90CNServer/pvf/internal/annotations/loader.go

package annotations

import (
	"bytes"
	"encoding/json"
	"fmt"
	"io"
	"path"
	"sort"
	"strings"
	"sync"

	appconfig "pvfine/config"
)

var (
	defaultEngineOnce sync.Once
	defaultEngine     *Engine
	defaultEngineErr  error
)

// LoadDefault parses the embedded rule set once and reuses the immutable
// engine. The generated path notes make the document large enough that a parse
// per preview or auxiliary core would be measurable.
func LoadDefault() (*Engine, error) {
	defaultEngineOnce.Do(func() {
		defaultEngine, defaultEngineErr = loadDefault()
	})
	return defaultEngine, defaultEngineErr
}

// generatedRuleFile is one bulk rule set that ships next to annotations.json.
type generatedRuleFile struct {
	// name is the sibling file that overrides the embedded copy at runtime.
	name  string
	label string
	data  []byte
}

// generatedRuleFiles lists the generated rule sets merged after the hand-written
// rules. Order only decides the primary title of a path or section, so the
// curated rules in annotations.json always stay first.
func generatedRuleFiles() []generatedRuleFile {
	return []generatedRuleFile{
		{name: "paths.json", label: "path 标注", data: appconfig.PathsJSON},
		{name: "sections.json", label: "section 标注", data: appconfig.SectionsJSON},
	}
}

func loadDefault() (*Engine, error) {
	document, err := parseDocument(appconfig.AnnotationsJSON)
	if err != nil {
		return nil, err
	}
	for _, generated := range generatedRuleFiles() {
		if err := mergeGeneratedRules(&document, generated.data, generated.label); err != nil {
			return nil, err
		}
	}
	lists, err := ParseLists(appconfig.ListsJSON)
	if err != nil {
		return nil, err
	}
	document.Relations = lists.Relations
	if err := Validate(document); err != nil {
		return nil, err
	}
	return Compile(document)
}

// mergeGeneratedRules appends a generated rule set after the hand-written rules,
// so an explicit rule keeps the primary title for a path or section. Rules whose
// ID is already present are skipped, which keeps merging idempotent when a user
// mirrors a generated file by hand.
func mergeGeneratedRules(document *Document, data []byte, label string) error {
	if len(data) == 0 {
		return nil
	}
	generated, err := parseDocument(data)
	if err != nil {
		return fmt.Errorf("解析 %s 失败: %w", label, err)
	}
	known := make(map[string]struct{}, len(document.Rules))
	for _, rule := range document.Rules {
		known[rule.ID] = struct{}{}
	}
	for _, rule := range generated.Rules {
		if _, exists := known[rule.ID]; exists {
			continue
		}
		known[rule.ID] = struct{}{}
		document.Rules = append(document.Rules, rule)
	}
	return nil
}

func Parse(data []byte) (Document, error) {
	document, err := parseDocument(data)
	if err != nil {
		return Document{}, err
	}
	if err := Validate(document); err != nil {
		return Document{}, err
	}
	return document, nil
}

func ParseRules(data []byte) (Document, error) {
	return parseDocument(data)
}

func parseDocument(data []byte) (Document, error) {
	var document Document
	decoder := json.NewDecoder(bytes.NewReader(data))
	decoder.DisallowUnknownFields()
	if err := decoder.Decode(&document); err != nil {
		return Document{}, fmt.Errorf("解析标注规则失败: %w", err)
	}
	if err := decoder.Decode(&struct{}{}); err != io.EOF {
		if err == nil {
			return Document{}, fmt.Errorf("标注规则只能包含一个 JSON 文档")
		}
		return Document{}, fmt.Errorf("解析标注规则失败: %w", err)
	}
	normalizeImageTargets(&document)
	return document, nil
}

func ParseLists(data []byte) (ListDocument, error) {
	var lists ListDocument
	decoder := json.NewDecoder(bytes.NewReader(data))
	decoder.DisallowUnknownFields()
	if err := decoder.Decode(&lists); err != nil {
		return ListDocument{}, fmt.Errorf("解析列表配置失败: %w", err)
	}
	if err := decoder.Decode(&struct{}{}); err != io.EOF {
		if err == nil {
			return ListDocument{}, fmt.Errorf("列表配置只能包含一个 JSON 文档")
		}
		return ListDocument{}, fmt.Errorf("解析列表配置失败: %w", err)
	}
	document := Document{Version: lists.Version, Relations: lists.Relations}
	if err := Validate(document); err != nil {
		return ListDocument{}, err
	}
	return lists, nil
}

func Marshal(document Document) ([]byte, error) {
	normalizeImageTargets(&document)
	if err := Validate(document); err != nil {
		return nil, err
	}
	data, err := json.MarshalIndent(document, "", "  ")
	if err != nil {
		return nil, err
	}
	return append(data, '\n'), nil
}

func MarshalRules(document Document) ([]byte, error) {
	normalizeImageTargets(&document)
	if err := Validate(document); err != nil {
		return nil, err
	}
	document.Relations = nil
	data, err := json.MarshalIndent(document, "", "  ")
	if err != nil {
		return nil, err
	}
	return append(data, '\n'), nil
}

func MarshalLists(lists ListDocument) ([]byte, error) {
	document := Document{Version: lists.Version, Relations: lists.Relations}
	if err := Validate(document); err != nil {
		return nil, err
	}
	data, err := json.MarshalIndent(lists, "", "  ")
	if err != nil {
		return nil, err
	}
	return append(data, '\n'), nil
}

func Validate(document Document) error {
	normalizeImageTargets(&document)
	problems := make([]string, 0)
	validateVersions := func(prefix string, versions []string) {
		for _, version := range versions {
			if !ValidPVFVersion(version) {
				problems = append(problems, fmt.Sprintf("%s.pvfVersions 包含无效版本 %q，只允许 90US、90CN、110US、115US、115CN、100CN、TW", prefix, version))
			}
		}
	}
	for i, rule := range document.Rules {
		validateVersions(fmt.Sprintf("rules[%d]", i), rule.PVFVersions)
	}
	for i, field := range document.Fields {
		validateVersions(fmt.Sprintf("fields[%d]", i), field.PVFVersions)
	}

	if document.Version != 1 {
		problems = append(problems, fmt.Sprintf("version 必须为 1，当前为 %d", document.Version))
	}

	relationNames := make([]string, 0, len(document.Relations))
	for name := range document.Relations {
		relationNames = append(relationNames, name)
	}
	sort.Strings(relationNames)
	for _, name := range relationNames {
		relation := document.Relations[name]
		prefix := fmt.Sprintf("relations.%s", name)
		if strings.TrimSpace(name) == "" {
			problems = append(problems, "relation 名称不能为空")
		}
		kind := relation.Kind
		if kind == "" {
			kind = "list"
		}
		if kind != "list" && kind != "contextual" && kind != "union" {
			problems = append(problems, prefix+".kind 必须是 list、contextual 或 union")
		}
		if kind == "union" {
			if len(relation.Relations) == 0 {
				problems = append(problems, prefix+".relations 不能为空")
			}
			for _, member := range relation.Relations {
				if strings.TrimSpace(member) == "" || member == name {
					problems = append(problems, prefix+".relations 包含无效关联类型")
					break
				}
			}
			continue
		}
		if kind == "list" && strings.TrimSpace(relation.ListPath) == "" {
			problems = append(problems, prefix+".listPath 不能为空")
		}
		if relation.IDToken < 0 || relation.PathToken < 0 || relation.ContextToken < 0 {
			problems = append(problems, prefix+" 的 token 下标不能为负数")
		}
		if kind == "contextual" {
			if len(relation.ContextPaths) == 0 {
				problems = append(problems, prefix+".contextPaths 不能为空")
			}
			for context, listPath := range relation.ContextPaths {
				if strings.TrimSpace(context) == "" || strings.TrimSpace(listPath) == "" {
					problems = append(problems, prefix+".contextPaths 的键和值不能为空")
					break
				}
			}
			if relation.RecordTokens <= relation.IDToken || relation.RecordTokens <= relation.PathToken {
				problems = append(problems, prefix+".recordTokens 必须覆盖 idToken 和 pathToken")
			}
		}
		recordTokens := relation.RecordTokens
		if recordTokens == 0 {
			recordTokens = max(relation.IDToken, relation.PathToken) + 1
		}
		if recordTokens <= relation.IDToken || recordTokens <= relation.PathToken {
			problems = append(problems, prefix+".recordTokens 必须覆盖 idToken 和 pathToken")
		}
		if strings.TrimSpace(relation.NameSection) == "" {
			problems = append(problems, prefix+".nameSection 不能为空")
		}
	}
	unionState := make(map[string]uint8)
	var visitUnion func(string)
	visitUnion = func(name string) {
		if unionState[name] == 2 {
			return
		}
		if unionState[name] == 1 {
			problems = append(problems, "relations."+name+" 存在循环引用")
			return
		}
		relation, ok := document.Relations[name]
		if !ok || relation.Kind != "union" {
			unionState[name] = 2
			return
		}
		unionState[name] = 1
		for _, member := range relation.Relations {
			if _, ok := document.Relations[member]; !ok {
				problems = append(problems, fmt.Sprintf("relations.%s 引用了不存在的 relation: %s", name, member))
				continue
			}
			visitUnion(member)
		}
		unionState[name] = 2
	}
	for _, name := range relationNames {
		visitUnion(name)
	}

	seenIDs := make(map[string]bool)
	for i, originalRule := range document.Rules {
		rule := originalRule
		if strings.TrimSpace(originalRule.Field) != "" {
			resolved, err := resolveFieldRule(originalRule, document.Fields)
			if err != nil {
				problems = append(problems, fmt.Sprintf("rules[%d]: %s", i, err))
				continue
			}
			rule = resolved
		}
		prefix := fmt.Sprintf("rules[%d]", i)
		if strings.TrimSpace(rule.ID) == "" {
			problems = append(problems, prefix+".id 不能为空")
		} else if seenIDs[rule.ID] {
			problems = append(problems, prefix+".id 重复: "+rule.ID)
		}
		seenIDs[rule.ID] = true

		for _, extension := range rule.Match.Extensions {
			if extension == "" || !strings.HasPrefix(extension, ".") || strings.ContainsAny(extension, "/\\") {
				problems = append(problems, prefix+".match.extensions 必须是以点开头的文件后缀")
				break
			}
		}
		if rule.Match.Glob != "" {
			if err := validateGlob(rule.Match.Glob); err != nil {
				problems = append(problems, prefix+".match.glob 无效: "+err.Error())
			}
		}

		switch rule.Target.Kind {
		case "path":
			if rule.Target.Offset != 0 || rule.Target.GroupOffset != 0 || len(rule.Target.StandaloneValues) > 0 || rule.Target.TokensPerLineIndex != nil {
				problems = append(problems, prefix+" 的 offset/tokensPerLineIndex 只允许用于 token 标注")
			}
			if rule.Annotation.Type != "text" {
				problems = append(problems, prefix+" 的 path 标注只支持 text 类型")
			}
		case "section":
			if rule.Target.Offset != 0 || rule.Target.GroupOffset != 0 || len(rule.Target.StandaloneValues) > 0 || rule.Target.TokensPerLineIndex != nil {
				problems = append(problems, prefix+" 的 offset/tokensPerLineIndex 只允许用于 token 标注")
			}
			if strings.TrimSpace(rule.Target.Section) == "" {
				problems = append(problems, prefix+".target.section 不能为空")
			}
			if rule.Annotation.Type != "text" {
				problems = append(problems, prefix+" 的 section 标注只支持 text 类型")
			}
		case "token":
			if rule.Annotation.Type != "image" && rule.Target.ImagePathToken != nil {
				problems = append(problems, prefix+".target.imagePathToken 只允许用于 image 标注")
			}
			if strings.TrimSpace(rule.Target.Section) == "" {
				problems = append(problems, prefix+".target.section 不能为空")
			}
			if (rule.Target.Index == nil) == (rule.Target.Range == nil) {
				problems = append(problems, prefix+" 的 token 目标必须且只能配置 index 或 range")
			}
			if rule.Target.Index != nil && *rule.Target.Index < 0 {
				problems = append(problems, prefix+".target.index 不能为负数")
			}
			if rule.Target.Offset < 0 {
				problems = append(problems, prefix+".target.offset 不能为负数")
			}
			if rule.Target.GroupOffset < 0 {
				problems = append(problems, prefix+".target.groupOffset 不能为负数")
			}
			if rule.Target.TokensPerLineIndex != nil && *rule.Target.TokensPerLineIndex < 0 {
				problems = append(problems, prefix+".target.tokensPerLineIndex 不能为负数")
			}
			if rule.Target.RecordTokens < 0 {
				problems = append(problems, prefix+".target.recordTokens 不能为负数")
			}
			if rule.Target.RecordTokens == 0 && rule.Target.ContextIndex != nil {
				problems = append(problems, prefix+".target.contextIndex 需要同时配置 recordTokens")
			}
			if rule.Target.RecordTokens > 0 {
				if rule.Target.Index == nil {
					problems = append(problems, prefix+".target.recordTokens 需要配合单个 index 使用")
				} else if *rule.Target.Index >= rule.Target.RecordTokens {
					problems = append(problems, prefix+".target.index 必须小于 recordTokens")
				}
				if rule.Target.ContextIndex != nil && (*rule.Target.ContextIndex < 0 || *rule.Target.ContextIndex >= rule.Target.RecordTokens) {
					problems = append(problems, prefix+".target.contextIndex 必须位于 recordTokens 范围内")
				}
			}
			if rule.Target.Offset != 0 || rule.Target.GroupOffset != 0 || len(rule.Target.StandaloneValues) > 0 || rule.Target.TokensPerLineIndex != nil {
				if rule.Target.RecordTokens <= 0 {
					problems = append(problems, prefix+".target.offset/tokensPerLineIndex 需要配置正数 recordTokens 作为回退值")
				}
				if rule.Target.Index == nil || rule.Target.Range != nil {
					problems = append(problems, prefix+".target.offset/tokensPerLineIndex 需要配合单个 index 使用")
				}
			}
			if rule.Target.Range != nil {
				if rule.Target.Range.Start < 0 || rule.Target.Range.EndExclusive <= rule.Target.Range.Start {
					problems = append(problems, prefix+".target.range 必须满足 0 <= start < endExclusive")
				}
				if rule.Annotation.Type != "text" {
					problems = append(problems, prefix+" 的 token 范围只支持 text 类型")
				}
			}
			if rule.Annotation.Type == "image" {
				if rule.Target.Index == nil || rule.Target.Range != nil {
					problems = append(problems, prefix+" 的 image 标注必须使用单个路径 token")
				}
				if rule.Target.RecordTokens <= 0 {
					problems = append(problems, prefix+" 的 image 标注必须配置正数 target.recordTokens")
				}
				if rule.Target.ImagePathToken == nil {
					problems = append(problems, prefix+" 的 image 标注必须配置 target.imagePathToken")
				} else if rule.Target.RecordTokens > 0 && (*rule.Target.ImagePathToken < 0 || *rule.Target.ImagePathToken >= rule.Target.RecordTokens) {
					problems = append(problems, prefix+".target.imagePathToken 必须位于 recordTokens 范围内")
				}
				if rule.Target.Index != nil && rule.Target.ImagePathToken != nil && *rule.Target.Index == *rule.Target.ImagePathToken {
					problems = append(problems, prefix+" 的 imagePathToken 不能与图片索引 token 相同")
				}
				if rule.Target.ContextIndex != nil {
					problems = append(problems, prefix+" 的 image 标注不支持 contextIndex")
				}
			}
		default:
			problems = append(problems, prefix+".target.kind 必须是 path、section 或 token")
		}

		switch rule.Annotation.Type {
		case "text":
		case "image":
		case "path":
			if root := strings.TrimSpace(strings.ReplaceAll(rule.Annotation.PathRoot, "\\", "/")); root != "" && (strings.HasPrefix(root, "/") || root == ".." || strings.HasPrefix(root, "../") || strings.Contains(root, "/../")) {
				problems = append(problems, prefix+".annotation.pathRoot 必须是归档内的相对目录")
			}
		case "enum":
			if len(rule.Annotation.Values) == 0 {
				problems = append(problems, prefix+".annotation.values 不能为空")
			}
		case "reference":
			relation, ok := document.Relations[rule.Annotation.Relation]
			if !ok {
				problems = append(problems, prefix+" 引用了不存在的 relation: "+rule.Annotation.Relation)
			} else if relation.Kind == "contextual" {
				if rule.Target.RecordTokens == 0 {
					problems = append(problems, prefix+" 的 contextual relation 需要配置 target.recordTokens")
				}
				contextIndex := relation.ContextToken
				if rule.Target.ContextIndex != nil {
					contextIndex = *rule.Target.ContextIndex
				}
				if rule.Target.RecordTokens > 0 && (contextIndex < 0 || contextIndex >= rule.Target.RecordTokens) {
					problems = append(problems, prefix+" 的 contextual relation 上下文 token 下标必须位于 target.recordTokens 范围内")
				}
			}
		default:
			problems = append(problems, prefix+".annotation.type 必须是 text、image、enum、reference 或 path")
		}
		if rule.Annotation.Type != "path" && rule.Annotation.PathRoot != "" {
			problems = append(problems, prefix+".annotation.pathRoot 只允许用于 path 标注")
		}
		if rule.Annotation.InlineImage && rule.Annotation.Type != "image" {
			problems = append(problems, prefix+".annotation.inlineImage 只允许用于 image 标注")
		}
	}

	seenFieldIDs := make(map[string]bool, len(document.Fields))
	for i, field := range document.Fields {
		prefix := fmt.Sprintf("fields[%d]", i)
		if strings.TrimSpace(field.ID) == "" {
			problems = append(problems, prefix+".id 不能为空")
		} else if seenFieldIDs[field.ID] {
			problems = append(problems, prefix+".id 重复: "+field.ID)
		}
		seenFieldIDs[field.ID] = true
		// Reuse the mature rule validator for the shared target and annotation
		// schema. Fields are not rules at runtime, so this temporary document
		// cannot recurse into the fields collection.
		if err := Validate(Document{
			Version: 1, Relations: document.Relations,
			Rules: []Rule{{ID: field.ID, Match: field.Match, Target: field.Target, Annotation: field.Annotation}},
		}); err != nil {
			problems = append(problems, prefix+": "+err.Error())
		}
		if field.Preview != nil {
			if len(field.Preview.providerNames()) == 0 {
				problems = append(problems, prefix+".preview.provider 不能为空")
			}
			if strings.TrimSpace(field.Preview.Role) == "" {
				problems = append(problems, prefix+".preview.role 不能为空")
			}
			if strings.TrimSpace(field.Preview.Group) == "" {
				problems = append(problems, prefix+".preview.group 不能为空")
			}
			if strings.TrimSpace(field.Preview.Format) == "" {
				problems = append(problems, prefix+".preview.format 不能为空")
			}
		}
	}

	if len(problems) > 0 {
		return fmt.Errorf("标注规则校验失败:\n- %s", strings.Join(problems, "\n- "))
	}
	return nil
}

// normalizeImageTargets migrates the original image rule shape. Originally
// target.index was the IMG path token and target.imageIndexToken was the
// numeric image index token. Image annotations now anchor on the numeric
// index and use imagePathToken for the associated IMG path token.
func normalizeImageTargets(document *Document) {
	if document == nil {
		return
	}
	document.Rules = append([]Rule(nil), document.Rules...)
	for i := range document.Rules {
		normalizeImageTarget(&document.Rules[i].Target, document.Rules[i].Annotation.Type)
	}
	document.Fields = append([]FieldDefinition(nil), document.Fields...)
	for i := range document.Fields {
		normalizeImageTarget(&document.Fields[i].Target, document.Fields[i].Annotation.Type)
	}
}

func normalizeImageTarget(target *TargetSpec, annotationType string) {
	if target == nil || annotationType != "image" || target.ImageIndexToken == nil {
		return
	}
	if target.ImagePathToken == nil {
		pathToken := target.Index
		target.Index = target.ImageIndexToken
		target.ImagePathToken = pathToken
	}
	target.ImageIndexToken = nil
}

func validateGlob(pattern string) error {
	for _, segment := range strings.Split(normalizePath(pattern), "/") {
		if segment == "**" {
			continue
		}
		if _, err := path.Match(segment, segment); err != nil {
			return err
		}
	}
	return nil
}

func max(a, b int) int {
	if a > b {
		return a
	}
	return b
}