knowledge-governance/internal/platform/postgres/dingtalk_space.go

package postgres

import (
	"context"
	"crypto/sha256"
	"encoding/hex"
	"errors"
	"fmt"
	"strings"
	"time"

	"github.com/jackc/pgx/v5"
)

// DingTalk space registration: a knowledge-base space discovered through the
// login sidecar becomes a source_connections row with a DETERMINISTIC id,
// derived the same way the catalog importer derives stable ids
// ('sc-' + sha256('dingtalk-'+root_key)[:32]). Re-registering the same space
// is an idempotent re-sync of name/root, never a duplicate row. New rows are
// enabled but carry no schedule (next_refresh_at stays NULL): automatic scans
// begin only after an admin sets an interval — same posture as the existing
// imported DingTalk sources.

// 登记失败的可行动分类（处理器据此返回不同错误码，前端才能说清"该怎么办"）：
//   - 同名来源**仍在使用**（未归档）：不能悄悄改名，也不能出现两个同名来源；
//   - 同一个登记键已指向另一个 root，而那个来源仍在使用。
var (
	ErrDingtalkSpaceNameTaken        = errors.New("dingtalk space name is already used by a live connection")
	ErrDingtalkSpaceIdentityConflict = errors.New("dingtalk space registration: live connection already serves another root")
)

// RegisterDingtalkSpaceInput describes one space resolved by the sidecar.
type RegisterDingtalkSpaceInput struct {
	RootKey string // sidecar inventory key, e.g. "atom" or "space-1f2e3d4c5b"
	Name    string // live space name from DingTalk spaceProfile
	SpaceID string // alidocs space id from the URL
	RootID  string // root dentry uuid
	Now     time.Time
}

func (in RegisterDingtalkSpaceInput) ConnectionID() string {
	sum := sha256.Sum256([]byte("dingtalk-" + in.RootKey))
	return "sc-" + hex.EncodeToString(sum[:])[:32]
}

func (in RegisterDingtalkSpaceInput) validate() error {
	if in.Now.IsZero() {
		return errors.New("dingtalk space registration: now is required")
	}
	if !validShortID(in.RootKey) || strings.TrimSpace(in.Name) == "" || len(in.Name) > 255 {
		return errors.New("dingtalk space registration: invalid key or name")
	}
	for _, id := range []string{in.SpaceID, in.RootID} {
		if id == "" || len(id) > 128 || strings.ContainsAny(id, " \t\r\n/") {
			return errors.New("dingtalk space registration: invalid identifier")
		}
	}
	return nil
}

func (s *Store) RegisterDingtalkSpace(ctx context.Context, in RegisterDingtalkSpaceInput) (string, error) {
	if err := in.validate(); err != nil {
		return "", err
	}
	connectionID := in.ConnectionID()
	var existingName, existingRoot string
	var existingArchived bool
	err := s.db.QueryRow(ctx, `SELECT name, base_url, archived_at IS NOT NULL FROM source_connections WHERE id = $1`, connectionID).
		Scan(&existingName, &existingRoot, &existingArchived)
	switch {
	case errors.Is(err, pgx.ErrNoRows):
		// fresh registration below
	case err != nil:
		return "", fmt.Errorf("load dingtalk connection: %w", err)
	default:
		// 同一个键解析到了另一个 root：**在用中**的行不能悄悄改写（可能是两个不同空间
		// 撞了同一个键）；已经归档的行可以就地改写 —— 这正是"去掉来源之后再用 dws
		// 重新获取"的语义。
		if !strings.HasPrefix(existingRoot, "spaces/"+in.RootID+"?") && !existingArchived {
			return "", ErrDingtalkSpaceIdentityConflict
		}
	}
	// 名字有 UNIQUE 约束，而**旧路径登记**的来源（例如早期直接建的钉钉知识库连接，
	// base_url 为空、id 与 dws 现在推导出的确定性 id 不同）会一直占着这个名字。
	// 归档行已经不显示在任何页面，就把它改名腾出位置（数据与审计一行不动）；
	// 仍在使用的同名来源则直接报可行动错误，绝不悄悄改名。
	var nameHolderID string
	var nameHolderArchived bool
	err = s.db.QueryRow(ctx, `SELECT id, archived_at IS NOT NULL FROM source_connections WHERE name = $1 AND id <> $2`, in.Name, connectionID).
		Scan(&nameHolderID, &nameHolderArchived)
	switch {
	case errors.Is(err, pgx.ErrNoRows):
	case err != nil:
		return "", fmt.Errorf("lookup dingtalk connection name: %w", err)
	default:
		if !nameHolderArchived {
			return "", ErrDingtalkSpaceNameTaken
		}
		if _, err := s.db.Exec(ctx, `
UPDATE source_connections SET name = $2, updated_at = $3, updated_by = $4 WHERE id = $1`,
			nameHolderID, archivedConnectionName(in.Name, nameHolderID), in.Now.UTC(), "dingtalk-registration"); err != nil {
			return "", fmt.Errorf("free archived dingtalk name: %w", err)
		}
	}
	// base_url stores a stable, non-secret locator (the root dentry uuid plus
	// the space id) so re-registration can verify identity and the UI can
	// explain what a DingTalk connection points at.
	locator := "spaces/" + in.RootID + "?space=" + in.SpaceID
	_, err = s.db.Exec(ctx, `
INSERT INTO source_connections (
    id, source_type, name, base_url, secret_ref, enabled,
    schedule, last_health, created_at, updated_at, updated_by
) VALUES ($1, 'dingtalk', $2, $3, '', TRUE, '{}'::jsonb, '{}'::jsonb, $4, $4, $5)
ON CONFLICT (id) DO UPDATE
SET name = EXCLUDED.name, base_url = EXCLUDED.base_url,
    -- 归档过的来源被重新登记（dws 重新获取）时原地复活：清掉归档标记，不建重复行 ——
    -- 目录快照与历史仍属于同一行。
    archived_at = NULL, archived_by = '',
    updated_at = EXCLUDED.updated_at, updated_by = EXCLUDED.updated_by`,
		connectionID, in.Name, locator, in.Now.UTC(), "dingtalk-registration")
	if err != nil {
		if isUniqueViolation(err) {
			return "", fmt.Errorf("dingtalk space registration: name %q already used by another connection", in.Name)
		}
		return "", fmt.Errorf("register dingtalk space: %w", err)
	}
	return connectionID, nil
}

// archivedConnectionName 给"已归档但名字仍占着唯一槽位"的行一个确定性新名字：
// 内建短 ID 后缀保证唯一，并按 name 列（255）截断原名。
func archivedConnectionName(name, connectionID string) string {
	suffix := "（已下架 · " + shortConnectionIDSuffix(connectionID) + "）"
	runes := []rune(name)
	if limit := 255 - len([]rune(suffix)); len(runes) > limit && limit > 0 {
		runes = runes[:limit]
	}
	return string(runes) + suffix
}

func shortConnectionIDSuffix(connectionID string) string {
	trimmed := strings.TrimPrefix(connectionID, "sc-")
	if len(trimmed) > 8 {
		trimmed = trimmed[:8]
	}
	if trimmed == "" {
		trimmed = "unknown"
	}
	return trimmed
}