import { describe, expect, test } from "vitest";
import {
  allFixtures,
  codingAgentFixture,
  contradictoryUpdatesFixture,
  mulberry32,
  toolSpamFixture,
} from "../fixtures";
import {
  checkViewStructure,
  gradeAnswer,
  latencyStats,
  reductionPercent,
} from "../metrics";
import { BenchRun, renderMarkdown, resultFileBase } from "../report";
import { planCompaction } from "../../src/plan";
import {
  CanonMessage,
  prefixHashes,
  roundBoundaries,
} from "../../src/view";

describe("mulberry32", () => {
  test("same seed yields the same sequence, different seeds differ", () => {
    const a1 = mulberry32(42);
    const a2 = mulberry32(42);
    const b = mulberry32(43);
    const seqA1 = [a1(), a1(), a1()];
    const seqA2 = [a2(), a2(), a2()];
    const seqB = [b(), b(), b()];
    expect(seqA1).toEqual(seqA2);
    expect(seqA1).not.toEqual(seqB);
    for (const v of seqA1) {
      expect(v).toBeGreaterThanOrEqual(0);
      expect(v).toBeLessThan(1);
    }
  });
});

describe("fixtures", () => {
  test("deterministic: same seed produces identical messages and hashes", () => {
    for (const [f1, f2] of allFixtures(7).map(
      (f, i) => [f, allFixtures(7)[i]] as const,
    )) {
      expect(f1.msgs).toEqual(f2.msgs);
      expect(prefixHashes(f1.msgs)).toEqual(prefixHashes(f2.msgs));
    }
  });

  test("every fixture has messages, facts, and starts with a user message", () => {
    for (const f of allFixtures()) {
      expect(f.msgs.length).toBeGreaterThan(10);
      expect(f.facts.length).toBeGreaterThanOrEqual(3);
      expect(f.msgs[0].role).toBe("user");
    }
  });

  test("tool results always directly follow their assistant tool call", () => {
    for (const f of allFixtures()) {
      f.msgs.forEach((m, i) => {
        if (m.role === "tool") {
          const prev = f.msgs[i - 1];
          const prevOk =
            prev.role === "tool" ||
            (prev.role === "assistant" && (prev.toolCalls ?? []).length > 0);
          expect(prevOk, `${f.name}[${i}]`).toBe(true);
        }
      });
    }
  });

  test("seeded facts appear verbatim in the transcript", () => {
    for (const f of allFixtures()) {
      const text = f.msgs
        .map(
          (m) =>
            m.text +
            (m.toolResults ?? []).map((r) => r.content).join(" "),
        )
        .join("\n");
      for (const fact of f.facts) {
        const found = fact.accepted.some((a) => text.includes(a));
        expect(found, `${f.name}: ${fact.id}`).toBe(true);
      }
    }
  });

  test("contradictory fixture accepts only the latest value", () => {
    const f = contradictoryUpdatesFixture();
    const revised = f.facts.find((fact) => fact.id.includes("revised"));
    expect(revised).toBeDefined();
  });

  test("tool spam fixture carries oversized tool results", () => {
    const f = toolSpamFixture();
    const biggest = Math.max(
      ...f.msgs.flatMap((m) =>
        (m.toolResults ?? []).map((r) => r.content.length),
      ),
    );
    expect(biggest).toBeGreaterThan(5000);
  });

  test("planCompaction cuts every fixture at valid boundaries", () => {
    for (const f of allFixtures()) {
      const tokens = f.msgs.map((m) => Math.ceil((m.text.length + 40) / 4) + 50);
      const boundaries = roundBoundaries(f.msgs);
      const plan = planCompaction({
        msgs: f.msgs,
        tokens,
        coveredUpTo: 0,
        summaryTokens: 0,
        estSummaryTokensPerChunk: 300,
        limit: 500,
        autoCompact: true,
        force: false,
        keepRecentTokens: 300,
        chunkTokens: 800,
      });
      expect(plan.cuts.length).toBeGreaterThan(0);
      for (const cut of plan.cuts) {
        expect(boundaries).toContain(cut);
      }
    }
  });
});

describe("gradeAnswer", () => {
  test("normalizes case and whitespace for substring matching", () => {
    expect(gradeAnswer("The codename is  VELVET-otter-19.", ["VELVET-OTTER-19"])).toBe(true);
    expect(gradeAnswer("no idea", ["VELVET-OTTER-19"])).toBe(false);
    expect(gradeAnswer("budget was $47,250 total", ["$47,250", "47250"])).toBe(true);
  });
});

describe("checkViewStructure", () => {
  const u = (text: string): CanonMessage => ({ role: "user", text });
  const a = (text: string): CanonMessage => ({ role: "assistant", text });

  test("passes a well-formed view", () => {
    const report = checkViewStructure("system+summary", [u("q"), a("r")]);
    expect(report.ok).toBe(true);
  });

  test("fails when the tail has no user message", () => {
    const report = checkViewStructure("sys", [a("r")]);
    expect(report.hasUserMessage).toBe(false);
    expect(report.ok).toBe(false);
  });

  test("fails on an orphan tool result at the tail start", () => {
    const tail: CanonMessage[] = [
      { role: "tool", text: "", toolResults: [{ content: "x" }] },
      u("q"),
    ];
    const report = checkViewStructure("sys", tail);
    expect(report.noOrphanToolResults).toBe(false);
    expect(report.ok).toBe(false);
  });

  test("fails on a system message inside the tail", () => {
    const tail: CanonMessage[] = [u("q"), { role: "system", text: "sneaky" }];
    expect(checkViewStructure("sys", tail).singleLeadingSystem).toBe(false);
  });
});

describe("reductionPercent and latencyStats", () => {
  test("reduction math", () => {
    expect(reductionPercent(1000, 250)).toBe(75);
    expect(reductionPercent(0, 0)).toBe(0);
  });

  test("latency stats", () => {
    const stats = latencyStats([100, 200, 300]);
    expect(stats.mean).toBe(200);
    expect(stats.p50).toBe(200);
    expect(stats.max).toBe(300);
  });
});

describe("report", () => {
  const run: BenchRun = {
    date: "2026-08-20",
    modelId: "qwen/qwen3.8-27b",
    settings: { chunkTokens: 800 },
    fixtures: [
      {
        fixture: "coding-agent",
        asked: 5,
        correctAfterCompaction: 5,
        correctAfterConsolidation: 4,
        reductionPercent: 82,
        latency: { mean: 1200, p50: 1100, max: 2000 },
        structural: {
          singleLeadingSystem: true,
          noOrphanToolResults: true,
          hasUserMessage: true,
          ok: true,
        },
        cacheDeterministic: true,
      },
    ],
  };

  test("markdown contains the essentials", () => {
    const md = renderMarkdown(run);
    expect(md).toContain("qwen/qwen3.8-27b");
    expect(md).toContain("coding-agent");
    expect(md).toContain("5/5");
    expect(md).toContain("82");
  });

  test("file base is filesystem-safe and stamped", () => {
    const base = resultFileBase(run);
    expect(base).toContain("2026-08-20");
    expect(base).not.toMatch(/[\/\\:]/);
  });
});

import { describe, expect, test } from "vitest";
import {
  allFixtures,
  codingAgentFixture,
  contradictoryUpdatesFixture,
  mulberry32,
  toolSpamFixture,
} from "../fixtures";
import {
  checkViewStructure,
  gradeAnswer,
  latencyStats,
  reductionPercent,
} from "../metrics";
import { BenchRun, renderMarkdown, resultFileBase } from "../report";
import { planCompaction } from "../../src/plan";
import {
  CanonMessage,
  prefixHashes,
  roundBoundaries,
} from "../../src/view";

describe("mulberry32", () => {
  test("same seed yields the same sequence, different seeds differ", () => {
    const a1 = mulberry32(42);
    const a2 = mulberry32(42);
    const b = mulberry32(43);
    const seqA1 = [a1(), a1(), a1()];
    const seqA2 = [a2(), a2(), a2()];
    const seqB = [b(), b(), b()];
    expect(seqA1).toEqual(seqA2);
    expect(seqA1).not.toEqual(seqB);
    for (const v of seqA1) {
      expect(v).toBeGreaterThanOrEqual(0);
      expect(v).toBeLessThan(1);
    }
  });
});

describe("fixtures", () => {
  test("deterministic: same seed produces identical messages and hashes", () => {
    for (const [f1, f2] of allFixtures(7).map(
      (f, i) => [f, allFixtures(7)[i]] as const,
    )) {
      expect(f1.msgs).toEqual(f2.msgs);
      expect(prefixHashes(f1.msgs)).toEqual(prefixHashes(f2.msgs));
    }
  });

  test("every fixture has messages, facts, and starts with a user message", () => {
    for (const f of allFixtures()) {
      expect(f.msgs.length).toBeGreaterThan(10);
      expect(f.facts.length).toBeGreaterThanOrEqual(3);
      expect(f.msgs[0].role).toBe("user");
    }
  });

  test("tool results always directly follow their assistant tool call", () => {
    for (const f of allFixtures()) {
      f.msgs.forEach((m, i) => {
        if (m.role === "tool") {
          const prev = f.msgs[i - 1];
          const prevOk =
            prev.role === "tool" ||
            (prev.role === "assistant" && (prev.toolCalls ?? []).length > 0);
          expect(prevOk, `${f.name}[${i}]`).toBe(true);
        }
      });
    }
  });

  test("seeded facts appear verbatim in the transcript", () => {
    for (const f of allFixtures()) {
      const text = f.msgs
        .map(
          (m) =>
            m.text +
            (m.toolResults ?? []).map((r) => r.content).join(" "),
        )
        .join("\n");
      for (const fact of f.facts) {
        const found = fact.accepted.some((a) => text.includes(a));
        expect(found, `${f.name}: ${fact.id}`).toBe(true);
      }
    }
  });

  test("contradictory fixture accepts only the latest value", () => {
    const f = contradictoryUpdatesFixture();
    const revised = f.facts.find((fact) => fact.id.includes("revised"));
    expect(revised).toBeDefined();
  });

  test("tool spam fixture carries oversized tool results", () => {
    const f = toolSpamFixture();
    const biggest = Math.max(
      ...f.msgs.flatMap((m) =>
        (m.toolResults ?? []).map((r) => r.content.length),
      ),
    );
    expect(biggest).toBeGreaterThan(5000);
  });

  test("planCompaction cuts every fixture at valid boundaries", () => {
    for (const f of allFixtures()) {
      const tokens = f.msgs.map((m) => Math.ceil((m.text.length + 40) / 4) + 50);
      const boundaries = roundBoundaries(f.msgs);
      const plan = planCompaction({
        msgs: f.msgs,
        tokens,
        coveredUpTo: 0,
        summaryTokens: 0,
        estSummaryTokensPerChunk: 300,
        limit: 500,
        autoCompact: true,
        force: false,
        keepRecentTokens: 300,
        chunkTokens: 800,
      });
      expect(plan.cuts.length).toBeGreaterThan(0);
      for (const cut of plan.cuts) {
        expect(boundaries).toContain(cut);
      }
    }
  });
});

describe("gradeAnswer", () => {
  test("normalizes case and whitespace for substring matching", () => {
    expect(gradeAnswer("The codename is  VELVET-otter-19.", ["VELVET-OTTER-19"])).toBe(true);
    expect(gradeAnswer("no idea", ["VELVET-OTTER-19"])).toBe(false);
    expect(gradeAnswer("budget was $47,250 total", ["$47,250", "47250"])).toBe(true);
  });
});

describe("checkViewStructure", () => {
  const u = (text: string): CanonMessage => ({ role: "user", text });
  const a = (text: string): CanonMessage => ({ role: "assistant", text });

  test("passes a well-formed view", () => {
    const report = checkViewStructure("system+summary", [u("q"), a("r")]);
    expect(report.ok).toBe(true);
  });

  test("fails when the tail has no user message", () => {
    const report = checkViewStructure("sys", [a("r")]);
    expect(report.hasUserMessage).toBe(false);
    expect(report.ok).toBe(false);
  });

  test("fails on an orphan tool result at the tail start", () => {
    const tail: CanonMessage[] = [
      { role: "tool", text: "", toolResults: [{ content: "x" }] },
      u("q"),
    ];
    const report = checkViewStructure("sys", tail);
    expect(report.noOrphanToolResults).toBe(false);
    expect(report.ok).toBe(false);
  });

  test("fails on a system message inside the tail", () => {
    const tail: CanonMessage[] = [u("q"), { role: "system", text: "sneaky" }];
    expect(checkViewStructure("sys", tail).singleLeadingSystem).toBe(false);
  });
});

describe("reductionPercent and latencyStats", () => {
  test("reduction math", () => {
    expect(reductionPercent(1000, 250)).toBe(75);
    expect(reductionPercent(0, 0)).toBe(0);
  });

  test("latency stats", () => {
    const stats = latencyStats([100, 200, 300]);
    expect(stats.mean).toBe(200);
    expect(stats.p50).toBe(200);
    expect(stats.max).toBe(300);
  });
});

describe("report", () => {
  const run: BenchRun = {
    date: "2026-08-20",
    modelId: "qwen/qwen3.8-27b",
    settings: { chunkTokens: 800 },
    fixtures: [
      {
        fixture: "coding-agent",
        asked: 5,
        correctAfterCompaction: 5,
        correctAfterConsolidation: 4,
        reductionPercent: 82,
        latency: { mean: 1200, p50: 1100, max: 2000 },
        structural: {
          singleLeadingSystem: true,
          noOrphanToolResults: true,
          hasUserMessage: true,
          ok: true,
        },
        cacheDeterministic: true,
      },
    ],
  };

  test("markdown contains the essentials", () => {
    const md = renderMarkdown(run);
    expect(md).toContain("qwen/qwen3.8-27b");
    expect(md).toContain("coding-agent");
    expect(md).toContain("5/5");
    expect(md).toContain("82");
  });

  test("file base is filesystem-safe and stamped", () => {
    const base = resultFileBase(run);
    expect(base).toContain("2026-08-20");
    expect(base).not.toMatch(/[\/\\:]/);
  });
});