@oh-my-pi/pi-coding-agent 16.1.13 → 16.1.15

package/src/tiny/worker.ts

@@ -1,4 +1,3 @@
-import { createRequire } from "node:module";
 import * as path from "node:path";
 import type {
 	ProgressInfo,
@@ -6,16 +5,18 @@ import type {
 	TextGenerationStringOutput,
 	StoppingCriteria as TransformersStoppingCriteria,
 } from "@huggingface/transformers";
-import {
-	ensureRuntimeInstalled,
-	getTinyModelsCacheDir,
-	installRuntimeModuleResolver,
-	isCompiledBinary,
-	prompt,
-	resolveRuntimeModule,
-} from "@oh-my-pi/pi-utils";
-import packageJson from "../../package.json" with { type: "json" };
+import { getTinyModelsCacheDir, prompt } from "@oh-my-pi/pi-utils";
 import tinyTitleSystemPrompt from "../prompts/system/tiny-title-system.md" with { type: "text" };
+import {
+	errorMessage,
+	errorText,
+	getTransformersVersionSpec,
+	loadTransformersRuntime,
+	MemoizedRuntime,
+	replayCachedReady,
+	sendLog,
+	sendProgress,
+} from "../subprocess/worker-runtime";
 import { resolveTinyModelDevicePreference, type TinyModelDevice, tinyModelDeviceLoadOrder } from "./device";
 import { resolveTinyModelDtypeOverride, type TinyModelDtype } from "./dtype";
 import {
@@ -25,7 +26,7 @@ import {
 	type TinyTitleLocalModelSpec,
 } from "./models";
 import { formatTitleUserMessage, normalizeGeneratedTitle } from "./text";
-import type { TinyTitleProgressEvent, TinyTitleTransport, TinyTitleWorkerInbound } from "./title-protocol";
+import type { TinyTitleTransport, TinyTitleWorkerInbound } from "./title-protocol";
 
 const TITLE_PREFILL = "<title>";
 const TITLE_CLOSE = "</title>";
@@ -34,9 +35,6 @@ const STOP_DECODE_WINDOW_TOKENS = 32;
 const MEMORY_COMPLETION_DEFAULT_MAX_NEW_TOKENS = 256;
 const COMPLETION_MAX_NEW_TOKENS = 1024;
 const TINY_TITLE_SYSTEM_PROMPT = prompt.render(tinyTitleSystemPrompt);
-const TRANSFORMERS_PACKAGE = "@huggingface/transformers";
-const COMPILED_TRANSFORMERS_VERSION = process.env.PI_TINY_TRANSFORMERS_VERSION;
-const sourceRequire = createRequire(import.meta.url);
 
 const tinyModelDevicePreference = resolveTinyModelDevicePreference();
 const tinyModelDtypeOverride = resolveTinyModelDtypeOverride();
@@ -64,51 +62,11 @@ interface TransformersRuntime {
 
 const pipelines = new Map<TinyLocalModelKey, Promise<TextGenerationPipeline>>();
 
-function resolveTransformersVersionSpec(): string {
-	const manifest = packageJson as {
-		optionalDependencies?: Record<string, string>;
-		dependencies?: Record<string, string>;
-	};
-	const versionSpec =
-		manifest.optionalDependencies?.[TRANSFORMERS_PACKAGE] ?? manifest.dependencies?.[TRANSFORMERS_PACKAGE];
-	if (!versionSpec) throw new Error(`${TRANSFORMERS_PACKAGE} is missing from package.json optionalDependencies`);
-	if (!versionSpec.startsWith("catalog:")) return versionSpec;
-	if (COMPILED_TRANSFORMERS_VERSION) return COMPILED_TRANSFORMERS_VERSION;
-	const installed = sourceRequire(`${TRANSFORMERS_PACKAGE}/package.json`) as { version: string };
-	return installed.version;
-}
-let cachedTransformersVersionSpec: string | undefined;
-/**
- * Lazily resolve (and memoize) the transformers version spec. In the
- * `catalog:` case {@link resolveTransformersVersionSpec} `require`s the
- * installed `@huggingface/transformers/package.json`, so touching it forces
- * the dependency to exist. Defer it to the compiled-binary runtime-install
- * path — which only runs when a local title model is actually generated or
- * downloaded — so loading this worker (smoke-test ping, online title path)
- * never triggers the transformers resolve/install dance.
- */
-function getTransformersVersionSpec(): string {
-	cachedTransformersVersionSpec ??= resolveTransformersVersionSpec();
-	return cachedTransformersVersionSpec;
-}
 function getTransformersRuntimeKey(): string {
 	return getTransformersVersionSpec().replace(/[^A-Za-z0-9._-]/g, "_");
 }
 let generateQueue = Promise.resolve();
-let transformersRuntime: Promise<TransformersRuntime> | null = null;
-
-function errorText(error: unknown): string {
-	return error instanceof Error ? (error.stack ?? error.message) : String(error);
-}
-
-function sendLog(
-	transport: TinyTitleTransport,
-	level: "debug" | "warn" | "error",
-	msg: string,
-	meta?: Record<string, unknown>,
-): void {
-	transport.send({ type: "log", level, msg, meta });
-}
+const transformersRuntime = new MemoizedRuntime<TransformersRuntime>();
 
 function getTinyTitleRuntimeDir(): string {
 	return path.join(
@@ -118,74 +76,6 @@ function getTinyTitleRuntimeDir(): string {
 	);
 }
 
-function sendRuntimeInstallProgress(
-	transport: TinyTitleTransport,
-	requestId: string,
-	modelKey: TinyLocalModelKey,
-	status: "initiate" | "download" | "done",
-): void {
-	transport.send({
-		type: "progress",
-		id: requestId,
-		event: {
-			modelKey,
-			status,
-			name: `${TRANSFORMERS_PACKAGE}@${getTransformersVersionSpec()}`,
-		},
-	});
-}
-
-/**
- * Prepare the freshly-installed compiled runtime for loading: stub `sharp`
- * (the tiny models are text-generation only, so the native image pipeline is
- * dead weight) and patch the module resolver so Transformers.js's bare requires
- * (`onnxruntime-node`, `onnxruntime-common`) resolve against the cache. Returns
- * the absolute Transformers.js entrypoint to `require`.
- */
-async function prepareCompiledRuntime(runtimeDir: string): Promise<string> {
-	const nodeModules = path.join(runtimeDir, "node_modules");
-	const sharpStub = path.join(runtimeDir, "omp-sharp-stub.cjs");
-	await Bun.write(sharpStub, "module.exports = {};\n");
-	installRuntimeModuleResolver({ runtimeNodeModules: nodeModules, stubs: { sharp: sharpStub } });
-	const entry = resolveRuntimeModule(nodeModules, TRANSFORMERS_PACKAGE);
-	if (!entry) throw new Error(`Unable to resolve ${TRANSFORMERS_PACKAGE} in compiled runtime at ${nodeModules}`);
-	return entry;
-}
-
-function configureTransformers(transformers: TransformersRuntime): TransformersRuntime {
-	transformers.env.cacheDir = getTinyModelsCacheDir();
-	transformers.env.allowLocalModels = false;
-	transformers.env.logLevel = transformers.LogLevel.ERROR;
-	return transformers;
-}
-
-async function loadTransformers(
-	transport: TinyTitleTransport,
-	requestId: string,
-	modelKey: TinyLocalModelKey,
-): Promise<TransformersRuntime> {
-	if (transformersRuntime) return transformersRuntime;
-	transformersRuntime = (async () => {
-		if (!isCompiledBinary()) return configureTransformers(sourceRequire(TRANSFORMERS_PACKAGE) as TransformersRuntime);
-		const runtimeDir = await ensureRuntimeInstalled({
-			runtimeDir: getTinyTitleRuntimeDir(),
-			install: {
-				dependencies: { [TRANSFORMERS_PACKAGE]: getTransformersVersionSpec() },
-				trustedDependencies: ["onnxruntime-node"],
-			},
-			probePackage: TRANSFORMERS_PACKAGE,
-			onPhase: phase => sendRuntimeInstallProgress(transport, requestId, modelKey, phase),
-		});
-		const entry = await prepareCompiledRuntime(runtimeDir);
-		const require_ = createRequire(entry);
-		return configureTransformers(require_(entry) as TransformersRuntime);
-	})().catch(error => {
-		transformersRuntime = null;
-		throw error;
-	});
-	return transformersRuntime;
-}
-
 function createStopOnTextCriteria(
 	transformers: TransformersRuntime,
 	tokenizer: TextGenerationPipeline["tokenizer"],
@@ -215,48 +105,6 @@ function createStopOnTextCriteria(
 	return new StopOnTextCriteria();
 }
 
-function toProgressEvent(modelKey: TinyLocalModelKey, info: ProgressInfo): TinyTitleProgressEvent {
-	if (info.status === "ready") {
-		return { modelKey, status: info.status, task: info.task, model: info.model };
-	}
-	if (info.status === "progress_total") {
-		return {
-			modelKey,
-			status: info.status,
-			name: info.name,
-			progress: info.progress,
-			loaded: info.loaded,
-			total: info.total,
-			files: info.files,
-		};
-	}
-	if (info.status === "progress") {
-		return {
-			modelKey,
-			status: info.status,
-			name: info.name,
-			file: info.file,
-			progress: info.progress,
-			loaded: info.loaded,
-			total: info.total,
-		};
-	}
-	return { modelKey, status: info.status, name: info.name, file: info.file };
-}
-
-function sendProgress(
-	transport: TinyTitleTransport,
-	id: string,
-	modelKey: TinyLocalModelKey,
-	info: ProgressInfo,
-): void {
-	transport.send({ type: "progress", id, event: toProgressEvent(modelKey, info) });
-}
-
-function errorMessage(error: unknown): string {
-	return error instanceof Error ? error.message : String(error);
-}
-
 async function loadPipelineOnDevice(
 	transformers: TransformersRuntime,
 	spec: TinyTitleLocalModelSpec,
@@ -318,21 +166,16 @@ async function loadPipeline(
 	const spec = getTinyLocalModelSpec(modelKey);
 	if (!spec) throw new Error(`Unknown tiny local model: ${modelKey}`);
 	if (spec.unsupportedReason) throw new Error(`${modelKey} is unavailable: ${spec.unsupportedReason}`);
-	const cached = pipelines.get(modelKey);
-	if (cached) {
-		void cached
-			.then(() => {
-				transport.send({
-					type: "progress",
-					id: requestId,
-					event: { modelKey, status: "ready", task: "text-generation", model: spec.repo },
-				});
-			})
-			.catch(() => undefined);
-		return cached;
-	}
+	const cached = replayCachedReady(pipelines, modelKey, transport, requestId, "text-generation", spec.repo);
+	if (cached) return cached;
 
-	const transformers = await loadTransformers(transport, requestId, modelKey);
+	const transformers = await loadTransformersRuntime(
+		transformersRuntime,
+		transport,
+		requestId,
+		modelKey,
+		getTinyTitleRuntimeDir,
+	);
 	const startedAt = performance.now();
 	const loaded = loadPipelineWithDeviceFallback(transformers, spec, modelKey, transport, requestId).then(
 		({ generator, device }) => {
@@ -393,7 +236,13 @@ async function generateTitle(
 ): Promise<string | null> {
 	const generator = await loadPipeline(modelKey, transport, requestId);
 	const promptText = buildPrompt(generator, message, systemPrompt);
-	const transformers = await loadTransformers(transport, requestId, modelKey);
+	const transformers = await loadTransformersRuntime(
+		transformersRuntime,
+		transport,
+		requestId,
+		modelKey,
+		getTinyTitleRuntimeDir,
+	);
 	const output = (await generator(promptText, {
 		max_new_tokens: TITLE_MAX_NEW_TOKENS,
 		do_sample: false,

package/src/tools/eval-backends.ts

@@ -4,24 +4,31 @@ import type { ToolSession } from ".";
 export interface EvalBackendsAllowance {
 	python: boolean;
 	js: boolean;
+	ruby: boolean;
+	julia: boolean;
 }
 
-/** Read per-backend allowance from settings (defaults true). */
+/** Read per-backend allowance from settings (py/js default on; rb/jl opt-in, default off). */
 export function readEvalBackendsAllowance(session: ToolSession): EvalBackendsAllowance {
 	return {
 		python: session.settings.get("eval.py") ?? true,
 		js: session.settings.get("eval.js") ?? true,
+		ruby: session.settings.get("eval.rb") ?? false,
+		julia: session.settings.get("eval.jl") ?? false,
 	};
 }
 
 /**
- * Materialize the active eval backend allowance: PI_PY / PI_JS env flags
- * override the per-key settings; otherwise settings (defaults true) win.
+ * Materialize the active eval backend allowance: PI_PY / PI_JS / PI_RB / PI_JL
+ * env flags override the per-key settings; otherwise settings win (py/js default
+ * on, rb/jl default off).
  */
 export function resolveEvalBackends(session: ToolSession): EvalBackendsAllowance {
 	const settings = readEvalBackendsAllowance(session);
 	return {
 		python: $flag("PI_PY", settings.python),
 		js: $flag("PI_JS", settings.js),
+		ruby: $flag("PI_RB", settings.ruby),
+		julia: $flag("PI_JL", settings.julia),
 	};
 }

package/src/tools/eval-render.ts

@@ -2,11 +2,12 @@
  * TUI rendering for the eval tool.
  *
  * Split out from `eval.ts` so the renderer can be imported by `renderers.ts`
- * without dragging the eval *runtime* (JS/Python backends -> agent bridge ->
- * task executor -> sdk -> extension loader -> root barrel) into the renderer
- * module graph. That transitive chain re-enters `renderers.ts` while `eval.ts`
- * is still initializing, which previously crashed module load with a TDZ
- * `Cannot access 'evalToolRenderer' before initialization`.
+ * without dragging the eval *runtime* (JS/Python/Ruby/Julia backends ->
+ * agent bridge -> task executor -> sdk -> extension loader -> root barrel)
+ * into the renderer module graph. That transitive chain re-enters
+ * `renderers.ts` while `eval.ts` is still initializing, which previously
+ * crashed module load with a TDZ `Cannot access 'evalToolRenderer' before
+ * initialization`.
  */
 import type { Component } from "@oh-my-pi/pi-tui";
 import { Markdown, Text } from "@oh-my-pi/pi-tui";
@@ -41,8 +42,11 @@ import {
 } from "./render-utils";
 export const EVAL_DEFAULT_PREVIEW_LINES = 10;
 
-function languageForHighlighter(language: EvalLanguage | undefined): "python" | "javascript" {
-	return language === "js" ? "javascript" : "python";
+function languageForHighlighter(language: EvalLanguage | undefined): "python" | "javascript" | "ruby" | "julia" {
+	if (language === "js") return "javascript";
+	if (language === "ruby") return "ruby";
+	if (language === "julia") return "julia";
+	return "python";
 }
 
 interface EvalRenderCellArg {
@@ -70,7 +74,10 @@ interface EvalRenderCell {
 }
 
 function normalizeRenderLanguage(value: string | undefined): EvalLanguage {
-	return value === "js" ? "js" : "python";
+	if (value === "js") return "js";
+	if (value === "rb" || value === "ruby") return "ruby";
+	if (value === "jl" || value === "julia") return "julia";
+	return "python";
 }
 
 function getRenderCells(args: EvalRenderArgs | undefined): EvalRenderCell[] {
@@ -506,6 +513,7 @@ export const evalToolRenderer = {
 						{
 							code: cell.code,
 							language: languageForHighlighter(cell.language),
+							showLanguage: true,
 							index: i,
 							total: cells.length,
 							title: cell.title,
@@ -608,6 +616,7 @@ export const evalToolRenderer = {
 							{
 								code: cell.code,
 								language: languageForHighlighter(cell.language ?? details?.language),
+								showLanguage: true,
 								index: i,
 								total: cellResults.length,
 								title: cell.title,