veylant/internal/pii/http.go

package pii

import (
	"encoding/json"
	"fmt"
	"net/http"
	"regexp"
	"sort"
	"strings"

	"go.uber.org/zap"

	"github.com/veylant/ia-gateway/internal/apierror"
)

// AnalyzeRequest is the JSON body accepted by POST /v1/pii/analyze.
type AnalyzeRequest struct {
	Text string `json:"text"`
}

// AnalyzeEntity is a single PII entity returned by the analyze endpoint.
type AnalyzeEntity struct {
	Type       string  `json:"type"`
	Start      int     `json:"start"`
	End        int     `json:"end"`
	Confidence float64 `json:"confidence"`
	Layer      string  `json:"layer"`
}

// AnalyzeResponse is the JSON response of POST /v1/pii/analyze.
type AnalyzeResponse struct {
	Anonymized string          `json:"anonymized"`
	Entities   []AnalyzeEntity `json:"entities"`
}

// AnalyzeHandler wraps a pii.Client as an HTTP handler for the playground.
// It is safe to call when client is nil: falls back to regex detection.
type AnalyzeHandler struct {
	client *Client
	logger *zap.Logger
}

// NewAnalyzeHandler creates a new AnalyzeHandler.
// client may be nil (PII service disabled) — the handler falls back to regex.
func NewAnalyzeHandler(client *Client, logger *zap.Logger) *AnalyzeHandler {
	return &AnalyzeHandler{client: client, logger: logger}
}

func (h *AnalyzeHandler) ServeHTTP(w http.ResponseWriter, r *http.Request) {
	var req AnalyzeRequest
	if err := json.NewDecoder(r.Body).Decode(&req); err != nil {
		apierror.WriteError(w, apierror.NewBadRequestError("invalid JSON: "+err.Error()))
		return
	}
	if req.Text == "" {
		w.Header().Set("Content-Type", "application/json")
		w.WriteHeader(http.StatusOK)
		_ = json.NewEncoder(w).Encode(AnalyzeResponse{Anonymized: "", Entities: []AnalyzeEntity{}})
		return
	}

	// Attempt real PII detection if service is available.
	// NOTE: when fail_open=true, Detect() returns (result, nil) even on RPC
	// failure, but sets result.Entities to nil to signal the degraded path.
	// A real successful response always has a non-nil Entities slice.
	if h.client != nil {
		resp, err := h.client.Detect(r.Context(), req.Text, "playground", "playground-analyze", true, false)
		if err == nil && resp.Entities != nil {
			// Real PII service response (may be empty if no PII detected).
			entities := make([]AnalyzeEntity, 0, len(resp.Entities))
			for _, e := range resp.Entities {
				entities = append(entities, AnalyzeEntity{
					Type:       e.EntityType,
					Start:      int(e.Start),
					End:        int(e.End),
					Confidence: float64(e.Confidence),
					Layer:      e.DetectionLayer,
				})
			}
			w.Header().Set("Content-Type", "application/json")
			w.WriteHeader(http.StatusOK)
			_ = json.NewEncoder(w).Encode(AnalyzeResponse{
				Anonymized: resp.AnonymizedText,
				Entities:   entities,
			})
			return
		}
		if err != nil {
			h.logger.Warn("PII service error — falling back to regex detection", zap.Error(err))
		} else {
			h.logger.Debug("PII service unavailable (fail-open) — falling back to regex detection")
		}
	}

	// Fallback: local regex detection so the playground stays useful when the
	// PII sidecar is not running (dev mode, demo environments).
	result := regexDetect(req.Text)
	w.Header().Set("Content-Type", "application/json")
	w.WriteHeader(http.StatusOK)
	_ = json.NewEncoder(w).Encode(result)
}

// ── Regex-based local detection ───────────────────────────────────────────────

// Package-level compiled regexes (compiled once at startup).
var (
	rePiiIBAN  = regexp.MustCompile(`(?i)FR\d{2}[\s]?\d{4}[\s]?\d{4}[\s]?\d{4}[\s]?\d{4}[\s]?\d{4}[\s]?\d{3}`)
	rePiiEmail = regexp.MustCompile(`[a-zA-Z0-9._%+\-]+@[a-zA-Z0-9.\-]+\.[a-zA-Z]{2,}`)
	rePiiPhone = regexp.MustCompile(`(?:\+33|0033|0)[\s]?[1-9](?:[\s.\-]?\d{2}){4}`)
	rePiiCC    = regexp.MustCompile(`(?:4\d{3}|5[1-5]\d{2}|3[47]\d{2})(?:[ \-]?\d{4}){3}`)
	rePiiSSN   = regexp.MustCompile(`\b[12][\s]?\d{2}[\s]?\d{2}[\s]?\d{2}[\s]?\d{3}[\s]?\d{3}[\s]?\d{2}\b`)
)

type regexPattern struct {
	re  *regexp.Regexp
	typ string
}

var piiPatterns = []regexPattern{
	{rePiiIBAN, "IBAN_CODE"},
	{rePiiEmail, "EMAIL_ADDRESS"},
	{rePiiPhone, "PHONE_NUMBER"},
	{rePiiCC, "CREDIT_CARD"},
	{rePiiSSN, "FR_SSN"},
}

type rawMatch struct {
	typ   string
	start int
	end   int
}

// regexDetect runs a set of compiled regexes over text and returns matched
// entities with their byte offsets, plus an anonymized version of the text.
func regexDetect(text string) AnalyzeResponse {
	var raw []rawMatch
	for _, p := range piiPatterns {
		for _, loc := range p.re.FindAllStringIndex(text, -1) {
			raw = append(raw, rawMatch{typ: p.typ, start: loc[0], end: loc[1]})
		}
	}
	// Sort by start position.
	sort.Slice(raw, func(i, j int) bool { return raw[i].start < raw[j].start })

	// Remove overlapping matches (keep the first / longest-starting one).
	filtered := raw[:0]
	cursor := 0
	for _, m := range raw {
		if m.start >= cursor {
			filtered = append(filtered, m)
			cursor = m.end
		}
	}

	if len(filtered) == 0 {
		return AnalyzeResponse{Anonymized: text, Entities: []AnalyzeEntity{}}
	}

	// Build anonymized text and entity list simultaneously.
	var sb strings.Builder
	counters := map[string]int{}
	entities := make([]AnalyzeEntity, 0, len(filtered))
	cursor = 0

	for _, m := range filtered {
		sb.WriteString(text[cursor:m.start])
		counters[m.typ]++
		sb.WriteString(fmt.Sprintf("[%s_%d]", m.typ, counters[m.typ]))
		entities = append(entities, AnalyzeEntity{
			Type:       m.typ,
			Start:      m.start,
			End:        m.end,
			Confidence: 0.95,
			Layer:      "regex-local",
		})
		cursor = m.end
	}
	sb.WriteString(text[cursor:])

	return AnalyzeResponse{
		Anonymized: sb.String(),
		Entities:   entities,
	}
}