seo-generator-server/lib/pipeline/PipelineExecutor.js

/**
 * PipelineExecutor.js
 *
 * Moteur d'exécution des pipelines modulaires flexibles.
 * Orchestre l'exécution séquentielle des modules avec gestion d'état.
 */

const { logSh } = require('../ErrorReporting');
const { tracer } = require('../trace');
const { PipelineDefinition } = require('./PipelineDefinition');
const { getPersonalities, readInstructionsData, selectPersonalityWithAI } = require('../BrainConfig');
const { extractElements, buildSmartHierarchy } = require('../ElementExtraction');
const { generateMissingKeywords } = require('../MissingKeywords');

// Modules d'exécution
const { generateSimple } = require('../selective-enhancement/SelectiveUtils');
const { applySelectiveLayer } = require('../selective-enhancement/SelectiveCore');
const { applyPredefinedStack: applySelectiveStack } = require('../selective-enhancement/SelectiveLayers');
const { applyAdversarialLayer } = require('../adversarial-generation/AdversarialCore');
const { applyPredefinedStack: applyAdversarialStack } = require('../adversarial-generation/AdversarialLayers');
const { applyHumanSimulationLayer } = require('../human-simulation/HumanSimulationCore');
const { applyPredefinedSimulation } = require('../human-simulation/HumanSimulationLayers');
const { applyPatternBreakingLayer } = require('../pattern-breaking/PatternBreakingCore');
const { applyPatternBreakingStack } = require('../pattern-breaking/PatternBreakingLayers');

/**
 * Classe PipelineExecutor
 */
class PipelineExecutor {
  constructor() {
    this.currentContent = null;
    this.executionLog = [];
    this.checkpoints = [];
    this.metadata = {
      startTime: null,
      endTime: null,
      totalDuration: 0,
      personality: null
    };
  }

  /**
   * Exécute un pipeline complet
   */
  async execute(pipelineConfig, rowNumber, options = {}) {
    return tracer.run('PipelineExecutor.execute', async () => {
      logSh(`🚀 Démarrage pipeline "${pipelineConfig.name}" (${pipelineConfig.pipeline.length} étapes)`, 'INFO');

      // Validation
      const validation = PipelineDefinition.validate(pipelineConfig);
      if (!validation.valid) {
        throw new Error(`Pipeline invalide: ${validation.errors.join(', ')}`);
      }

      this.metadata.startTime = Date.now();
      this.executionLog = [];
      this.checkpoints = [];

      // Charger les données
      const csvData = await this.loadData(rowNumber);

      // Exécuter les étapes
      const enabledSteps = pipelineConfig.pipeline.filter(s => s.enabled !== false);

      for (let i = 0; i < enabledSteps.length; i++) {
        const step = enabledSteps[i];

        try {
          logSh(`▶ Étape ${step.step}/${pipelineConfig.pipeline.length}: ${step.module} (${step.mode})`, 'INFO');

          const stepStartTime = Date.now();
          const result = await this.executeStep(step, csvData, options);
          const stepDuration = Date.now() - stepStartTime;

          // Log l'étape
          this.executionLog.push({
            step: step.step,
            module: step.module,
            mode: step.mode,
            intensity: step.intensity,
            duration: stepDuration,
            modifications: result.modifications || 0,
            success: true,
            timestamp: new Date().toISOString()
          });

          // Mise à jour du contenu
          if (result.content) {
            this.currentContent = result.content;
          }

          // Checkpoint si demandé
          if (step.saveCheckpoint) {
            this.checkpoints.push({
              step: step.step,
              content: this.currentContent,
              timestamp: new Date().toISOString()
            });
            logSh(`💾 Checkpoint sauvegardé (étape ${step.step})`, 'DEBUG');
          }

          logSh(`✔ Étape ${step.step} terminée (${stepDuration}ms, ${result.modifications || 0} modifs)`, 'INFO');

        } catch (error) {
          logSh(`✖ Erreur étape ${step.step}: ${error.message}`, 'ERROR');

          this.executionLog.push({
            step: step.step,
            module: step.module,
            mode: step.mode,
            success: false,
            error: error.message,
            timestamp: new Date().toISOString()
          });

          // Propager l'erreur ou continuer selon options
          if (options.stopOnError !== false) {
            throw error;
          }
        }
      }

      this.metadata.endTime = Date.now();
      this.metadata.totalDuration = this.metadata.endTime - this.metadata.startTime;

      logSh(`✅ Pipeline terminé: ${this.metadata.totalDuration}ms`, 'INFO');

      return {
        success: true,
        finalContent: this.currentContent,
        executionLog: this.executionLog,
        checkpoints: this.checkpoints,
        metadata: {
          ...this.metadata,
          pipelineName: pipelineConfig.name,
          totalSteps: enabledSteps.length,
          successfulSteps: this.executionLog.filter(l => l.success).length
        }
      };
    }, { pipelineName: pipelineConfig.name, rowNumber });
  }

  /**
   * Charge les données depuis Google Sheets
   */
  async loadData(rowNumber) {
    return tracer.run('PipelineExecutor.loadData', async () => {
      const csvData = await readInstructionsData(rowNumber);

      // Charger personnalité si besoin
      const personalities = await getPersonalities();
      const personality = await selectPersonalityWithAI(
        csvData.mc0,
        csvData.t0,
        personalities
      );

      csvData.personality = personality;
      this.metadata.personality = personality.nom;

      logSh(`📊 Données chargées: ${csvData.mc0}, personnalité: ${personality.nom}`, 'DEBUG');

      return csvData;
    }, { rowNumber });
  }

  /**
   * Exécute une étape individuelle
   */
  async executeStep(step, csvData, options) {
    return tracer.run(`PipelineExecutor.executeStep.${step.module}`, async () => {

      switch (step.module) {
        case 'generation':
          return await this.runGeneration(step, csvData);

        case 'selective':
          return await this.runSelective(step, csvData);

        case 'adversarial':
          return await this.runAdversarial(step, csvData);

        case 'human':
          return await this.runHumanSimulation(step, csvData);

        case 'pattern':
          return await this.runPatternBreaking(step, csvData);

        default:
          throw new Error(`Module inconnu: ${step.module}`);
      }

    }, { step: step.step, module: step.module, mode: step.mode });
  }

  /**
   * Exécute la génération initiale
   */
  async runGeneration(step, csvData) {
    return tracer.run('PipelineExecutor.runGeneration', async () => {

      if (this.currentContent) {
        logSh('⚠️ Contenu déjà généré, génération ignorée', 'WARN');
        return { content: this.currentContent, modifications: 0 };
      }

      // Étape 1: Extraire les éléments depuis le template XML
      const elements = await extractElements(csvData.xmlTemplate, csvData);
      logSh(`✓ Extraction: ${elements.length} éléments extraits`, 'DEBUG');

      // Étape 2: Générer les mots-clés manquants
      const finalElements = await generateMissingKeywords(elements, csvData);

      // Étape 3: Construire la hiérarchie
      const elementsArray = Array.isArray(finalElements) ? finalElements :
        (finalElements && typeof finalElements === 'object') ? Object.values(finalElements) : [];
      const hierarchy = await buildSmartHierarchy(elementsArray);
      logSh(`✓ Hiérarchie: ${Object.keys(hierarchy).length} sections`, 'DEBUG');

      // Étape 4: Génération simple avec LLM configurable
      const llmProvider = step.parameters?.llmProvider || 'claude';
      const result = await generateSimple(hierarchy, csvData, { llmProvider });

      logSh(`✓ Génération: ${Object.keys(result.content || {}).length} éléments créés avec ${llmProvider}`, 'DEBUG');

      return {
        content: result.content,
        modifications: Object.keys(result.content || {}).length
      };

    }, { mode: step.mode });
  }

  /**
   * Exécute l'enhancement sélectif
   */
  async runSelective(step, csvData) {
    return tracer.run('PipelineExecutor.runSelective', async () => {

      if (!this.currentContent) {
        throw new Error('Aucun contenu à améliorer. Génération requise avant selective enhancement');
      }

      // Configuration de la couche
      const llmProvider = step.parameters?.llmProvider || 'openai';
      const config = {
        csvData,
        personality: csvData.personality,
        intensity: step.intensity || 1.0,
        llmProvider: llmProvider,
        ...step.parameters
      };

      let result;

      // Utiliser le stack si c'est un mode prédéfini
      const predefinedStacks = ['lightEnhancement', 'standardEnhancement', 'fullEnhancement', 'personalityFocus', 'fluidityFocus', 'adaptive'];

      if (predefinedStacks.includes(step.mode)) {
        result = await applySelectiveStack(this.currentContent, step.mode, config);
      } else {
        // Sinon utiliser la couche directe
        result = await applySelectiveLayer(this.currentContent, config);
      }

      logSh(`✓ Selective: modifications appliquées avec ${llmProvider}`, 'DEBUG');

      return {
        content: result.content || result,
        modifications: result.modificationsCount || 0
      };

    }, { mode: step.mode, intensity: step.intensity });
  }

  /**
   * Exécute l'adversarial generation
   */
  async runAdversarial(step, csvData) {
    return tracer.run('PipelineExecutor.runAdversarial', async () => {

      if (!this.currentContent) {
        throw new Error('Aucun contenu à traiter. Génération requise avant adversarial');
      }

      if (step.mode === 'none') {
        logSh('Adversarial mode = none, ignoré', 'DEBUG');
        return { content: this.currentContent, modifications: 0 };
      }

      const llmProvider = step.parameters?.llmProvider || 'gemini';
      const config = {
        csvData,
        detectorTarget: step.parameters?.detector || 'general',
        method: step.parameters?.method || 'regeneration',
        intensity: step.intensity || 1.0,
        llmProvider: llmProvider
      };

      let result;

      // Mapper les noms user-friendly vers les vrais noms de stacks
      const stackMapping = {
        'light': 'lightDefense',
        'standard': 'standardDefense',
        'heavy': 'heavyDefense',
        'adaptive': 'adaptive'
      };

      // Utiliser le stack si c'est un mode prédéfini
      if (stackMapping[step.mode]) {
        const stackName = stackMapping[step.mode];

        if (stackName === 'adaptive') {
          // Mode adaptatif utilise la couche directe
          result = await applyAdversarialLayer(this.currentContent, config);
        } else {
          result = await applyAdversarialStack(this.currentContent, stackName, config);
        }
      } else {
        // Sinon utiliser la couche directe
        result = await applyAdversarialLayer(this.currentContent, config);
      }

      logSh(`✓ Adversarial: modifications appliquées avec ${llmProvider}`, 'DEBUG');

      return {
        content: result.content || result,
        modifications: result.modificationsCount || 0
      };

    }, { mode: step.mode, detector: step.parameters?.detector });
  }

  /**
   * Exécute la simulation humaine
   */
  async runHumanSimulation(step, csvData) {
    return tracer.run('PipelineExecutor.runHumanSimulation', async () => {

      if (!this.currentContent) {
        throw new Error('Aucun contenu à traiter. Génération requise avant human simulation');
      }

      if (step.mode === 'none') {
        logSh('Human simulation mode = none, ignoré', 'DEBUG');
        return { content: this.currentContent, modifications: 0 };
      }

      const llmProvider = step.parameters?.llmProvider || 'mistral';
      const config = {
        csvData,
        personality: csvData.personality,
        intensity: step.intensity || 1.0,
        fatigueLevel: step.parameters?.fatigueLevel || 0.5,
        errorRate: step.parameters?.errorRate || 0.3,
        llmProvider: llmProvider
      };

      let result;

      // Utiliser le stack si c'est un mode prédéfini
      const predefinedModes = ['lightSimulation', 'standardSimulation', 'heavySimulation', 'adaptiveSimulation', 'personalityFocus', 'temporalFocus'];

      if (predefinedModes.includes(step.mode)) {
        result = await applyPredefinedSimulation(this.currentContent, step.mode, config);
      } else {
        // Sinon utiliser la couche directe
        result = await applyHumanSimulationLayer(this.currentContent, config);
      }

      logSh(`✓ Human Simulation: modifications appliquées avec ${llmProvider}`, 'DEBUG');

      return {
        content: result.content || result,
        modifications: result.modificationsCount || 0
      };

    }, { mode: step.mode, intensity: step.intensity });
  }

  /**
   * Exécute le pattern breaking
   */
  async runPatternBreaking(step, csvData) {
    return tracer.run('PipelineExecutor.runPatternBreaking', async () => {

      if (!this.currentContent) {
        throw new Error('Aucun contenu à traiter. Génération requise avant pattern breaking');
      }

      if (step.mode === 'none') {
        logSh('Pattern breaking mode = none, ignoré', 'DEBUG');
        return { content: this.currentContent, modifications: 0 };
      }

      const llmProvider = step.parameters?.llmProvider || 'deepseek';
      const config = {
        csvData,
        personality: csvData.personality,
        intensity: step.intensity || 1.0,
        focus: step.parameters?.focus || 'both',
        llmProvider: llmProvider
      };

      let result;

      // Utiliser le stack si c'est un mode prédéfini
      const predefinedModes = ['lightPatternBreaking', 'standardPatternBreaking', 'heavyPatternBreaking', 'adaptivePatternBreaking', 'syntaxFocus', 'connectorsFocus'];

      if (predefinedModes.includes(step.mode)) {
        result = await applyPatternBreakingStack(step.mode, this.currentContent, config);
      } else {
        // Sinon utiliser la couche directe
        result = await applyPatternBreakingLayer(this.currentContent, config);
      }

      logSh(`✓ Pattern Breaking: modifications appliquées avec ${llmProvider}`, 'DEBUG');

      return {
        content: result.content || result,
        modifications: result.modificationsCount || 0
      };

    }, { mode: step.mode, intensity: step.intensity });
  }

  /**
   * Obtient le contenu actuel
   */
  getCurrentContent() {
    return this.currentContent;
  }

  /**
   * Obtient le log d'exécution
   */
  getExecutionLog() {
    return this.executionLog;
  }

  /**
   * Obtient les checkpoints sauvegardés
   */
  getCheckpoints() {
    return this.checkpoints;
  }

  /**
   * Obtient les métadonnées d'exécution
   */
  getMetadata() {
    return this.metadata;
  }

  /**
   * Reset l'état de l'executor
   */
  reset() {
    this.currentContent = null;
    this.executionLog = [];
    this.checkpoints = [];
    this.metadata = {
      startTime: null,
      endTime: null,
      totalDuration: 0,
      personality: null
    };
  }
}

module.exports = { PipelineExecutor };