import { afterEach, describe, expect, it, vi } from 'vitest'

vi.mock('../../server/gateway-capabilities', () => ({
  BEARER_TOKEN: '',
  CLAUDE_API: 'http://127.0.0.1:8642',
  dashboardFetch: vi.fn(),
  ensureGatewayProbed: vi.fn(async () => ({ dashboard: { available: false } })),
  getCapabilities: vi.fn(() => ({ dashboard: { available: false } })),
}))

vi.mock('../../server/claude-api', () => ({
  listSessions: vi.fn(async () => []),
}))

vi.mock('../../server/local-session-store', () => ({
  getLocalMessages: vi.fn(() => []),
  getLocalSession: vi.fn(() => null),
}))

import {
  dashboardFetch,
  ensureGatewayProbed,
  getCapabilities,
} from '../../server/gateway-capabilities'
import { listSessions } from '../../server/claude-api'
import { getLocalMessages, getLocalSession } from '../../server/local-session-store'
import {
  estimateContextTokensFromCacheRead,
  estimateContextTokensFromMessages,
  readContextUsage,
} from '../../server/context-usage'

afterEach(() => {
  vi.unstubAllGlobals()
  vi.clearAllMocks()
})

describe('context usage estimation', () => {
  it('prefers live gateway runtime snapshots when the vanilla runtime endpoint is available', async () => {
    const fetchMock = vi.fn(async () =>
      new Response(
        JSON.stringify({
          model: 'anthropic/claude-sonnet-4-20250514',
          context_tokens: 4321,
          context_length: 200000,
          context_percent: 2,
          prompt_tokens: 111,
          completion_tokens: 22,
          total_tokens: 133,
        }),
        { status: 200, headers: { 'Content-Type': 'application/json' } },
      ),
    )
    vi.stubGlobal('fetch', fetchMock)

    const snapshot = await readContextUsage('session-123')

    expect(fetchMock).toHaveBeenCalledWith(
      'http://127.0.0.1:8642/api/sessions/session-123/runtime',
      expect.objectContaining({ headers: {}, signal: expect.any(AbortSignal) }),
    )
    expect(snapshot).toMatchObject({
      ok: true,
      model: 'anthropic/claude-sonnet-4-20250514',
      usedTokens: 4321,
      maxTokens: 200000,
      contextPercent: 2,
      conversationTokens: 4321,
    })
  })

  it('resolves main to the canonical Hermes session id before reading runtime usage', async () => {
    vi.mocked(listSessions).mockResolvedValue([
      {
        id: 'session-abc',
        title: 'Live chat',
        message_count: 12,
      },
    ] as any)

    const fetchMock = vi.fn(async (input: RequestInfo | URL) => {
      const url = String(input)
      if (url.includes('/api/sessions/session-abc/runtime')) {
        return new Response(
          JSON.stringify({
            model: 'gpt-5.4',
            context_tokens: 86397,
            context_length: 512000,
            context_percent: 17,
          }),
          { status: 200, headers: { 'Content-Type': 'application/json' } },
        )
      }
      return new Response('not found', { status: 404 })
    })
    vi.stubGlobal('fetch', fetchMock)

    const snapshot = await readContextUsage('main')

    expect(listSessions).toHaveBeenCalledWith(30, 0)
    expect(fetchMock).toHaveBeenCalledWith(
      'http://127.0.0.1:8642/api/sessions/session-abc/runtime',
      expect.objectContaining({ headers: {}, signal: expect.any(AbortSignal) }),
    )
    expect(snapshot).toMatchObject({
      ok: true,
      model: 'gpt-5.4',
      usedTokens: 86397,
      maxTokens: 512000,
      contextPercent: 17,
    })
  })

  it('prefers configured dashboard context length for local Workspace-only chats', async () => {
    vi.mocked(getLocalSession).mockReturnValue({ id: 'local-1', model: null } as any)
    vi.mocked(getLocalMessages).mockReturnValue([
      { content: 'x'.repeat(700) },
    ] as any)
    vi.mocked(getCapabilities).mockReturnValue({
      dashboard: { available: true },
    } as any)
    vi.mocked(ensureGatewayProbed).mockResolvedValue({
      dashboard: { available: true },
    } as any)

    const fetchMock = vi.fn(async (input: RequestInfo | URL) => {
      const url = String(input)
      if (url.includes('/api/sessions/local-1/runtime')) {
        return new Response('not found', { status: 404 })
      }
      return new Response('unexpected', { status: 500 })
    })
    vi.stubGlobal('fetch', fetchMock)
    vi.mocked(dashboardFetch).mockResolvedValue(
      new Response(
        JSON.stringify({
          model: 'gpt-5.4',
          effective_context_length: 512000,
          config_context_length: 512000,
        }),
        { status: 200, headers: { 'Content-Type': 'application/json' } },
      ) as any,
    )

    const snapshot = await readContextUsage('local-1')

    expect(snapshot).toMatchObject({
      ok: true,
      model: 'gpt-5.4',
      maxTokens: 512000,
      usedTokens: 200,
      contextPercent: 0,
    })
    expect(dashboardFetch).toHaveBeenCalledWith(
      '/api/model/info',
      expect.objectContaining({ signal: expect.any(AbortSignal) }),
    )
  })

  it('maps a mirrored local chat to the nearest real Hermes runtime session even when the runtime session has zero stored messages', async () => {
    vi.mocked(getLocalSession).mockReturnValue({
      id: 'local-mirror',
      model: null,
      createdAt: 1_000_000,
      updatedAt: 2_000_000,
    } as any)
    vi.mocked(getLocalMessages).mockReturnValue([
      { role: 'user', content: 'hello' },
      { role: 'assistant', content: 'world' },
    ] as any)

    vi.mocked(listSessions).mockResolvedValue([
      {
        id: 'runtime-nearest',
        started_at: 2000.01,
        last_active: 2100,
        message_count: 0,
      },
      {
        id: 'runtime-older',
        started_at: 1800,
        last_active: 1900,
        message_count: 12,
      },
    ] as any)

    const fetchMock = vi.fn(async (input: RequestInfo | URL) => {
      const url = String(input)
      if (url.includes('/api/sessions/local-mirror/runtime')) {
        return new Response('not found', { status: 404 })
      }
      if (url.includes('/api/sessions/runtime-nearest/runtime')) {
        return new Response(
          JSON.stringify({
            model: 'gpt-5.4',
            context_tokens: 85028,
            context_length: 512000,
            context_percent: 17,
          }),
          { status: 200, headers: { 'Content-Type': 'application/json' } },
        )
      }
      if (url.includes('/api/sessions/runtime-older/runtime')) {
        return new Response(
          JSON.stringify({
            model: 'gpt-5.4',
            context_tokens: 19266,
            context_length: 512000,
            context_percent: 4,
          }),
          { status: 200, headers: { 'Content-Type': 'application/json' } },
        )
      }
      return new Response('not found', { status: 404 })
    })
    vi.stubGlobal('fetch', fetchMock)

    const snapshot = await readContextUsage('local-mirror')

    expect(snapshot).toMatchObject({
      ok: true,
      model: 'gpt-5.4',
      usedTokens: 85028,
      maxTokens: 512000,
      contextPercent: 17,
    })
  })

  it('keeps configured context length visible for unresolved synthetic sessions like new', async () => {
    vi.mocked(getLocalSession).mockReturnValue(null as any)
    vi.mocked(getLocalMessages).mockReturnValue([] as any)
    vi.mocked(getCapabilities).mockReturnValue({
      dashboard: { available: true },
    } as any)
    vi.mocked(ensureGatewayProbed).mockResolvedValue({
      dashboard: { available: true },
    } as any)

    const fetchMock = vi.fn(async () => new Response('not found', { status: 404 }))
    vi.stubGlobal('fetch', fetchMock)
    vi.mocked(dashboardFetch).mockResolvedValue(
      new Response(
        JSON.stringify({
          model: 'gpt-5.4',
          effective_context_length: 512000,
          config_context_length: 512000,
        }),
        { status: 200, headers: { 'Content-Type': 'application/json' } },
      ) as any,
    )

    const snapshot = await readContextUsage('new')

    expect(snapshot).toMatchObject({
      ok: true,
      model: 'gpt-5.4',
      maxTokens: 512000,
      usedTokens: 0,
      contextPercent: 0,
    })
  })

  it('counts serialized content arrays and tool results instead of only string lengths', () => {
    const tokens = estimateContextTokensFromMessages([
      {
        content: [{ type: 'text', text: 'hello world' }],
      },
      {
        content: [
          {
            type: 'tool_result',
            text: 'x'.repeat(400),
          },
        ],
      },
    ])

    expect(tokens).toBeGreaterThan(100)
  })

  it('uses structured message estimation for local sessions instead of string-only content lengths', async () => {
    vi.mocked(getLocalSession).mockReturnValue({ id: 'local-structured', model: null } as any)
    vi.mocked(getLocalMessages).mockReturnValue([
      {
        content: [{ type: 'tool_result', text: 'x'.repeat(400) }],
      },
    ] as any)
    vi.mocked(getCapabilities).mockReturnValue({
      dashboard: { available: true },
    } as any)
    vi.mocked(ensureGatewayProbed).mockResolvedValue({
      dashboard: { available: true },
    } as any)

    const fetchMock = vi.fn(async () => new Response('not found', { status: 404 }))
    vi.stubGlobal('fetch', fetchMock)
    vi.mocked(dashboardFetch).mockResolvedValue(
      new Response(
        JSON.stringify({
          model: 'gpt-5.4',
          effective_context_length: 512000,
          config_context_length: 512000,
        }),
        { status: 200, headers: { 'Content-Type': 'application/json' } },
      ) as any,
    )

    const snapshot = await readContextUsage('local-structured')

    expect(snapshot.usedTokens).toBeGreaterThan(0)
    expect(snapshot.maxTokens).toBe(512000)
  })

  it('does not double-count top-level text when it mirrors structured content', () => {
    const mirroredToolOutput = JSON.stringify({ output: 'x'.repeat(4000) })
    const withMirroredText = estimateContextTokensFromMessages([
      {
        content: [{ type: 'tool_result', text: mirroredToolOutput }],
        text: mirroredToolOutput,
      },
    ])
    const contentOnly = estimateContextTokensFromMessages([
      {
        content: [{ type: 'tool_result', text: mirroredToolOutput }],
      },
    ])

    expect(withMirroredText).toBe(contentOnly)
  })

  it('keeps cumulative cache-read totals as a fallback, not the primary estimate', () => {
    const messageEstimate = estimateContextTokensFromMessages([
      { content: 'x'.repeat(4_000) },
    ])
    const cacheEstimate = estimateContextTokensFromCacheRead(14_100_480, 123)

    expect(messageEstimate).toBeLessThan(cacheEstimate)
    expect(messageEstimate).toBeGreaterThan(1000)
    expect(messageEstimate).toBeLessThan(1200)
  })
})