@juvio15/pi-cliproxyapi-provider 0.0.0-stage → 0.1.1

package/src/extension/tps.ts

@@ -0,0 +1,215 @@
+import type {
+  ExtensionAPI,
+  ExtensionContext,
+} from "@earendil-works/pi-coding-agent";
+
+import type { PauseController } from "./pause.ts";
+
+const STATUS_KEY = "tps";
+const REFRESH_INTERVAL_MS = 1000;
+
+const primaryUi = (ctx: ExtensionContext): boolean =>
+  ctx.hasUI && ctx.mode === "tui";
+
+const formatElapsed = (totalSeconds: number): string => {
+  const safe = Math.max(0, Math.floor(totalSeconds));
+
+  const units = [
+    { suffix: "d", value: Math.floor(safe / 86_400) },
+    { suffix: "h", value: Math.floor((safe % 86_400) / 3600) },
+    { suffix: "m", value: Math.floor((safe % 3600) / 60) },
+    { suffix: "s", value: safe % 60 },
+  ];
+
+  const first = units.findIndex(
+    (unit, index) => unit.value > 0 || index === units.length - 1
+  );
+
+  return units
+    .slice(first)
+    .map((unit) => `${unit.value}${unit.suffix}`)
+    .join(" ");
+};
+
+export interface TpsUsage {
+  cacheRead: number;
+  cacheWrite: number;
+  input: number;
+  output: number;
+  total: number;
+}
+
+const PERCENT = 100;
+const MS_PER_SECOND = 1000;
+
+const count = (value: number): string => value.toLocaleString("en-US");
+
+const seconds = (ms: number): string => (ms / MS_PER_SECOND).toFixed(1);
+
+/** The run summary: output speed over generation time, run length, and cache hit rate. */
+export const formatTpsSummary = (
+  usage: TpsUsage,
+  generationMs: number,
+  wallMs: number
+): string => {
+  const tps =
+    generationMs > 0
+      ? (usage.output / (generationMs / MS_PER_SECOND)).toFixed(1)
+      : "--";
+
+  const prompt = usage.input + usage.cacheRead;
+
+  const cacheHit =
+    prompt > 0 ? `${Math.round((usage.cacheRead / prompt) * PERCENT)}%` : "--";
+
+  return `TPS ${tps} tok/s (${seconds(generationMs)}s generating of ${seconds(wallMs)}s). out ${count(usage.output)}, in ${count(usage.input)}, cache hit ${cacheHit} (r/w ${count(usage.cacheRead)}/${count(usage.cacheWrite)}), total ${count(usage.total)}`;
+};
+
+const emptyUsage = (): TpsUsage => ({
+  cacheRead: 0,
+  cacheWrite: 0,
+  input: 0,
+  output: 0,
+  total: 0,
+});
+
+const isProviderAssistant = (
+  message: { role: string; provider?: unknown },
+  providerId: string
+): boolean => message.role === "assistant" && message.provider === providerId;
+
+export const registerTpsFooter = (
+  pi: ExtensionAPI,
+  pause: PauseController,
+  providerId: string
+): void => {
+  let startedAt: number | undefined;
+  let pausedAtStart = 0;
+  let pauseAtStart = false;
+  let timer: ReturnType<typeof setInterval> | undefined;
+  let statusContext: ExtensionContext | undefined;
+  let usage = emptyUsage();
+  let generationMs = 0;
+  let generationStartedAt: number | undefined;
+
+  const clearTimer = (): void => {
+    if (timer) {
+      clearInterval(timer);
+      timer = undefined;
+    }
+  };
+
+  const elapsed = (): number => {
+    if (startedAt === undefined) {
+      return 0;
+    }
+
+    return pauseAtStart
+      ? pause.getElapsed(startedAt, pausedAtStart)
+      : Math.max(0, Date.now() - startedAt);
+  };
+
+  const showElapsed = (ctx: ExtensionContext): void => {
+    if (primaryUi(ctx)) {
+      ctx.ui.setStatus(
+        STATUS_KEY,
+        ctx.ui.theme.fg("dim", `Elapsed ${formatElapsed(elapsed() / 1000)}`)
+      );
+    }
+  };
+
+  pi.on("before_agent_start", (_event, ctx) => {
+    if (!primaryUi(ctx)) {
+      return;
+    }
+
+    statusContext = ctx;
+
+    if (startedAt !== undefined) {
+      return;
+    }
+
+    startedAt = Date.now();
+    pauseAtStart = pause.isEnabled();
+    pausedAtStart = pause.getPausedDuration(startedAt);
+    usage = emptyUsage();
+    generationMs = 0;
+    generationStartedAt = undefined;
+    showElapsed(ctx);
+    clearTimer();
+    timer = setInterval(() => {
+      if (statusContext) {
+        showElapsed(statusContext);
+      }
+    }, REFRESH_INTERVAL_MS);
+    timer.unref();
+  });
+
+  pi.on("message_start", (event) => {
+    if (
+      startedAt !== undefined &&
+      isProviderAssistant(event.message, providerId)
+    ) {
+      generationStartedAt = Date.now();
+    }
+  });
+
+  pi.on("message_end", (event) => {
+    if (
+      generationStartedAt === undefined ||
+      !isProviderAssistant(event.message, providerId)
+    ) {
+      return;
+    }
+
+    generationMs += Math.max(0, Date.now() - generationStartedAt);
+    generationStartedAt = undefined;
+  });
+
+  pi.on("agent_end", (event, ctx) => {
+    if (startedAt === undefined || !primaryUi(ctx)) {
+      return;
+    }
+
+    for (const message of event.messages) {
+      if (message.role !== "assistant" || message.provider !== providerId) {
+        continue;
+      }
+
+      usage.input += message.usage.input;
+      usage.output += message.usage.output;
+      usage.cacheRead += message.usage.cacheRead;
+      usage.cacheWrite += message.usage.cacheWrite;
+      usage.total += message.usage.totalTokens;
+    }
+  });
+
+  pi.on("agent_settled", (_event, ctx) => {
+    if (startedAt === undefined || !primaryUi(ctx)) {
+      return;
+    }
+
+    const elapsedMs = elapsed();
+    showElapsed(ctx);
+    startedAt = undefined;
+    clearTimer();
+    statusContext = ctx;
+
+    if (elapsedMs <= 0) {
+      return;
+    }
+
+    ctx.ui.notify(formatTpsSummary(usage, generationMs, elapsedMs), "info");
+  });
+
+  pi.on("session_shutdown", (_event, ctx) => {
+    clearTimer();
+
+    if (primaryUi(ctx)) {
+      ctx.ui.setStatus(STATUS_KEY, undefined);
+    }
+
+    startedAt = undefined;
+    statusContext = undefined;
+  });
+};