package engine

import (
	"context"
	"fmt"
	"log/slog"
	"strings"
	"time"

	"github.com/local/glpi-neural-brain/internal/model"
	"github.com/local/glpi-neural-brain/internal/ollama"
	"github.com/local/glpi-neural-brain/internal/sourceagent"
)

func (e *Engine) SetSourceInbox(store *sourceagent.Store) { e.SourceInbox = store }

func (e *Engine) evidenceAcquisitionEnabled() bool {
	return (e.SourceInbox != nil && e.Cfg.SourceInboxEnabled) || e.ResearchEnabledForRuntime()
}

func (e *Engine) sourceInboxLoop(ctx context.Context) {
	ticker := time.NewTicker(e.Cfg.SourceInboxInterval)
	defer ticker.Stop()
	for {
		select {
		case <-ctx.Done():
			return
		case <-ticker.C:
			e.processSourceInbox(ctx)
		}
	}
}

func (e *Engine) processSourceInbox(ctx context.Context) {
	if e.SourceInbox == nil || !e.Cfg.SourceInboxEnabled {
		return
	}
	items, err := e.SourceInbox.ClaimInbox(ctx, e.Cfg.SourceInboxBatchSize)
	if err != nil {
		slog.Warn("source inbox claim failed", "error", err)
		return
	}
	if len(items) == 0 {
		return
	}
	texts := make([]string, 0, len(items))
	for _, item := range items {
		text := item.Document.Title + "\n" + strings.Join(item.Document.Categories, " · ") + "\n" + item.Document.Text
		if len([]rune(text)) > 12000 {
			text = string([]rune(text)[:12000])
		}
		texts = append(texts, text)
	}
	cctx, cancel := context.WithTimeout(ollama.WithLowPriority(ctx), 4*time.Minute)
	vectors, embedErr := e.Ollama.Embed(cctx, texts)
	cancel()
	if embedErr != nil || len(vectors) != len(items) {
		for _, item := range items {
			_ = e.SourceInbox.ReleaseInbox(ctx, item.ID, fmt.Sprint(embedErr))
		}
		return
	}
	candidateCount := 0
	for i, item := range items {
		hits, stats := e.similarKnowledge(vectors[i], 1, e.effectiveLearningFilter(), 0)
		if len(hits) == 0 {
			_ = e.SourceInbox.ReleaseInbox(ctx, item.ID, "knowledge vectors are not ready")
			continue
		}
		relevance := 0.0
		matched := ""
		if len(hits) > 0 {
			relevance = hits[0].Score
			matched = hits[0].NodeID
		}
		status := "archived"
		if relevance >= e.Cfg.SourceInboxMinSimilarity {
			status = "candidate"
			candidateCount++
		}
		meta := make(map[string]any, len(item.Metadata)+4)
		for key, value := range item.Metadata {
			meta[key] = value
		}
		meta["processing_mode"] = e.RuntimeSettings().ProcessingMode
		meta["exact_comparisons"] = stats.ExactComparisons
		meta["coarse_comparisons"] = stats.CoarseComparisons
		meta["candidate_pool"] = stats.CandidatePool
		_ = e.SourceInbox.CompleteClassification(ctx, item.ID, status, relevance, matched, meta)
	}
	if e.Broker != nil {
		e.Broker.Publish(model.Activity{Type: "source.inbox.classified", Source: "brain", Phase: "source-inbox", Message: fmt.Sprintf("Source-Inbox: %d Dokumente geprüft · %d als Wissenskandidaten vorgemerkt", len(items), candidateCount), Strength: .36, Metadata: map[string]any{"documents": len(items), "candidates": candidateCount, "minimum_similarity": e.Cfg.SourceInboxMinSimilarity}})
	}
}

func (e *Engine) sourceInboxResearch(ctx context.Context, query string, limit int, freshnessSensitive bool) []model.ResearchResult {
	if e.SourceInbox == nil || !e.Cfg.SourceInboxEnabled || limit < 1 {
		return nil
	}
	maxAge := time.Duration(0)
	if freshnessSensitive {
		maxAge = e.Cfg.SourceInboxFreshMaxAge
	}
	items, err := e.SourceInbox.SearchCandidates(ctx, query, limit, maxAge)
	if err != nil {
		return nil
	}
	out := make([]model.ResearchResult, 0, len(items))
	for _, item := range items {
		if item.QueryScore < .18 {
			continue
		}
		d := item.Document
		out = append(out, model.ResearchResult{Title: d.Title, URL: d.CanonicalURL, Snippet: clamp(d.Text, 1000), Content: d.Text, ContentType: d.ContentType, Query: query, Language: d.Language, Fetched: true, Relevant: true, Relevance: item.QueryScore, SourceQuality: "source_inbox", SourceQualityScore: .68, AssessmentReason: "Vorab durch Source-Agent gesammelt und vom Brain als thematisch passend zur Knowledgebase klassifiziert."})
	}
	if len(out) > 0 && e.Broker != nil {
		e.Broker.Publish(model.Activity{Type: "article.research.inbox", Source: "brain", Phase: "knowledge-research-routing", Message: fmt.Sprintf("Source-Inbox liefert %d bereits gecrawlte Kandidaten vor SearXNG", len(out)), Strength: .56, Metadata: map[string]any{"query": query, "results": len(out), "freshness_sensitive": freshnessSensitive}})
	}
	return out
}