agent-management/frontend/src/utils/ioInference.js

/**
 * 工作流节点 IO 变量推断引擎
 *
 * 职责：
 * 1. 从节点数据中提取 inputs / outputs 定义
 * 2. 连线时自动匹配源节点输出 → 目标节点输入
 * 3. 多源汇聚时计算变量分配
 */

// ========== 类型常量 ==========

export const IO_TYPES = [
  { label: '字符串', value: 'string' },
  { label: '数字', value: 'number' },
  { label: '布尔值', value: 'boolean' },
  { label: '数组', value: 'array' },
  { label: '对象', value: 'object' },
  { label: '文件路径', value: 'filePath' },
  { label: '目录路径', value: 'directoryPath' }
]

// 边状态 → 样式
export const EDGE_STATUS_STYLE = {
  ok:       { stroke: '#22c55e', strokeWidth: 2 },
  unmapped: { stroke: '#666',    strokeWidth: 2 },
  partial:  { stroke: '#f59e0b', strokeWidth: 2 },
  mismatch: { stroke: '#ef4444', strokeWidth: 2 }
}

// ========== 模板变量提取 ==========

/**
 * 从 {{变量名}} 模板中提取变量名列表（去重）
 */
export function extractTemplateVariables(text) {
  if (!text) return []
  const matches = text.matchAll(/\{\{\s*([\w.]+)\s*\}\}/g)
  const seen = new Set()
  const result = []
  for (const m of matches) {
    if (!seen.has(m[1])) {
      seen.add(m[1])
      result.push(m[1])
    }
  }
  return result
}

// ========== 节点 IO 获取 ==========

/**
 * 获取节点的输出字段列表
 */
export function getNodeOutputs(node) {
  const d = node.data || {}
  switch (node.type) {
    case 'userInput':
      return (d.variables || []).map(v => ({
        name: v.name,
        label: v.label || v.name,
        type: v.type || 'string',
        description: v.description || ''
      }))
    case 'llm':
      if (d.outputs && d.outputs.length) return d.outputs
      return [{ name: 'result', label: 'LLM 输出', type: 'string', description: '大模型响应文本' }]
    case 'agent':
      return d.outputs || []
    case 'skill':
      return d.outputs || []
    case 'knowledgeRetrieval':
      if (d.outputs && d.outputs.length) return d.outputs
      return [
        { name: 'evidences', label: '检索证据列表', type: 'array', description: '命中的知识片段集合' },
        { name: 'evidenceCount', label: '证据数量', type: 'number', description: '命中证据条数' },
        { name: 'sourceType', label: '来源类型', type: 'string', description: 'DOCUMENT / STRUCTURED_DATA / GRAPH / HYBRID' },
        { name: 'diagnostics', label: '诊断信息', type: 'object', description: '检索过程的诊断信息' }
      ]
    case 'output':
      return []
    case 'condition':
      // 条件节点透传：输出 = 所有输入
      return getNodeInputs(node)
    default:
      return d.outputs || []
  }
}

/**
 * 获取节点的输入字段列表
 */
export function getNodeInputs(node) {
  const d = node.data || {}
  switch (node.type) {
    case 'userInput':
      return []
    case 'llm': {
      // 优先使用手动定义的 inputs
      if (d.inputs && d.inputs.length) return d.inputs
      // 回退：从模板提取
      const vars = new Set([
        ...extractTemplateVariables(d.systemPrompt),
        ...extractTemplateVariables(d.userPrompt)
      ])
      return [...vars].map(name => ({ name, label: name, type: 'string', description: '' }))
    }
    case 'agent':
      return d.inputs || []
    case 'skill':
      return d.inputs || []
    case 'smartAction': {
      // 优先使用手动定义的 inputs
      if (d.inputs && d.inputs.length) return d.inputs
      // 回退：从操作要求模板提取
      const saVars = new Set(extractTemplateVariables(d.actionPrompt))
      return [...saVars].map(name => ({ name, label: name, type: 'string', description: '' }))
    }
    case 'knowledgeRetrieval': {
      // 优先使用手动定义的 inputs
      if (d.inputs && d.inputs.length) return d.inputs
      // 回退：从检索语句模板提取
      const krVars = new Set(extractTemplateVariables(d.query))
      return [...krVars].map(name => ({ name, label: name, type: 'string', description: '' }))
    }
    case 'output':
      return d.fields || []
    case 'condition': {
      // 优先使用手动定义的 inputs
      if (d.inputs && d.inputs.length) return d.inputs
      // 回退：从条件表达式提取
      const vars = new Set()
      for (const c of (d.conditions || [])) {
        for (const v of extractTemplateVariables(c.expression)) {
          vars.add(v)
        }
      }
      return [...vars].map(name => ({ name, label: name, type: 'string', description: '' }))
    }
    default:
      return d.inputs || []
  }
}

// ========== 类型兼容性 ==========

/**
 * 判断源类型是否可以赋值给目标类型
 */
export function isTypeCompatible(sourceType, targetType) {
  if (sourceType === targetType) return true
  // 以下类型可隐式转为 string
  const stringLike = ['filePath', 'directoryPath', 'number', 'boolean']
  if (stringLike.includes(sourceType) && targetType === 'string') return true
  // array → object 兼容
  if (sourceType === 'array' && targetType === 'object') return true
  return false
}

// ========== 单边匹配 ==========

/**
 * 匹配源节点输出与目标节点输入
 * @returns {{ status, mapping, unmatchedSource, unmatchedTarget, typeMismatches }}
 */
export function matchIO(sourceOutputs, targetInputs, existingMapping) {
  const mapping = []
  const unmatchedSource = [...sourceOutputs]
  const unmatchedTarget = [...targetInputs]
  const typeMismatches = []

  // 第一轮：按变量名精确匹配
  for (let si = unmatchedSource.length - 1; si >= 0; si--) {
    const srcField = unmatchedSource[si]
    const ti = unmatchedTarget.findIndex(t => t.name === srcField.name)
    if (ti !== -1) {
      const tgtField = unmatchedTarget[ti]
      if (isTypeCompatible(srcField.type, tgtField.type)) {
        mapping.push({ sourceField: srcField.name, targetField: tgtField.name })
      } else {
        typeMismatches.push({ source: srcField, target: tgtField })
      }
      unmatchedSource.splice(si, 1)
      unmatchedTarget.splice(ti, 1)
    }
  }

  // 第二轮：保留已有的手动映射（不与新映射冲突的部分）
  if (existingMapping) {
    for (const em of existingMapping) {
      if (mapping.some(m => m.sourceField === em.sourceField && m.targetField === em.targetField)) continue
      if (mapping.some(m => m.sourceField === em.sourceField || m.targetField === em.targetField)) continue
      mapping.push(em)
      const ti = unmatchedTarget.findIndex(t => t.name === em.targetField)
      if (ti !== -1) unmatchedTarget.splice(ti, 1)
    }
  }

  // 判定状态
  let status
  if (targetInputs.length === 0 && sourceOutputs.length === 0) {
    status = 'unmapped'
  } else if (unmatchedTarget.length === 0 && typeMismatches.length === 0) {
    status = 'ok'
  } else if (typeMismatches.length > 0) {
    status = 'mismatch'
  } else if (unmatchedTarget.length > 0 && unmatchedSource.length === 0) {
    status = 'partial'
  } else if (unmatchedTarget.length === 0 && unmatchedSource.length > 0) {
    status = 'ok' // 源有多余输出，但目标全部满足
  } else {
    status = 'mismatch'
  }

  return { status, mapping, unmatchedSource, unmatchedTarget, typeMismatches }
}

// ========== 推断调度 ==========

/**
 * 推断一条边的映射状态
 * mapping 仍然基于直接源节点的输出（精确映射），
 * 但状态判断基于所有可达前驱节点的合并输出（因为上下文是累积的）
 * @param {object} sourceNode - 源节点
 * @param {object} targetNode - 目标节点
 * @param {object} [existingEdge] - 已有边数据
 * @param {Array} [allNodes] - 全部节点（用于可达前驱计算）
 * @param {Array} [allEdges] - 全部边（用于可达前驱计算）
 * @returns {{ status, mapping, unmatchedSource, unmatchedTarget, typeMismatches }}
 */
export function inferEdgeMapping(sourceNode, targetNode, existingEdge, allNodes, allEdges) {
  const sourceOutputs = getNodeOutputs(sourceNode)
  const targetInputs = getNodeInputs(targetNode)
  const existingMapping = existingEdge?.data?.mapping

  // 直接源→目标的精确映射（用于 mapping 字段）
  const directResult = matchIO(sourceOutputs, targetInputs, existingMapping)

  // 如果没有传入全图数据，降级为只看直接源
  if (!allNodes || !allEdges) {
    return directResult
  }

  // 收集所有可达前驱节点的合并输出（用于状态判断）
  const reachableOutputs = collectReachableOutputs(targetNode.id, allNodes, allEdges)
  if (reachableOutputs.length === 0 && targetInputs.length === 0) {
    return { ...directResult, status: 'unmapped' }
  }

  // 用合并输出判断目标输入是否全部满足
  const unmatched = []
  const typeMismatches = []
  for (const tgtField of targetInputs) {
    const srcField = reachableOutputs.find(o => o.name === tgtField.name)
    if (!srcField) {
      unmatched.push(tgtField)
    } else if (!isTypeCompatible(srcField.type, tgtField.type)) {
      typeMismatches.push({ source: srcField, target: tgtField })
    }
  }

  let status
  if (targetInputs.length === 0 && sourceOutputs.length === 0) {
    status = 'unmapped'
  } else if (unmatched.length === 0 && typeMismatches.length === 0) {
    status = 'ok'
  } else if (typeMismatches.length > 0) {
    status = 'mismatch'
  } else {
    status = 'partial'
  }

  return { ...directResult, status }
}

/**
 * 收集目标节点所有可达前驱节点的合并输出（BFS 回溯边图）
 * @param {string} targetId - 目标节点 ID
 * @param {Array} allNodes - 全部节点
 * @param {Array} allEdges - 全部边
 * @returns {Array} 合并后的输出字段列表（去重，先出现的优先）
 */
function collectReachableOutputs(targetId, allNodes, allEdges) {
  const nodeMap = new Map(allNodes.map(n => [n.id, n]))
  const visited = new Set()
  const queue = [targetId]
  const merged = new Map() // name → field

  while (queue.length > 0) {
    const currentId = queue.shift()
    if (visited.has(currentId)) continue
    visited.add(currentId)

    // 找到所有指向当前节点的边
    for (const edge of allEdges) {
      if (edge.target === currentId && !visited.has(edge.source)) {
        const sourceNode = nodeMap.get(edge.source)
        if (sourceNode) {
          for (const field of getNodeOutputs(sourceNode)) {
            if (!merged.has(field.name)) {
              merged.set(field.name, field)
            }
          }
          queue.push(edge.source)
        }
      }
    }
  }

  return [...merged.values()]
}

/**
 * 获取目标节点的所有可达前驱节点（通过边反向 BFS）
 *
 * 用于"前置数据关联"下拉选项：节点输入不仅可关联直接上游，
 * 也可关联画布中所有可达的前序节点输出。
 *
 * @param {string} targetId - 目标节点 ID
 * @param {Array} allNodes - 全部节点
 * @param {Array} allEdges - 全部边
 * @returns {Array} 可达前驱节点列表（按 BFS 发现顺序，不含 targetId 本身，不重复）
 */
export function getReachablePredecessors(targetId, allNodes, allEdges) {
  const nodeMap = new Map(allNodes.map(n => [n.id, n]))
  const visited = new Set([targetId])
  const queue = [targetId]
  const result = []

  while (queue.length > 0) {
    const currentId = queue.shift()
    for (const edge of allEdges) {
      if (edge.target !== currentId) continue
      if (visited.has(edge.source)) continue
      visited.add(edge.source)
      const sourceNode = nodeMap.get(edge.source)
      if (sourceNode) {
        result.push(sourceNode)
        queue.push(edge.source)
      }
    }
  }

  return result
}

/**
 * 获取起始节点的所有可达后继节点（通过边正向 BFS）
 *
 * 用于"输出变量引入"：节点输出可关联到所有可达后继节点的输入字段。
 *
 * @param {string} sourceId - 起始节点 ID
 * @param {Array} allNodes - 全部节点
 * @param {Array} allEdges - 全部边
 * @returns {Array} 可达后继节点列表（按 BFS 发现顺序，不含 sourceId 本身，不重复）
 */
export function getReachableSuccessors(sourceId, allNodes, allEdges) {
  const nodeMap = new Map(allNodes.map(n => [n.id, n]))
  const visited = new Set([sourceId])
  const queue = [sourceId]
  const result = []

  while (queue.length > 0) {
    const currentId = queue.shift()
    for (const edge of allEdges) {
      if (edge.source !== currentId) continue
      if (visited.has(edge.target)) continue
      visited.add(edge.target)
      const targetNode = nodeMap.get(edge.target)
      if (targetNode) {
        result.push(targetNode)
        queue.push(edge.target)
      }
    }
  }

  return result
}

/**
 * 推断指定边的样式
 */
export function getEdgeStyle(status) {
  return EDGE_STATUS_STYLE[status] || EDGE_STATUS_STYLE.unmapped
}

/**
 * 刷新图中与指定节点关联的所有边的映射
 * 返回需要更新的边列表
 * @param {string} nodeId - 变更的节点 ID
 * @param {Array} nodes - 所有节点
 * @param {Array} edges - 所有边
 * @returns {Array} - 需要更新的边 [{ id, data, style }]
 */
export function refreshMappingsForNode(nodeId, nodes, edges) {
  const updates = []
  const nodeMap = new Map(nodes.map(n => [n.id, n]))

  for (const edge of edges) {
    const isRelevant = edge.source === nodeId || edge.target === nodeId
    if (!isRelevant) continue

    const sourceNode = nodeMap.get(edge.source)
    const targetNode = nodeMap.get(edge.target)
    if (!sourceNode || !targetNode) continue

    const result = inferEdgeMapping(sourceNode, targetNode, edge, nodes, edges)
    updates.push({
      id: edge.id,
      data: {
        ...edge.data,
        mapping: result.mapping,
        status: result.status,
        unmatchedSource: result.unmatchedSource,
        unmatchedTarget: result.unmatchedTarget,
        typeMismatches: result.typeMismatches
      },
      style: getEdgeStyle(result.status)
    })
  }

  return updates
}

/**
 * 推断一条新边的映射（用于 onConnect）
 */
export function inferNewEdge(sourceNode, targetNode, allNodes, allEdges) {
  const result = inferEdgeMapping(sourceNode, targetNode, undefined, allNodes, allEdges)
  return {
    data: {
      mapping: result.mapping,
      status: result.status,
      unmatchedSource: result.unmatchedSource,
      unmatchedTarget: result.unmatchedTarget,
      typeMismatches: result.typeMismatches
    },
    style: getEdgeStyle(result.status)
  }
}

// ========== 多源汇聚分配 ==========

/**
 * 计算多源汇聚时的变量分配方案
 * @param {Array} sources - 源节点列表
 * @param {object} target - 目标节点
 * @param {Array} edges - 连接到目标的边列表
 * @returns {{ edges: Array<{ edgeId, mapping, provided, missing }> }}
 */
export function resolveMultiSource(sources, target, edges) {
  const targetInputs = getNodeInputs(target)
  if (targetInputs.length === 0) {
    return { edges: edges.map(e => ({ edgeId: e.id, mapping: [], provided: [], missing: [] })) }
  }

  const allSourceOutputs = sources.map(s => getNodeOutputs(s))
  const results = []

  // 计算所有源的合并输出
  const mergedOutputs = new Map()
  for (let i = 0; i < sources.length; i++) {
    for (const field of allSourceOutputs[i]) {
      if (!mergedOutputs.has(field.name)) {
        mergedOutputs.set(field.name, { field, sourceIndex: i })
      }
    }
  }

  // 尝试匹配每个目标输入字段
  const assigned = new Map() // targetField -> sourceIndex
  for (const tgtField of targetInputs) {
    const entry = mergedOutputs.get(tgtField.name)
    if (entry && isTypeCompatible(entry.field.type, tgtField.type)) {
      assigned.set(tgtField.name, entry.sourceIndex)
    }
  }

  // 分配回每条边
  for (let i = 0; i < edges.length; i++) {
    const sourceOutputs = allSourceOutputs[i] || []
    const edgeMapping = []
    const provided = []

    for (const tgtField of targetInputs) {
      if (assigned.get(tgtField.name) === i) {
        const srcField = sourceOutputs.find(f => f.name === tgtField.name)
        if (srcField) {
          edgeMapping.push({ sourceField: srcField.name, targetField: tgtField.name })
          provided.push(tgtField.name)
        }
      }
    }

    const missing = targetInputs.filter(t => !assigned.has(t.name)).map(t => t.name)

    results.push({
      edgeId: edges[i].id,
      mapping: edgeMapping,
      provided,
      missing
    })
  }

  return { edges: results }
}