|
|
|
|
/** Runtime dirs already warmed in this Orca process; the daemon outlives one pane. */
|
|
const warmed = new Set<string>()
|
|
|
|
export function resetJcodeDaemonPrewarmForTests(): void {
|
|
warmed.clear()
|
|
}
|
|
|
|
/** The runtime dir to warm, or null when this pane is not a local jcode launch.
|
|
* Why non-Windows only: the runtime dir is a unix-socket directory, and Orca only
|
|
* stamps it off Windows (see shouldInjectJcodeRuntimeDir). */
|
|
function prewarmTarget({ launchAgent, runtimeDir, platform }: JcodeDaemonPrewarm): string | null {
|
|
const unsupported = launchAgent !== 'jcode' || (platform ?? process.platform) === 'win32'
|
|
return unsupported || !runtimeDir ? null : runtimeDir
|
|
}
|
|
|
|
/**
|
|
* Start this pane's jcode daemon in the background, at most once per runtime dir.
|
|
*
|
|
* Fire-and-forget by contract: jcode's client starts its own server when none is
|
|
* listening, so a pre-warm that fails costs nothing beyond the cold start Orca
|
|
* already had. Never throws, and never blocks the spawn path.
|
|
*/
|
|
export function prewarmJcodeDaemon(args: JcodeDaemonPrewarm): boolean {
|
|
const runtimeDir = prewarmTarget(args)
|
|
if (runtimeDir === null || warmed.has(runtimeDir)) {
|
|
return false
|
|
}
|
|
warmed.add(runtimeDir)
|
|
// A missing binary or a spawn refusal just means no head start; forget the dir so
|
|
// the next pane on it can try again.
|
|
const giveUp = (): boolean => (warmed.delete(runtimeDir), false)
|
|
try {
|
|
const child = spawnProcess({
|
|
program: getTuiAgentLaunchCommand(TUI_AGENT_CONFIG.jcode, args.platform ?? process.platform),
|
|
// Why --no-update: an update check here would delay the very socket the client
|
|
// is about to wait on. Why stdio ignore + unref: the daemon is jcode's to own
|
|
// and must outlive this spawn, so Orca keeps no handle on it.
|
|
args: ['--no-update', 'serve'],
|
|
cwd: args.cwd,
|
|
env: { ...args.env, JCODE_RUNTIME_DIR: runtimeDir },
|
|
detached: true,
|
|
stdio: 'ignore'
|
|
})
|
|
child.unref()
|
|
child.on('error', giveUp)
|
|
return true
|
|
} catch {
|
|
return giveUp()
|
|
}
|
|
}
|