@torrent-tv/proxy 2.87.10 → 2.87.12

package/test/move-cost.test.js

@@ -1,216 +1,216 @@
-/**
- * @file What moving a running encoder costs, and what the plan does while
- * nobody has measured it.
- *
- * Field 2026-09-08: 39 moves in one session, 24 of them between three adjacent
- * numbers — #58 to #59, #59 to #58, #58 to #60, #60 to #58, six times each,
- * about 0.8 s apart — while the viewer's picture stood still for 116.7 s in
- * three interruptions, the worst of them 91.8 s. The zone the viewer's own
- * position defines slides forward one number at a time, and every slide made
- * standing one number behind it score worse than standing in it.
- *
- * THREE FAULTS IN THE ARITHMETIC, and no threshold anywhere. There was one for
- * a while — "a move must beat staying by at least what moving costs" — and it
- * was a prop under a comparison that was wrong rather than indifferent. It is
- * gone.
- *
- * 1. **A body was charged a whole piece for the one it was already making.**
- *    `arrival = delay + (index - at + 1) / rate` is right for a body that does
- *    not exist yet and wrong for a run 0.8 s into a 0.9 s piece. `delaySec` is
- *    now when the body finishes the piece it STANDS ON, so from #58 reaching #59
- *    costs 0.14 + 0.94 = 1.08 s against 1.88 s for a kill and a cold start — a
- *    decision by eight hundred milliseconds, where the double charge had made it
- *    a coin flip lost by ten.
- * 2. **The piece was priced at the unpenalised rate** while arrivals used the
- *    penalised one, so every extra body looked cheaper than it is and the plan
- *    bought a second encoder where one served. One rate, the one in force.
- * 3. **`withinSeconds: null` was read through `Number()`**, where it is 0, so
- *    the film BEHIND the viewers was due immediately and was the most urgent
- *    material in the file. It bought encoders and it took the run standing in
- *    front of the viewer, because that run was the nearest body to it.
- *
- * And what a move costs is `Infinity` until something has been measured, because
- * a move is irreversible while leaving the encoder alone is always available. A
- * run killed before producing anything is a measurement too — a lower bound on
- * the first output — which is the only reading a thrash can supply.
- */
-
-import test from "node:test";
-import assert from "node:assert/strict";
-import { RunCosts } from "../services/encode/run-costs.js";
-import { planEncoders } from "../services/encode/EncodePlan.js";
-import { CoverageMap } from "../services/encode/CoverageMap.js";
-
-test("nothing measured is a plain zero, and the floor is derived where the arithmetic is", () => {
-  const costs = new RunCosts();
-
-  const { firstByteWaitSec, killCostSec } = costs.seconds();
-  assert.equal(firstByteWaitSec, 0, "no reading is said as none, not as a guess");
-  assert.equal(killCostSec, 0);
-  // There was an `Infinity` here — the cost of a move, made unaffordable until
-  // something had been measured, on the reasoning that an unmeasured price must
-  // not license an irreversible act. It was an exception in a model that needs
-  // none: a first piece cannot appear faster than it takes to ENCODE one, and
-  // how fast this host encodes is measured before any viewer exists, so the
-  // floor belongs where the arithmetic is.
-  assert.equal("moveCostSec" in costs.seconds(), false, "no such figure any more");
-});
-
-test("a run killed before producing anything is a lower bound on the first output", () => {
-  const costs = new RunCosts();
-
-  // Exactly what a thrash supplies: a run that lived 800 ms and finished
-  // nothing. It says the first output takes AT LEAST that long, which is a fact
-  // and the only reading a thrash can give — every run in one is killed before
-  // it produces.
-  costs.note({ livedMs: 800, dyingMs: 40 });
-
-  const { firstByteWaitSec, killCostSec } = costs.seconds();
-  assert.ok(Math.abs(firstByteWaitSec - 0.8) < 0.001, `got ${firstByteWaitSec}`);
-  assert.ok(Math.abs(killCostSec - 0.04) < 0.001, `got ${killCostSec}`);
-});
-
-test("a run that produced something is measured by its first output, not its life", () => {
-  const costs = new RunCosts();
-
-  costs.note({ livedMs: 60_000, firstOutputMs: 1260, dyingMs: 100 });
-
-  const { firstByteWaitSec } = costs.seconds();
-  assert.ok(Math.abs(firstByteWaitSec - 1.26) < 0.001, `got ${firstByteWaitSec}`);
-});
-
-/**
- * The map's real shape: one segment at the viewer, doubling zones ahead down to
- * p91, and everything behind them at p1 with no deadline. Written out because a
- * fixture of two zones is not this, and the difference decides the answer: with
- * nothing stated past the viewer's own zone, a run one segment behind it is
- * compared on that zone alone and loses by a tenth of a second.
- *
- * @param {number} head - The segment the viewer is on.
- * @param {number} count
- * @returns {object[]}
- */
-function mapAt(head, count) {
-  const zones = [];
-  if (head > 0) {
-    zones.push({ from: 0, to: head - 1, priority: 1, withinSeconds: null, behind: true });
-  }
-  let from = head;
-  let width = 1;
-  let rank = 100;
-  while (from < count && rank > 90) {
-    const to = Math.min(count - 1, from + width - 1);
-    zones.push({ from, to, priority: rank, withinSeconds: (from - head) * 4.2, behind: false });
-    from = to + 1;
-    width *= 2;
-    rank -= 1;
-  }
-  if (from < count) {
-    zones.push({ from, to: count - 1, priority: 90, withinSeconds: (from - head) * 4.2, behind: false });
-  }
-  return zones;
-}
-
-test("an encoder is left alone while the viewer is still at or before it", () => {
-  // Every slide of the viewer's zone used to make standing one number behind it
-  // score worse than standing in it — by ten milliseconds, which is nothing but
-  // the double charge for a piece already being made. What holds now is the
-  // narrower and true statement: while the viewer's own zone still contains the
-  // encoder's position, it is left alone. A viewer BEFORE it is a different
-  // case entirely and correctly moves it back — encoders only go forward, so
-  // one standing past a viewer never reaches them.
-  //
-  // Once the viewer has PASSED it, moving forward is correct and happens once: a
-  // run that has produced nothing in 0.8 s of a 1.26 s warm-up owes 0.46 s
-  // before its piece exists, while a fresh one at the viewer's own number owes
-  // 0.32 s of spawn and then the piece — so the viewer is served sooner, and the
-  // number left behind is in nobody's zone.
-  const coverage = new CoverageMap();
-  coverage.setSegmentCount(482);
-  const run = { from: 58, to: -1, head: 58, speedX: 4.45, isAlive: true, startedAt: 1_000_000 };
-  coverage.claim(run, 58, -1);
-
-  for (const viewerAt of [58]) {
-    const actions = planEncoders({
-      coverage,
-      windows: mapAt(viewerAt, 482),
-      runs: [run],
-      maxRuns: 3,
-      segmentSeconds: 4.2,
-      speedX: 4.45,
-      killCostSec: 0.04,
-      // Measured on the addon host: a run started at 15:50:15.521 and its first
-      // piece existed at 15:50:16.785.
-      firstByteWaitSec: 1.26,
-      refetchSecPerFilmSecond: 0,
-      // 1.98 at 1920x1080, measured: a second encoder takes very nearly all of
-      // the first's speed.
-      contentionPenaltyFor: (others) => (others <= 0 ? 1 : 1.98 ** others),
-      now: 1_000_000 + 800
-    });
-
-    assert.deepEqual(
-      actions.filter((one) => one.type === "move"),
-      [],
-      `the viewer at #${viewerAt} does not cost the encoder its place`
-    );
-  }
-
-  // And once they are past it, exactly one move — not one per slide.
-  const past = [59, 60, 61].map((viewerAt) => planEncoders({
-    coverage,
-    windows: mapAt(viewerAt, 482),
-    runs: [run],
-    maxRuns: 3,
-    segmentSeconds: 4.2,
-    speedX: 4.45,
-    killCostSec: 0.04,
-    firstByteWaitSec: 1.26,
-    refetchSecPerFilmSecond: 0,
-    contentionPenaltyFor: (others) => (others <= 0 ? 1 : 1.98 ** others),
-    now: 1_000_000 + 800
-  }).filter((one) => one.type === "move"));
-
-  assert.deepEqual(
-    past.map((moves) => moves.length),
-    [1, 1, 1],
-    "one move to where the viewer now is, whichever number that is"
-  );
-  assert.deepEqual(
-    past.map((moves) => moves[0].from),
-    [59, 60, 61],
-    "and it goes to the viewer's own number, not one past it"
-  );
-});
-
-test("a move that genuinely saves the viewer time still happens", () => {
-  // The other half: the viewer jumped fourteen segments ahead, and driving there
-  // at 4.45x would take 13 s while a cold start takes 0.94 s. Refusing this
-  // would be the opposite fault.
-  const coverage = new CoverageMap();
-  coverage.setSegmentCount(482);
-  const run = { from: 0, to: 481, head: 44, speedX: 4.45, isAlive: true };
-  coverage.claim(run, 0, 481);
-
-  const actions = planEncoders({
-    coverage,
-    windows: [{ from: 58, to: 59, priority: 100, withinSeconds: 0, behind: false }],
-    runs: [run],
-    maxRuns: 3,
-    segmentSeconds: 4.2,
-    speedX: 4.45,
-    killCostSec: 0.04,
-    firstByteWaitSec: 1.26,
-    refetchSecPerFilmSecond: 0,
-    contentionPenaltyFor: () => 1
-  });
-
-  // WHERE the encoder ends up, not how it got there: the run at #44 reaches
-  // nothing anybody waits for, so the plan may either take it to #58 or stop it
-  // and start one there. Both are one process at #58, and which is cheaper is
-  // the measured difference between a kill and a cold start.
-  const placed = actions
-    .filter((one) => one.type === "move" || one.type === "start")
-    .map((one) => one.from);
-  assert.deepEqual(placed, [58], "fourteen segments of driving is worth a cold start");
-});
+/**
+ * @file What moving a running encoder costs, and what the plan does while
+ * nobody has measured it.
+ *
+ * Field 2026-09-08: 39 moves in one session, 24 of them between three adjacent
+ * numbers — #58 to #59, #59 to #58, #58 to #60, #60 to #58, six times each,
+ * about 0.8 s apart — while the viewer's picture stood still for 116.7 s in
+ * three interruptions, the worst of them 91.8 s. The zone the viewer's own
+ * position defines slides forward one number at a time, and every slide made
+ * standing one number behind it score worse than standing in it.
+ *
+ * THREE FAULTS IN THE ARITHMETIC, and no threshold anywhere. There was one for
+ * a while — "a move must beat staying by at least what moving costs" — and it
+ * was a prop under a comparison that was wrong rather than indifferent. It is
+ * gone.
+ *
+ * 1. **A body was charged a whole piece for the one it was already making.**
+ *    `arrival = delay + (index - at + 1) / rate` is right for a body that does
+ *    not exist yet and wrong for a run 0.8 s into a 0.9 s piece. `delaySec` is
+ *    now when the body finishes the piece it STANDS ON, so from #58 reaching #59
+ *    costs 0.14 + 0.94 = 1.08 s against 1.88 s for a kill and a cold start — a
+ *    decision by eight hundred milliseconds, where the double charge had made it
+ *    a coin flip lost by ten.
+ * 2. **The piece was priced at the unpenalised rate** while arrivals used the
+ *    penalised one, so every extra body looked cheaper than it is and the plan
+ *    bought a second encoder where one served. One rate, the one in force.
+ * 3. **`withinSeconds: null` was read through `Number()`**, where it is 0, so
+ *    the film BEHIND the viewers was due immediately and was the most urgent
+ *    material in the file. It bought encoders and it took the run standing in
+ *    front of the viewer, because that run was the nearest body to it.
+ *
+ * And what a move costs is `Infinity` until something has been measured, because
+ * a move is irreversible while leaving the encoder alone is always available. A
+ * run killed before producing anything is a measurement too — a lower bound on
+ * the first output — which is the only reading a thrash can supply.
+ */
+
+import test from "node:test";
+import assert from "node:assert/strict";
+import { RunCosts } from "../services/encode/run-costs.js";
+import { planEncoders } from "../services/encode/EncodePlan.js";
+import { CoverageMap } from "../services/encode/CoverageMap.js";
+
+test("nothing measured is a plain zero, and the floor is derived where the arithmetic is", () => {
+  const costs = new RunCosts();
+
+  const { firstByteWaitSec, killCostSec } = costs.seconds();
+  assert.equal(firstByteWaitSec, 0, "no reading is said as none, not as a guess");
+  assert.equal(killCostSec, 0);
+  // There was an `Infinity` here — the cost of a move, made unaffordable until
+  // something had been measured, on the reasoning that an unmeasured price must
+  // not license an irreversible act. It was an exception in a model that needs
+  // none: a first piece cannot appear faster than it takes to ENCODE one, and
+  // how fast this host encodes is measured before any viewer exists, so the
+  // floor belongs where the arithmetic is.
+  assert.equal("moveCostSec" in costs.seconds(), false, "no such figure any more");
+});
+
+test("a run killed before producing anything is a lower bound on the first output", () => {
+  const costs = new RunCosts();
+
+  // Exactly what a thrash supplies: a run that lived 800 ms and finished
+  // nothing. It says the first output takes AT LEAST that long, which is a fact
+  // and the only reading a thrash can give — every run in one is killed before
+  // it produces.
+  costs.note({ livedMs: 800, dyingMs: 40 });
+
+  const { firstByteWaitSec, killCostSec } = costs.seconds();
+  assert.ok(Math.abs(firstByteWaitSec - 0.8) < 0.001, `got ${firstByteWaitSec}`);
+  assert.ok(Math.abs(killCostSec - 0.04) < 0.001, `got ${killCostSec}`);
+});
+
+test("a run that produced something is measured by its first output, not its life", () => {
+  const costs = new RunCosts();
+
+  costs.note({ livedMs: 60_000, firstOutputMs: 1260, dyingMs: 100 });
+
+  const { firstByteWaitSec } = costs.seconds();
+  assert.ok(Math.abs(firstByteWaitSec - 1.26) < 0.001, `got ${firstByteWaitSec}`);
+});
+
+/**
+ * The map's real shape: one segment at the viewer, doubling zones ahead down to
+ * p91, and everything behind them at p1 with no deadline. Written out because a
+ * fixture of two zones is not this, and the difference decides the answer: with
+ * nothing stated past the viewer's own zone, a run one segment behind it is
+ * compared on that zone alone and loses by a tenth of a second.
+ *
+ * @param {number} head - The segment the viewer is on.
+ * @param {number} count
+ * @returns {object[]}
+ */
+function mapAt(head, count) {
+  const zones = [];
+  if (head > 0) {
+    zones.push({ from: 0, to: head - 1, priority: 1, withinSeconds: null, behind: true });
+  }
+  let from = head;
+  let width = 1;
+  let rank = 100;
+  while (from < count && rank > 90) {
+    const to = Math.min(count - 1, from + width - 1);
+    zones.push({ from, to, priority: rank, withinSeconds: (from - head) * 4.2, behind: false });
+    from = to + 1;
+    width *= 2;
+    rank -= 1;
+  }
+  if (from < count) {
+    zones.push({ from, to: count - 1, priority: 90, withinSeconds: (from - head) * 4.2, behind: false });
+  }
+  return zones;
+}
+
+test("an encoder is left alone while the viewer is still at or before it", () => {
+  // Every slide of the viewer's zone used to make standing one number behind it
+  // score worse than standing in it — by ten milliseconds, which is nothing but
+  // the double charge for a piece already being made. What holds now is the
+  // narrower and true statement: while the viewer's own zone still contains the
+  // encoder's position, it is left alone. A viewer BEFORE it is a different
+  // case entirely and correctly moves it back — encoders only go forward, so
+  // one standing past a viewer never reaches them.
+  //
+  // Once the viewer has PASSED it, moving forward is correct and happens once: a
+  // run that has produced nothing in 0.8 s of a 1.26 s warm-up owes 0.46 s
+  // before its piece exists, while a fresh one at the viewer's own number owes
+  // 0.32 s of spawn and then the piece — so the viewer is served sooner, and the
+  // number left behind is in nobody's zone.
+  const coverage = new CoverageMap();
+  coverage.setSegmentCount(482);
+  const run = { from: 58, to: -1, head: 58, speedX: 4.45, isAlive: true, startedAt: 1_000_000 };
+  coverage.claim(run, 58, -1);
+
+  for (const viewerAt of [58]) {
+    const actions = planEncoders({
+      coverage,
+      windows: mapAt(viewerAt, 482),
+      runs: [run],
+      maxRuns: 3,
+      segmentSeconds: 4.2,
+      speedX: 4.45,
+      killCostSec: 0.04,
+      // Measured on the addon host: a run started at 15:50:15.521 and its first
+      // piece existed at 15:50:16.785.
+      firstByteWaitSec: 1.26,
+      refetchSecPerFilmSecond: 0,
+      // 1.98 at 1920x1080, measured: a second encoder takes very nearly all of
+      // the first's speed.
+      contentionPenaltyFor: (others) => (others <= 0 ? 1 : 1.98 ** others),
+      now: 1_000_000 + 800
+    });
+
+    assert.deepEqual(
+      actions.filter((one) => one.type === "move"),
+      [],
+      `the viewer at #${viewerAt} does not cost the encoder its place`
+    );
+  }
+
+  // And once they are past it, exactly one move — not one per slide.
+  const past = [59, 60, 61].map((viewerAt) => planEncoders({
+    coverage,
+    windows: mapAt(viewerAt, 482),
+    runs: [run],
+    maxRuns: 3,
+    segmentSeconds: 4.2,
+    speedX: 4.45,
+    killCostSec: 0.04,
+    firstByteWaitSec: 1.26,
+    refetchSecPerFilmSecond: 0,
+    contentionPenaltyFor: (others) => (others <= 0 ? 1 : 1.98 ** others),
+    now: 1_000_000 + 800
+  }).filter((one) => one.type === "move"));
+
+  assert.deepEqual(
+    past.map((moves) => moves.length),
+    [1, 1, 1],
+    "one move to where the viewer now is, whichever number that is"
+  );
+  assert.deepEqual(
+    past.map((moves) => moves[0].from),
+    [59, 60, 61],
+    "and it goes to the viewer's own number, not one past it"
+  );
+});
+
+test("a move that genuinely saves the viewer time still happens", () => {
+  // The other half: the viewer jumped fourteen segments ahead, and driving there
+  // at 4.45x would take 13 s while a cold start takes 0.94 s. Refusing this
+  // would be the opposite fault.
+  const coverage = new CoverageMap();
+  coverage.setSegmentCount(482);
+  const run = { from: 0, to: 481, head: 44, speedX: 4.45, isAlive: true };
+  coverage.claim(run, 0, 481);
+
+  const actions = planEncoders({
+    coverage,
+    windows: [{ from: 58, to: 59, priority: 100, withinSeconds: 0, behind: false }],
+    runs: [run],
+    maxRuns: 3,
+    segmentSeconds: 4.2,
+    speedX: 4.45,
+    killCostSec: 0.04,
+    firstByteWaitSec: 1.26,
+    refetchSecPerFilmSecond: 0,
+    contentionPenaltyFor: () => 1
+  });
+
+  // WHERE the encoder ends up, not how it got there: the run at #44 reaches
+  // nothing anybody waits for, so the plan may either take it to #58 or stop it
+  // and start one there. Both are one process at #58, and which is cheaper is
+  // the measured difference between a kill and a cold start.
+  const placed = actions
+    .filter((one) => one.type === "move" || one.type === "start")
+    .map((one) => one.from);
+  assert.deepEqual(placed, [58], "fourteen segments of driving is worth a cold start");
+});