history: index short terms and resolve mentions

This commit is contained in:
2026-08-06 22:04:13 +08:00
parent c8cdea6fab
commit 19475b3ee8
9 changed files with 433 additions and 102 deletions
+4 -3
View File
@@ -238,10 +238,11 @@ searchHistoryMaxRecords: 5000
聊天记录与联系人快照保存在插件数据目录的 `chat-history.sqlite` 中,并使用 SQLite WAL 模式支持记录与查询并行进行。 聊天记录与联系人快照保存在插件数据目录的 `chat-history.sqlite` 中,并使用 SQLite WAL 模式支持记录与查询并行进行。
数据库由插件在首次启动时自动创建和维护,无需安装额外的聊天记录插件。 数据库由插件在首次启动时自动创建和维护,无需安装额外的聊天记录插件。
聊天记录搜索使用 SQLite FTS5 中文三元索引。升级已有数据库时只会新增独立的派生搜索文本表和索引表, 聊天记录搜索使用 SQLite FTS5 中文二元 + 三元索引。升级已有数据库时只会新增独立的派生搜索文本表和索引表,
不会改列、重建或删除 `message_record` 原始记录;旧记录的搜索文本会在插件启动后分批后台回填。回填期间新消息仍会正常保存, 不会改列、重建或删除 `message_record` 原始记录;旧记录的搜索文本会在插件启动后分批后台回填。回填期间新消息仍会正常保存,
搜索结果会明确标注索引尚未完成。搜索工具默认只查询当前群聊或当前私聊,明确指定 `from``to` 后可以跨越任意历史时间范围, 搜索结果会明确标注索引尚未完成。搜索工具默认只查询当前群聊或当前私聊,明确指定 `from``to` 后可以跨越任意历史时间范围,
并通过 `nextCursor` 分页。升级后新收到的消息还会记录发送者在该会话中使用过的名称,名称搜索不依赖对方仍是当前群成员 并通过 `nextCursor` 分页。普通文字和可解析的 `@` 目标会合并检索,不要求模型预先判断查询类型
升级后新收到的消息还会记录发送者在该会话中使用过的名称,名称搜索不依赖对方仍是当前群成员;
`getChatHistoryContext` 可按 `messageId` 获取命中消息前后的对话上下文。 `getChatHistoryContext` 可按 `messageId` 获取命中消息前后的对话上下文。
联系人刷新在后台通过 Overflow 的公开 `RemoteBot.executeAction` 调用 OneBot 联系人刷新在后台通过 Overflow 的公开 `RemoteBot.executeAction` 调用 OneBot
@@ -506,7 +507,7 @@ fallbackCooldownMinutes: 5
14. **GetChatHistoryContext** - 根据历史消息 ID 获取目标消息前后的对话上下文 14. **GetChatHistoryContext** - 根据历史消息 ID 获取目标消息前后的对话上下文
15. **QueryUserProfile** - 按当前发送者、QQ 号、群名片、昵称或好友备注读取长期画像,并针对该联系人按需读取公开资料卡 15. **QueryUserProfile** - 按当前发送者、QQ 号、群名片、昵称或好友备注读取长期画像,并针对该联系人按需读取公开资料卡
`SearchChatHistory` 的常用参数只有消息内容、发送者、时间范围和翻页游标;结果默认按最近消息返回, `SearchChatHistory` 的常用参数只有消息内容、发送者、时间范围和翻页游标;普通文字和可解析的 `@` 目标会合并检索,结果默认按最近消息返回,
单页大小和匹配策略由插件控制。`senderId` 与 `senderName` 可以同时提供,此时 QQ 号作为精确筛选条件;有 QQ 号时名称不参与筛选。 单页大小和匹配策略由插件控制。`senderId` 与 `senderName` 可以同时提供,此时 QQ 号作为精确筛选条件;有 QQ 号时名称不参与筛选。
### GitHub 查询工具 ### GitHub 查询工具
@@ -42,6 +42,7 @@ data class ChatHistoryCursor(
data class ChatHistorySearchRequest( data class ChatHistorySearchRequest(
val subject: ChatHistorySubject, val subject: ChatHistorySubject,
val query: String? = null, val query: String? = null,
val atTargetIds: Set<Long> = emptySet(),
val matchMode: ChatHistoryMatchMode = ChatHistoryMatchMode.ALL, val matchMode: ChatHistoryMatchMode = ChatHistoryMatchMode.ALL,
val fromId: Long? = null, val fromId: Long? = null,
val start: Int? = null, val start: Int? = null,
+53 -14
View File
@@ -18,28 +18,49 @@ object ChatHistorySearchText {
private val json = Json { ignoreUnknownKeys = true } private val json = Json { ignoreUnknownKeys = true }
private val whitespace = Regex("\\s+") private val whitespace = Regex("\\s+")
fun extract(code: String): String { fun extract(code: String, atNames: Map<Long, String> = emptyMap()): String {
val rendered = runCatching { renderJsonCode(code) } val rendered = runCatching { renderJsonCode(code, atNames) }
.recoverCatching { decodeMessageCode(code).content } .recoverCatching { decodeMessageCode(code).content }
.getOrDefault("") .getOrDefault("")
return rendered.normalize().takeUtf16Safely(MAX_INDEXED_CHARS) return rendered.normalize().takeUtf16Safely(MAX_INDEXED_CHARS)
} }
private fun renderJsonCode(code: String): String { fun extractAtTargets(code: String): Set<Long> = runCatching {
val messages = json.parseToJsonElement(code) as? JsonArray val messages = json.parseToJsonElement(code) as? JsonArray ?: return@runCatching emptySet()
?: throw IllegalArgumentException("消息记录不是 JSON array") buildSet { collectAtTargets(messages, this) }
return renderMessages(messages) }.getOrDefault(emptySet())
fun bigrams(text: String): String {
val codePoints = text.lowercase().codePoints().toArray()
if (codePoints.size < 2) return ""
return buildString(codePoints.size * 3) {
var tokenCount = 0
for (index in 0 until codePoints.lastIndex) {
if (Character.isWhitespace(codePoints[index]) || Character.isWhitespace(codePoints[index + 1])) {
continue
}
if (tokenCount++ > 0) append(' ')
appendCodePoint(codePoints[index])
appendCodePoint(codePoints[index + 1])
}
}
} }
private fun renderMessages(messages: JsonArray): String = messages.joinToString("") { element -> private fun renderJsonCode(code: String, atNames: Map<Long, String>): String {
val messages = json.parseToJsonElement(code) as? JsonArray
?: throw IllegalArgumentException("消息记录不是 JSON array")
return renderMessages(messages, atNames)
}
private fun renderMessages(messages: JsonArray, atNames: Map<Long, String>): String = messages.joinToString("") { element ->
val message = element as? JsonObject ?: return@joinToString "" val message = element as? JsonObject ?: return@joinToString ""
when (val type = message.string("type")) { when (val type = message.string("type")) {
"PlainText" -> message.string("content").orEmpty() "PlainText" -> message.string("content").orEmpty()
"At" -> message.long("target")?.let { "@$it" }.orEmpty() "At" -> message.long("target")?.let { target -> "@${atNames[target] ?: target}" }.orEmpty()
"AtAll" -> "@全体成员" "AtAll" -> "@全体成员"
"Image", "FlashImage" -> if (message.boolean("isEmoji") == true) "[表情包]" else "[图片]" "Image", "FlashImage" -> if (message.boolean("isEmoji") == true) "[表情包]" else "[图片]"
"QuoteReply" -> renderQuote(message) "QuoteReply" -> renderQuote(message, atNames)
"ForwardMessage" -> renderForward(message) "ForwardMessage" -> renderForward(message, atNames)
"MessageOrigin", "MessageSource", "ShowImageFlag" -> "" "MessageOrigin", "MessageSource", "ShowImageFlag" -> ""
"Face", "MarketFace", "VipFace" -> "[表情]" "Face", "MarketFace", "VipFace" -> "[表情]"
"Audio" -> "[语音]" "Audio" -> "[语音]"
@@ -51,18 +72,18 @@ object ChatHistorySearchText {
} }
} }
private fun renderQuote(message: JsonObject): String { private fun renderQuote(message: JsonObject, atNames: Map<Long, String>): String {
val source = message["source"] as? JsonObject ?: return "[引用消息]" val source = message["source"] as? JsonObject ?: return "[引用消息]"
val author = source.long("fromId")?.toString() ?: "其他用户" val author = source.long("fromId")?.toString() ?: "其他用户"
val original = (source["originalMessage"] as? JsonArray) val original = (source["originalMessage"] as? JsonArray)
?.let(::renderMessages) ?.let { renderMessages(it, atNames) }
.orEmpty() .orEmpty()
.normalize() .normalize()
.takeUtf16Safely(MAX_QUOTED_CHARS) .takeUtf16Safely(MAX_QUOTED_CHARS)
return "[引用 $author: $original]" return "[引用 $author: $original]"
} }
private fun renderForward(message: JsonObject): String = buildString { private fun renderForward(message: JsonObject, atNames: Map<Long, String>): String = buildString {
append("[转发消息]") append("[转发消息]")
val nodes = message["nodeList"] as? JsonArray ?: return@buildString val nodes = message["nodeList"] as? JsonArray ?: return@buildString
nodes.take(MAX_FORWARD_NODES).forEach { element -> nodes.take(MAX_FORWARD_NODES).forEach { element ->
@@ -71,7 +92,7 @@ object ChatHistorySearchText {
val chain = node["messageChain"] as? JsonArray val chain = node["messageChain"] as? JsonArray
append(' ').append(sender).append(": ") append(' ').append(sender).append(": ")
append( append(
chain?.let(::renderMessages) chain?.let { renderMessages(it, atNames) }
.orEmpty() .orEmpty()
.normalize() .normalize()
.takeUtf16Safely(MAX_FORWARD_NODE_CHARS) .takeUtf16Safely(MAX_FORWARD_NODE_CHARS)
@@ -80,6 +101,24 @@ object ChatHistorySearchText {
if (nodes.size > MAX_FORWARD_NODES) append(" ...[转发内容截断]") if (nodes.size > MAX_FORWARD_NODES) append(" ...[转发内容截断]")
} }
private fun collectAtTargets(messages: JsonArray, targets: MutableSet<Long>) {
messages.forEach { element ->
val message = element as? JsonObject ?: return@forEach
when (message.string("type")) {
"At" -> message.long("target")?.let(targets::add)
"QuoteReply" -> {
val original = (message["source"] as? JsonObject)
?.get("originalMessage") as? JsonArray
original?.let { collectAtTargets(it, targets) }
}
"ForwardMessage" -> (message["nodeList"] as? JsonArray).orEmpty().forEach { node ->
((node as? JsonObject)?.get("messageChain") as? JsonArray)
?.let { collectAtTargets(it, targets) }
}
}
}
}
private fun String.normalize(): String = whitespace.replace(this, " ") private fun String.normalize(): String = whitespace.replace(this, " ")
.trim() .trim()
.replaceUnpairedSurrogates() .replaceUnpairedSurrogates()
+217 -54
View File
@@ -24,13 +24,16 @@ import java.sql.ResultSet
* 写连接由 [writeLock] 串行保护;SQLite 使用 WAL + NORMAL synchronous,允许并发读取。 * 写连接由 [writeLock] 串行保护;SQLite 使用 WAL + NORMAL synchronous,允许并发读取。
*/ */
object ChatHistoryStore { object ChatHistoryStore {
private const val SCHEMA_VERSION = 3 private const val SCHEMA_VERSION = 4
private const val BUSY_TIMEOUT_MS = 30_000 private const val BUSY_TIMEOUT_MS = 30_000
private const val SEARCH_FTS_TABLE = "message_record_fts" private const val SEARCH_FTS_TABLE = "message_record_fts"
private const val SEARCH_BIGRAM_FTS_TABLE = "message_record_bigram_fts"
private const val SEARCH_TEXT_TABLE = "message_record_search" private const val SEARCH_TEXT_TABLE = "message_record_search"
private const val SENDER_ALIAS_TABLE = "chat_history_sender_alias" private const val SENDER_ALIAS_TABLE = "chat_history_sender_alias"
private const val SEARCH_BACKFILL_CURSOR_KEY = "search_backfill_cursor" private const val SEARCH_BACKFILL_CURSOR_KEY = "search_backfill_cursor"
private const val SEARCH_INDEX_COMPLETE_KEY = "search_index_complete" private const val SEARCH_INDEX_COMPLETE_KEY = "search_index_complete"
private const val SEARCH_BIGRAM_VERSION_KEY = "search_bigram_version"
private const val SEARCH_BIGRAM_VERSION = "1"
private const val DEFAULT_BACKFILL_BATCH_SIZE = 250 private const val DEFAULT_BACKFILL_BATCH_SIZE = 250
private const val MAX_PAGE_SIZE = 200 private const val MAX_PAGE_SIZE = 200
private const val MAX_CONTEXT_MESSAGES = 100 private const val MAX_CONTEXT_MESSAGES = 100
@@ -43,6 +46,8 @@ object ChatHistoryStore {
@Volatile @Volatile
private var searchIndexAvailable = false private var searchIndexAvailable = false
@Volatile @Volatile
private var bigramIndexAvailable = false
@Volatile
private var searchIndexReady = false private var searchIndexReady = false
private lateinit var databaseFile: File private lateinit var databaseFile: File
private var writeConnection: Connection? = null private var writeConnection: Connection? = null
@@ -79,6 +84,7 @@ object ChatHistoryStore {
writeConnection = connection writeConnection = connection
initialized = true initialized = true
searchIndexAvailable = hasSearchIndex(connection) searchIndexAvailable = hasSearchIndex(connection)
bigramIndexAvailable = hasBigramSearchIndex(connection)
searchIndexReady = searchIndexAvailable && readMeta(connection, SEARCH_INDEX_COMPLETE_KEY) == "1" searchIndexReady = searchIndexAvailable && readMeta(connection, SEARCH_INDEX_COMPLETE_KEY) == "1"
} catch (cause: Throwable) { } catch (cause: Throwable) {
connection.close() connection.close()
@@ -102,6 +108,7 @@ object ChatHistoryStore {
writeConnection = null writeConnection = null
initialized = false initialized = false
searchIndexAvailable = false searchIndexAvailable = false
bigramIndexAvailable = false
searchIndexReady = false searchIndexReady = false
warningLogger = null warningLogger = null
} }
@@ -299,17 +306,17 @@ object ChatHistoryStore {
} }
val query = request.query?.trim().orEmpty() val query = request.query?.trim().orEmpty()
val textPredicate = buildTextPredicate(query, request.matchMode, parameters) val textPredicate = buildTextPredicate(query, request.matchMode, request.atTargetIds, parameters)
if ((textPredicate.usesFts || textPredicate.usesSearchText) && !isSearchIndexAvailable) { if ((textPredicate.usesFts || textPredicate.usesSearchText) && !isSearchIndexAvailable) {
error("聊天记录全文索引尚未就绪") error("聊天记录全文索引尚未就绪")
} }
if (textPredicate.usesBigram && !bigramIndexAvailable) {
error("聊天记录二元索引尚未就绪")
}
val joins = buildString { val joins = buildString {
if (textPredicate.usesSearchText) { if (textPredicate.usesSearchText) {
append("JOIN $SEARCH_TEXT_TABLE search ON search.message_id = mr.id ") append("JOIN $SEARCH_TEXT_TABLE search ON search.message_id = mr.id ")
} }
if (textPredicate.usesFts) {
append("JOIN $SEARCH_FTS_TABLE ON $SEARCH_FTS_TABLE.rowid = mr.id")
}
} }
textPredicate.sql?.let(conditions::add) textPredicate.sql?.let(conditions::add)
@@ -499,16 +506,24 @@ object ChatHistoryStore {
var cursor = withWriteConnectionResult { connection -> var cursor = withWriteConnectionResult { connection ->
readMeta(connection, SEARCH_BACKFILL_CURSOR_KEY)?.toLongOrNull() ?: 0L readMeta(connection, SEARCH_BACKFILL_CURSOR_KEY)?.toLongOrNull() ?: 0L
} }
val hasBigramIndex = bigramIndexAvailable
while (shouldContinue()) { while (shouldContinue()) {
val batch = openReadConnection().use { connection -> val batch = openReadConnection().use { connection ->
val missingBigramExpression = if (hasBigramIndex) {
"NOT EXISTS (SELECT 1 FROM $SEARCH_BIGRAM_FTS_TABLE bigram WHERE bigram.rowid = mr.id)"
} else {
"0"
}
val missingBigramCondition = if (hasBigramIndex) "OR $missingBigramExpression" else ""
connection.prepareStatement( connection.prepareStatement(
""" """
SELECT mr.id, mr.code SELECT mr.id, mr.code,
search.message_id IS NULL AS missing_search,
$missingBigramExpression AS missing_bigram
FROM message_record mr FROM message_record mr
LEFT JOIN $SEARCH_TEXT_TABLE search ON search.message_id = mr.id
WHERE mr.id > ? WHERE mr.id > ?
AND NOT EXISTS ( AND (search.message_id IS NULL $missingBigramCondition)
SELECT 1 FROM $SEARCH_TEXT_TABLE search WHERE search.message_id = mr.id
)
ORDER BY mr.id ORDER BY mr.id
LIMIT ? LIMIT ?
""".trimIndent() """.trimIndent()
@@ -518,7 +533,14 @@ object ChatHistoryStore {
statement.executeQuery().use { results -> statement.executeQuery().use { results ->
buildList { buildList {
while (results.next()) { while (results.next()) {
add(results.getLong("id") to results.getString("code")) add(
SearchBackfillRow(
id = results.getLong("id"),
code = results.getString("code"),
missingSearch = results.getBoolean("missing_search"),
missingBigram = results.getBoolean("missing_bigram"),
)
)
} }
} }
} }
@@ -541,20 +563,39 @@ object ChatHistoryStore {
connection.prepareStatement( connection.prepareStatement(
"INSERT INTO $SEARCH_FTS_TABLE(rowid, search_text) VALUES (?, ?)" "INSERT INTO $SEARCH_FTS_TABLE(rowid, search_text) VALUES (?, ?)"
).use { insertFts -> ).use { insertFts ->
batch.forEach { (id, code) -> val insertBigram = if (hasBigramIndex) {
val text = ChatHistorySearchText.extract(code) connection.prepareStatement(
insertText.setLong(1, id) "INSERT INTO $SEARCH_BIGRAM_FTS_TABLE(rowid, bigram_text) VALUES (?, ?)"
)
} else {
null
}
try {
batch.forEach { row ->
val text = ChatHistorySearchText.extract(row.code)
if (row.missingSearch) {
insertText.setLong(1, row.id)
insertText.setString(2, text) insertText.setString(2, text)
insertText.addBatch() insertText.addBatch()
insertFts.setLong(1, id) insertFts.setLong(1, row.id)
insertFts.setString(2, text) insertFts.setString(2, text)
insertFts.addBatch() insertFts.addBatch()
} }
if (row.missingBigram) {
insertBigram?.setLong(1, row.id)
insertBigram?.setString(2, ChatHistorySearchText.bigrams(text))
insertBigram?.addBatch()
}
}
insertText.executeBatch() insertText.executeBatch()
insertFts.executeBatch() insertFts.executeBatch()
insertBigram?.executeBatch()
} finally {
insertBigram?.close()
} }
} }
cursor = batch.last().first }
cursor = batch.last().id
writeMeta(connection, SEARCH_BACKFILL_CURSOR_KEY, cursor.toString()) writeMeta(connection, SEARCH_BACKFILL_CURSOR_KEY, cursor.toString())
writeMeta(connection, SEARCH_INDEX_COMPLETE_KEY, "0") writeMeta(connection, SEARCH_INDEX_COMPLETE_KEY, "0")
connection.commit() connection.commit()
@@ -571,6 +612,13 @@ object ChatHistoryStore {
return processed return processed
} }
private data class SearchBackfillRow(
val id: Long,
val code: String,
val missingSearch: Boolean,
val missingBigram: Boolean,
)
internal fun insertRecord(record: ChatMessageRecord, senderName: String = "") { internal fun insertRecord(record: ChatMessageRecord, senderName: String = "") {
var recordId = 0L var recordId = 0L
var indexFailure: Throwable? = null var indexFailure: Throwable? = null
@@ -639,6 +687,15 @@ object ChatHistoryStore {
statement.setString(2, searchText) statement.setString(2, searchText)
statement.executeUpdate() statement.executeUpdate()
} }
if (bigramIndexAvailable) {
connection.prepareStatement(
"INSERT INTO $SEARCH_BIGRAM_FTS_TABLE(rowid, bigram_text) VALUES (?, ?)"
).use { statement ->
statement.setLong(1, recordId)
statement.setString(2, ChatHistorySearchText.bigrams(searchText))
statement.executeUpdate()
}
}
connection.commit() connection.commit()
} catch (cause: Throwable) { } catch (cause: Throwable) {
connection.rollback() connection.rollback()
@@ -810,6 +867,20 @@ object ChatHistoryStore {
} }
}.onFailure { warn("初始化 SQLite FTS5 聊天记录索引失败,消息记录仍会继续保存", it) } }.onFailure { warn("初始化 SQLite FTS5 聊天记录索引失败,消息记录仍会继续保存", it) }
runCatching {
connection.createStatement().use { statement ->
statement.executeUpdate(
"""
CREATE VIRTUAL TABLE IF NOT EXISTS $SEARCH_BIGRAM_FTS_TABLE USING fts5(
bigram_text,
content='',
tokenize='unicode61'
)
""".trimIndent()
)
}
}.onFailure { warn("初始化 SQLite FTS5 聊天记录二元索引失败,将回退到短词扫描", it) }
if (hasSearchIndex(connection)) { if (hasSearchIndex(connection)) {
val hasMessages = connection.createStatement().use { statement -> val hasMessages = connection.createStatement().use { statement ->
statement.executeQuery("SELECT 1 FROM message_record LIMIT 1").use(ResultSet::next) statement.executeQuery("SELECT 1 FROM message_record LIMIT 1").use(ResultSet::next)
@@ -825,6 +896,27 @@ object ChatHistoryStore {
statement.executeUpdate() statement.executeUpdate()
} }
} }
if (hasBigramSearchIndex(connection) &&
readMeta(connection, SEARCH_BIGRAM_VERSION_KEY) != SEARCH_BIGRAM_VERSION
) {
val hasMessages = connection.createStatement().use { statement ->
statement.executeQuery("SELECT 1 FROM message_record LIMIT 1").use(ResultSet::next)
}
val oldAutoCommitForMigration = connection.autoCommit
connection.autoCommit = false
try {
writeMeta(connection, SEARCH_BACKFILL_CURSOR_KEY, "0")
writeMeta(connection, SEARCH_INDEX_COMPLETE_KEY, if (hasMessages) "0" else "1")
writeMeta(connection, SEARCH_BIGRAM_VERSION_KEY, SEARCH_BIGRAM_VERSION)
connection.commit()
} catch (cause: Throwable) {
connection.rollback()
throw cause
} finally {
connection.autoCommit = oldAutoCommitForMigration
}
}
} }
private fun subjectConditions( private fun subjectConditions(
@@ -852,9 +944,40 @@ object ChatHistoryStore {
val sql: String?, val sql: String?,
val usesFts: Boolean, val usesFts: Boolean,
val usesSearchText: Boolean, val usesSearchText: Boolean,
val usesBigram: Boolean = false,
) )
private fun buildTextPredicate( private fun buildTextPredicate(
query: String,
mode: ChatHistoryMatchMode,
atTargetIds: Set<Long>,
parameters: MutableList<Any>,
): TextPredicate {
if (query.isBlank() && atTargetIds.isEmpty()) return TextPredicate(null, false, false)
val contentPredicate = if (query.isBlank()) {
TextPredicate(null, false, false)
} else {
buildContentTextPredicate(query, mode, parameters)
}
if (atTargetIds.isEmpty()) return contentPredicate
val mentionPredicate = atTargetIds.sorted().joinToString(" OR ", prefix = "(", postfix = ")") {
parameters += quoteFtsTerm("@$it")
ftsRowPredicate(SEARCH_FTS_TABLE)
}
val sql = when {
contentPredicate.sql == null -> mentionPredicate
else -> "(${contentPredicate.sql} OR $mentionPredicate)"
}
return TextPredicate(
sql = sql,
usesFts = true,
usesSearchText = contentPredicate.usesSearchText,
usesBigram = contentPredicate.usesBigram,
)
}
private fun buildContentTextPredicate(
query: String, query: String,
mode: ChatHistoryMatchMode, mode: ChatHistoryMatchMode,
parameters: MutableList<Any>, parameters: MutableList<Any>,
@@ -862,48 +985,77 @@ object ChatHistoryStore {
if (query.isBlank()) return TextPredicate(null, false, false) if (query.isBlank()) return TextPredicate(null, false, false)
val normalized = query.lowercase() val normalized = query.lowercase()
val terms = normalized.split(Regex("\\s+")).filter(String::isNotBlank) val terms = normalized.split(Regex("\\s+")).filter(String::isNotBlank)
val hasShortTerm = terms.any { it.codePointCount(0, it.length) < 3 } if (mode == ChatHistoryMatchMode.PHRASE) {
if (mode == ChatHistoryMatchMode.ANY && hasShortTerm) { val codePointCount = normalized.codePointCount(0, normalized.length)
val predicates = terms.map { return when {
parameters += it codePointCount >= 3 -> {
"instr(lower(search.search_text), ?) > 0" parameters += quoteFtsTerm(normalized)
TextPredicate(ftsRowPredicate(SEARCH_FTS_TABLE), usesFts = true, usesSearchText = false)
} }
return TextPredicate(predicates.joinToString(" OR ", prefix = "(", postfix = ")"), false, true) codePointCount == 2 && bigramIndexAvailable -> {
} parameters += quoteFtsTerm(normalized)
val ftsTerms = if (mode == ChatHistoryMatchMode.PHRASE) { TextPredicate(
listOf(normalized) ftsRowPredicate(SEARCH_BIGRAM_FTS_TABLE),
} else { usesFts = false,
terms usesSearchText = false,
} usesBigram = true,
val ftsOnly = ftsTerms.filter { it.codePointCount(0, it.length) >= 3 }
if (ftsOnly.isEmpty()) {
val predicates = terms.map {
parameters += it
"instr(lower(search.search_text), ?) > 0"
}
val joiner = if (mode == ChatHistoryMatchMode.ANY) " OR " else " AND "
return TextPredicate(predicates.joinToString(joiner, prefix = "(", postfix = ")"), false, true)
}
val ftsQuery = when (mode) {
ChatHistoryMatchMode.PHRASE -> quoteFtsTerm(normalized)
ChatHistoryMatchMode.ALL -> ftsOnly.joinToString(" AND ", transform = ::quoteFtsTerm)
ChatHistoryMatchMode.ANY -> ftsOnly.joinToString(" OR ", transform = ::quoteFtsTerm)
}
parameters += ftsQuery
val ftsPredicate = "$SEARCH_FTS_TABLE MATCH ?"
if (mode != ChatHistoryMatchMode.ALL || !hasShortTerm) {
return TextPredicate(ftsPredicate, true, false)
}
val shortPredicates = terms.filter { it.codePointCount(0, it.length) < 3 }.map {
parameters += it
"instr(lower(search.search_text), ?) > 0"
}
return TextPredicate(
(listOf(ftsPredicate) + shortPredicates).joinToString(" AND ", prefix = "(", postfix = ")"),
true,
true,
) )
} }
else -> {
parameters += normalized
TextPredicate("instr(lower(search.search_text), ?) > 0", false, true)
}
}
}
val longTerms = terms.filter { it.codePointCount(0, it.length) >= 3 }
val bigramTerms = terms.filter { it.codePointCount(0, it.length) == 2 }
val shortTerms = terms.filter { it.codePointCount(0, it.length) < 2 }
val predicates = mutableListOf<String>()
var usesFts = false
var usesBigram = false
var usesSearchText = false
if (longTerms.isNotEmpty()) {
parameters += longTerms.joinToString(
if (mode == ChatHistoryMatchMode.ANY) " OR " else " AND ",
transform = ::quoteFtsTerm,
)
predicates += ftsRowPredicate(SEARCH_FTS_TABLE)
usesFts = true
}
if (bigramTerms.isNotEmpty()) {
if (bigramIndexAvailable) {
parameters += bigramTerms.joinToString(
if (mode == ChatHistoryMatchMode.ANY) " OR " else " AND ",
transform = ::quoteFtsTerm,
)
predicates += ftsRowPredicate(SEARCH_BIGRAM_FTS_TABLE)
usesBigram = true
} else {
bigramTerms.forEach {
parameters += it
predicates += "instr(lower(search.search_text), ?) > 0"
}
usesSearchText = true
}
}
shortTerms.forEach {
parameters += it
predicates += "instr(lower(search.search_text), ?) > 0"
usesSearchText = true
}
val joiner = if (mode == ChatHistoryMatchMode.ANY) " OR " else " AND "
return TextPredicate(
predicates.joinToString(joiner, prefix = "(", postfix = ")"),
usesFts,
usesSearchText,
usesBigram,
)
}
private fun ftsRowPredicate(table: String): String =
"mr.id IN (SELECT rowid FROM $table WHERE $table MATCH ?)"
private fun quoteFtsTerm(term: String): String = private fun quoteFtsTerm(term: String): String =
"\"${term.replace("\"", "\"\"")}\"" "\"${term.replace("\"", "\"\"")}\""
@@ -948,6 +1100,17 @@ object ChatHistoryStore {
} }
} }
private fun hasBigramSearchIndex(connection: Connection): Boolean {
connection.prepareStatement(
"SELECT 1 FROM sqlite_master WHERE type = 'table' AND name = ? LIMIT 1"
).use { statement ->
statement.setString(1, SEARCH_BIGRAM_FTS_TABLE)
statement.executeQuery().use { results ->
return results.next()
}
}
}
private fun readMeta(connection: Connection, key: String): String? { private fun readMeta(connection: Connection, key: String): String? {
connection.prepareStatement( connection.prepareStatement(
"SELECT value FROM chat_history_meta WHERE key = ?" "SELECT value FROM chat_history_meta WHERE key = ?"
@@ -26,6 +26,16 @@ internal object ChatHistoryToolFormatter {
val snapshotNames = runCatching { val snapshotNames = runCatching {
ContactSnapshotStore.loadDisplayNames(event.bot.id, group?.id, userIds) ContactSnapshotStore.loadDisplayNames(event.bot.id, group?.id, userIds)
}.getOrDefault(emptyMap()) }.getOrDefault(emptyMap())
val atNames = if (group != null) {
val targetIds = records.asSequence()
.flatMap { ChatHistorySearchText.extractAtTargets(it.code).asSequence() }
.toSet()
runCatching {
ContactSnapshotStore.loadDisplayNames(event.bot.id, group.id, targetIds)
}.getOrDefault(emptyMap())
} else {
emptyMap()
}
records.forEach { record -> records.forEach { record ->
val marker = if (record.id == targetId) ">>> " else "" val marker = if (record.id == targetId) ">>> " else ""
@@ -40,7 +50,7 @@ internal object ChatHistoryToolFormatter {
val time = timeFormatter.format( val time = timeFormatter.format(
Instant.ofEpochSecond(record.time.toLong()).atZone(ZoneId.systemDefault()) Instant.ofEpochSecond(record.time.toLong()).atZone(ZoneId.systemDefault())
) )
val content = buildSnippet(ChatHistorySearchText.extract(record.code), query) val content = buildSnippet(ChatHistorySearchText.extract(record.code, atNames), query)
output.append(marker) output.append(marker)
.append("[messageId=").append(record.id).append("] ") .append("[messageId=").append(record.id).append("] ")
.append(time).append(' ') .append(time).append(' ')
@@ -18,22 +18,21 @@ import top.jie65535.mirai.data.ChatHistorySubject
class GetChatHistoryContext : BaseAgent( class GetChatHistoryContext : BaseAgent(
tool = Tool.function( tool = Tool.function(
name = "getChatHistoryContext", name = "getChatHistoryContext",
description = "根据 searchChatHistory 返回的 messageId,读取该消息在当前群聊或私聊中的前后文" + description = "读取某条聊天记录前后的消息",
"结果按时间顺序排列,并用 >>> 标记目标消息。",
parameters = Parameters.buildJsonObject { parameters = Parameters.buildJsonObject {
put("type", "object") put("type", "object")
putJsonObject("properties") { putJsonObject("properties") {
putJsonObject("messageId") { putJsonObject("messageId") {
put("type", "integer") put("type", "integer")
put("description", "searchChatHistory 返回的稳定消息 ID") put("description", "搜索结果中的 messageId")
} }
putJsonObject("before") { putJsonObject("before") {
put("type", "integer") put("type", "integer")
put("description", "目标消息之前的消息数,默认8,最大15") put("description", "前文条数,默认8,最大15")
} }
putJsonObject("after") { putJsonObject("after") {
put("type", "integer") put("type", "integer")
put("description", "目标消息之后的消息数,默认8,最大15") put("description", "后文条数,默认8,最大15")
} }
} }
putJsonArray("required") { add(JsonPrimitive("messageId")) } putJsonArray("required") { add(JsonPrimitive("messageId")) }
+26 -17
View File
@@ -33,35 +33,33 @@ import java.util.Base64
class SearchChatHistory : BaseAgent( class SearchChatHistory : BaseAgent(
tool = Tool.function( tool = Tool.function(
name = "searchChatHistory", name = "searchChatHistory",
description = "在当前群聊或私聊的 SQLite 历史中搜索消息,可按内容、发送者时间筛选" + description = "搜索当前聊天的历史消息,支持文本、发送者时间和分页;普通文本与可解析的 @ 目标会一起匹配",
"默认返回最近匹配并在结果过多时提供 cursor;需要查看命中消息前后讨论时调用 getChatHistoryContext。" +
"未指定 from 时默认搜索配置的近期天数,明确指定时间时可以搜索任意历史跨度。",
parameters = Parameters.buildJsonObject { parameters = Parameters.buildJsonObject {
put("type", "object") put("type", "object")
putJsonObject("properties") { putJsonObject("properties") {
putJsonObject("query") { putJsonObject("query") {
put("type", "string") put("type", "string")
put("description", "要搜索的消息文本;省略时按其他条件浏览时间线") put("description", "消息文本;也会匹配名称对应的 @ 目标")
} }
putJsonObject("senderId") { putJsonObject("senderId") {
put("type", "integer") put("type", "integer")
put("description", "发送者 QQ 号;仅在用户明确给出或上下文可靠提供 QQ 号时使用,不要根据昵称猜测") put("description", "发送者 QQ 号;用户明确给出时使用")
} }
putJsonObject("senderName") { putJsonObject("senderName") {
put("type", "string") put("type", "string")
put("description", "用户用名称指代发送者时优先原样传入群名片、昵称或好友备注;可与 senderId 同时提供,没有 senderId 时解析发送者,有重名时返回候选 QQ 号") put("description", "发送者名称;用户按昵称或群名片称呼时使用")
} }
putJsonObject("from") { putJsonObject("from") {
put("type", "string") put("type", "string")
put("description", "起始时间,格式 yyyy-MM-dd HH:mm 或 yyyy-MM-dd") put("description", "起始时间yyyy-MM-dd HH:mm 或 yyyy-MM-dd")
} }
putJsonObject("to") { putJsonObject("to") {
put("type", "string") put("type", "string")
put("description", "结束时间,格式 yyyy-MM-dd HH:mm 或 yyyy-MM-dd;仅日期表示当天结束") put("description", "结束时间yyyy-MM-dd HH:mm 或 yyyy-MM-dd")
} }
putJsonObject("cursor") { putJsonObject("cursor") {
put("type", "string") put("type", "string")
put("description", "上一页返回的 nextCursor;继续同一搜索时原样传回") put("description", "上一页的 nextCursor")
} }
} }
}, },
@@ -119,9 +117,11 @@ class SearchChatHistory : BaseAgent(
return "聊天记录全文索引当前不可用,无法执行关键词搜索" return "聊天记录全文索引当前不可用,无法执行关键词搜索"
} }
val atTargetIds = query?.let { resolveMentionTargets(it, event) }.orEmpty()
val request = ChatHistorySearchRequest( val request = ChatHistorySearchRequest(
subject = ChatHistorySubject.from(event.subject), subject = ChatHistorySubject.from(event.subject),
query = query, query = query,
atTargetIds = atTargetIds,
matchMode = matchMode, matchMode = matchMode,
fromId = resolvedSenderId, fromId = resolvedSenderId,
start = start.toEpochSecond().toInt(), start = start.toEpochSecond().toInt(),
@@ -182,6 +182,20 @@ class SearchChatHistory : BaseAgent(
} }
private fun resolveSender(name: String, event: MessageEvent): SenderResolution { private fun resolveSender(name: String, event: MessageEvent): SenderResolution {
val sorted = findNameMatches(name, event)
val first = sorted.firstOrNull() ?: return SenderResolution.NotFound
val second = sorted.getOrNull(1)
return if (second == null || first.matchRank < second.matchRank) {
SenderResolution.Found(first.userId)
} else {
SenderResolution.Ambiguous(sorted)
}
}
private fun resolveMentionTargets(name: String, event: MessageEvent): Set<Long> =
findNameMatches(name, event).mapTo(linkedSetOf(), ContactNameMatch::userId)
private fun findNameMatches(name: String, event: MessageEvent): List<ContactNameMatch> {
val groupId = (event as? GroupMessageEvent)?.group?.id val groupId = (event as? GroupMessageEvent)?.group?.id
val matches = runCatching { val matches = runCatching {
ContactSnapshotStore.findUsersByName(event.bot.id, groupId, name, limit = 6) ContactSnapshotStore.findUsersByName(event.bot.id, groupId, name, limit = 6)
@@ -202,14 +216,8 @@ class SearchChatHistory : BaseAgent(
} }
} }
} }
val sorted = matches.sortedWith(compareBy({ it.matchRank }, { it.userId })) return matches.sortedWith(compareBy({ it.matchRank }, { it.userId }))
val first = sorted.firstOrNull() ?: return SenderResolution.NotFound .take(MAX_NAME_MATCHES)
val second = sorted.getOrNull(1)
return if (second == null || first.matchRank < second.matchRank) {
SenderResolution.Found(first.userId)
} else {
SenderResolution.Ambiguous(sorted)
}
} }
private sealed interface SenderResolution { private sealed interface SenderResolution {
@@ -221,6 +229,7 @@ class SearchChatHistory : BaseAgent(
companion object { companion object {
private const val DEFAULT_PAGE_SIZE = 20 private const val DEFAULT_PAGE_SIZE = 20
private const val DIAGNOSTIC_PAGE_SIZE = 6 private const val DIAGNOSTIC_PAGE_SIZE = 6
private const val MAX_NAME_MATCHES = 12
private const val MAX_PAGE_SIZE = 200 private const val MAX_PAGE_SIZE = 200
private const val OUTPUT_RESERVE_CHARS = 1_000 private const val OUTPUT_RESERVE_CHARS = 1_000
private const val APPROXIMATE_RECORD_CHARS = 450 private const val APPROXIMATE_RECORD_CHARS = 450
@@ -0,0 +1,17 @@
package top.jie65535.mirai.data
import org.junit.jupiter.api.Test
import kotlin.test.assertEquals
class ChatHistorySearchTextTest {
@Test
fun rendersAtTargetNamesAndBuildsBigrams() {
val code = """
[{"type":"At","target":2180487691}, {"type":"PlainText","content":" 筱玥"}]
""".trimIndent()
assertEquals(setOf(2180487691L), ChatHistorySearchText.extractAtTargets(code))
assertEquals("@筱玥 筱玥", ChatHistorySearchText.extract(code, mapOf(2180487691L to "筱玥")))
assertEquals("筱玥 筱玥", ChatHistorySearchText.bigrams("筱玥 筱玥"))
}
}
+101 -9
View File
@@ -62,15 +62,16 @@ class ChatHistoryStoreTest {
"SELECT value FROM chat_history_meta WHERE key = 'schema_version'" "SELECT value FROM chat_history_meta WHERE key = 'schema_version'"
).use { results -> ).use { results ->
assertTrue(results.next()) assertTrue(results.next())
assertEquals("3", results.getString(1)) assertEquals("4", results.getString(1))
} }
statement.executeQuery( statement.executeQuery(
"SELECT COUNT(*) FROM sqlite_master " + "SELECT COUNT(*) FROM sqlite_master " +
"WHERE type = 'table' AND name IN " + "WHERE type = 'table' AND name IN " +
"('message_record_search', 'message_record_fts', 'chat_history_sender_alias')" "('message_record_search', 'message_record_fts', 'message_record_bigram_fts', " +
"'chat_history_sender_alias')"
).use { results -> ).use { results ->
assertTrue(results.next()) assertTrue(results.next())
assertEquals(3, results.getInt(1)) assertEquals(4, results.getInt(1))
} }
} }
} }
@@ -137,6 +138,33 @@ class ChatHistoryStoreTest {
assertEquals(2L, shortQuery.totalMatches) assertEquals(2L, shortQuery.totalMatches)
assertEquals(listOf(first, third), shortQuery.records.map { it.id }) assertEquals(listOf(first, third), shortQuery.records.map { it.id })
val mention = insert(
directory,
time = 103,
fromId = 11,
text = "",
code = buildJsonArray {
add(buildJsonObject {
put("type", "At")
put("target", 2180487691L)
})
}.toString(),
)
val literal = insert(directory, 104, 11, "筱玥检查一下")
val mentionQuery = ChatHistoryStore.search(
ChatHistorySearchRequest(
subject = subject,
query = "筱玥",
atTargetIds = setOf(2180487691L),
fromId = 11,
start = 0,
end = 200,
limit = 10,
)
)
assertEquals(2L, mentionQuery.totalMatches)
assertEquals(listOf(literal, mention), mentionQuery.records.map { it.id })
val aliases = ChatHistoryStore.findSenderAliases(subject, "旧名") val aliases = ChatHistoryStore.findSenderAliases(subject, "旧名")
assertEquals(listOf(ChatHistorySenderAliasMatch(11, "旧名张三", 1)), aliases) assertEquals(listOf(ChatHistorySenderAliasMatch(11, "旧名张三", 1)), aliases)
} finally { } finally {
@@ -245,12 +273,81 @@ class ChatHistoryStoreTest {
} }
} }
@Test
fun migratesCompletedTrigramIndexAndBackfillsBigramIndex() {
val directory = Files.createTempDirectory("jchatgpt-history-bigram-migration-test-")
val database = directory.resolve("chat-history.sqlite")
try {
ChatHistoryStore.init(directory.toFile())
val messageId = insert(directory, 100, 11, "部署完成")
ChatHistoryStore.close()
DriverManager.getConnection("jdbc:sqlite:${database.absolutePathString()}").use { connection ->
connection.createStatement().use { statement ->
statement.executeUpdate("DROP TABLE message_record_bigram_fts")
statement.executeUpdate(
"DELETE FROM chat_history_meta WHERE key = 'search_bigram_version'"
)
statement.executeUpdate(
"UPDATE chat_history_meta SET value = '3' WHERE key = 'schema_version'"
)
statement.executeUpdate(
"UPDATE chat_history_meta SET value = '1' WHERE key = 'search_index_complete'"
)
statement.executeUpdate(
"UPDATE chat_history_meta SET value = '$messageId' WHERE key = 'search_backfill_cursor'"
)
}
}
ChatHistoryStore.init(directory.toFile())
assertFalse(ChatHistoryStore.isSearchIndexReady)
assertEquals(1, ChatHistoryStore.backfillSearchIndex(batchSize = 1))
assertTrue(ChatHistoryStore.isSearchIndexReady)
val search = ChatHistoryStore.search(
ChatHistorySearchRequest(
subject = ChatHistorySubject(1, MessageSourceKind.GROUP, 1000),
query = "部署",
start = 0,
end = 200,
)
)
assertEquals(listOf(messageId), search.records.map { it.id })
DriverManager.getConnection("jdbc:sqlite:${database.absolutePathString()}").use { connection ->
connection.createStatement().use { statement ->
statement.executeQuery(
"SELECT search_text FROM message_record_search WHERE message_id = $messageId"
).use { results ->
assertTrue(results.next())
assertEquals("部署完成", results.getString(1))
}
statement.executeQuery(
"SELECT rowid FROM message_record_bigram_fts WHERE rowid = $messageId"
).use { results ->
assertTrue(results.next())
}
}
}
} finally {
ChatHistoryStore.close()
directory.toFile().deleteRecursively()
}
}
private fun insert( private fun insert(
directory: java.nio.file.Path, directory: java.nio.file.Path,
time: Int, time: Int,
fromId: Long, fromId: Long,
text: String, text: String,
senderName: String = "", senderName: String = "",
code: String = buildJsonArray {
add(buildJsonObject {
put("type", "PlainText")
put("content", text)
})
}.toString(),
): Long { ): Long {
val record = ChatMessageRecord( val record = ChatMessageRecord(
botId = 1, botId = 1,
@@ -260,12 +357,7 @@ class ChatHistoryStoreTest {
internalIds = "$time", internalIds = "$time",
time = time, time = time,
kind = MessageSourceKind.GROUP, kind = MessageSourceKind.GROUP,
code = buildJsonArray { code = code,
add(buildJsonObject {
put("type", "PlainText")
put("content", text)
})
}.toString(),
) )
ChatHistoryStore.insertRecord(record, senderName) ChatHistoryStore.insertRecord(record, senderName)
DriverManager.getConnection("jdbc:sqlite:${directory.resolve("chat-history.sqlite").absolutePathString()}").use { DriverManager.getConnection("jdbc:sqlite:${directory.resolve("chat-history.sqlite").absolutePathString()}").use {