• 简介

• 查询

  • 查询流程
  • 构建查询 SQL 语句
  • 条件语句
  • 小结

• search 结构体

  • search 的定义
  • search 的方法
  • 小结
• 总结

简介

GORM 源码解读, 基于 v1.9.11 版本.

查询

上一节中, 我们已经探究过了模型是如何定义的, 以及数据表是如何创建的.
这次, 看一下查询是如何实现的.

查询涉及到很大的一块内容, 因为要支持各种类型的方法.
先看一下官方文档中提供的最简单的几个查询方法.

// 根据主键查询第一条记录
db.First(&user)
//// SELECT * FROM users ORDER BY id LIMIT 1;

// 随机获取一条记录
db.Take(&user)
//// SELECT * FROM users LIMIT 1;

// 根据主键查询最后一条记录
db.Last(&user)
//// SELECT * FROM users ORDER BY id DESC LIMIT 1;

// 查询所有的记录
db.Find(&users)
//// SELECT * FROM users;

// 查询指定的某条记录(仅当主键为整型时可用)
db.First(&user, 10)
//// SELECT * FROM users WHERE id = 10;

以 First 方法为例, 看一下它的实现:

// First find first record that match given conditions, order by primary key
func (s *DB) First(out interface{}, where ...interface{}) *DB {
    newScope := s.NewScope(out)
    newScope.Search.Limit(1)

    return newScope.Set("gorm:order_by_primary_key", "ASC").
        inlineCondition(where...).callCallbacks(s.parent.callbacks.queries).db
}

First 方法从数据库中获取第一条数据, 以 primary key 升序排序.

前面介绍过, 具体的数据库操作实现是依靠 callbacks 的. 这里用到了 callbacks.queries.

在默认的 callbacks 中, 注册了三个不同的 query 回调函数.

// Define callbacks for querying
func init() {
    DefaultCallback.Query().Register("gorm:query", queryCallback)
    DefaultCallback.Query().Register("gorm:preload", preloadCallback)
    DefaultCallback.Query().Register("gorm:after_query", afterQueryCallback)
}

查询流程

先来看一下最主要的 queryCallback 函数.

// queryCallback used to query data from database
func queryCallback(scope *Scope) {
    if _, skip := scope.InstanceGet("gorm:skip_query_callback"); skip {
        return
    }

    //we are only preloading relations, dont touch base model
    if _, skip := scope.InstanceGet("gorm:only_preload"); skip {
        return
    }

    defer scope.trace(scope.db.nowFunc())

    var (
        isSlice, isPtr bool
        resultType     reflect.Type
        results        = scope.IndirectValue()
    )

    if orderBy, ok := scope.Get("gorm:order_by_primary_key"); ok {
        if primaryField := scope.PrimaryField(); primaryField != nil {
            scope.Search.Order(fmt.Sprintf("%v.%v %v", scope.QuotedTableName(), scope.Quote(primaryField.DBName), orderBy))
        }
    }

    if value, ok := scope.Get("gorm:query_destination"); ok {
        results = indirect(reflect.ValueOf(value))
    }

    if kind := results.Kind(); kind == reflect.Slice {
        isSlice = true
        resultType = results.Type().Elem()
        results.Set(reflect.MakeSlice(results.Type(), 0, 0))

        if resultType.Kind() == reflect.Ptr {
            isPtr = true
            resultType = resultType.Elem()
        }
    } else if kind != reflect.Struct {
        scope.Err(errors.New("unsupported destination, should be slice or struct"))
        return
    }

    scope.prepareQuerySQL()

    if !scope.HasError() {
        scope.db.RowsAffected = 0
        if str, ok := scope.Get("gorm:query_option"); ok {
            scope.SQL += addExtraSpaceIfExist(fmt.Sprint(str))
        }

        if rows, err := scope.SQLDB().Query(scope.SQL, scope.SQLVars...); scope.Err(err) == nil {
            defer rows.Close()

            columns, _ := rows.Columns()
            for rows.Next() {
                scope.db.RowsAffected++

                elem := results
                if isSlice {
                    elem = reflect.New(resultType).Elem()
                }

                scope.scan(rows, columns, scope.New(elem.Addr().Interface()).Fields())

                if isSlice {
                    if isPtr {
                        results.Set(reflect.Append(results, elem.Addr()))
                    } else {
                        results.Set(reflect.Append(results, elem))
                    }
                }
            }

            if err := rows.Err(); err != nil {
                scope.Err(err)
            } else if scope.db.RowsAffected == 0 && !isSlice {
                scope.Err(ErrRecordNotFound)
            }
        }
    }
}

核心的步骤在于 scope.prepareQuerySQL() 构建 SQL 语句.
然后通过 rows, err := scope.SQLDB().Query(scope.SQL, scope.SQLVars...), 执行了数据库查询.

那么查询到的结果是如何传递的, 传递给谁呢?

函数的开头定义了 results = scope.IndirectValue(), 这就是最终查询结果的归属地.

results 只能是结构体或者是结构体的切片.

if kind := results.Kind(); kind == reflect.Slice {
  isSlice = true
  resultType = results.Type().Elem()
  results.Set(reflect.MakeSlice(results.Type(), 0, 0))

  if resultType.Kind() == reflect.Ptr {
    isPtr = true
    resultType = resultType.Elem()
  }
} else if kind != reflect.Struct {
  scope.Err(errors.New("unsupported destination, should be slice or struct"))
  return
}

具体如何处理查询到的结果是在下面这部分代码中:

columns, _ := rows.Columns()
for rows.Next() {
  scope.db.RowsAffected++

  elem := results
  if isSlice {
    elem = reflect.New(resultType).Elem()
  }

  scope.scan(rows, columns, scope.New(elem.Addr().Interface()).Fields())

  if isSlice {
    if isPtr {
      results.Set(reflect.Append(results, elem.Addr()))
    } else {
      results.Set(reflect.Append(results, elem))
    }
  }
}

这部分代码的核心语句在于 scope.scan, 看一下这个方法的定义:

func (scope *Scope) scan(rows *sql.Rows, columns []string, fields []*Field) {
    var (
        ignored            interface{}
        values             = make([]interface{}, len(columns))
        selectFields       []*Field
        selectedColumnsMap = map[string]int{}
        resetFields        = map[int]*Field{}
    )

    for index, column := range columns {
        values[index] = &ignored

        selectFields = fields
        offset := 0
        if idx, ok := selectedColumnsMap[column]; ok {
            offset = idx + 1
            selectFields = selectFields[offset:]
        }

        for fieldIndex, field := range selectFields {
            if field.DBName == column {
                if field.Field.Kind() == reflect.Ptr {
                    values[index] = field.Field.Addr().Interface()
                } else {
                    reflectValue := reflect.New(reflect.PtrTo(field.Struct.Type))
                    reflectValue.Elem().Set(field.Field.Addr())
                    values[index] = reflectValue.Interface()
                    resetFields[index] = field
                }

                selectedColumnsMap[column] = offset + fieldIndex

                if field.IsNormal {
                    break
                }
            }
        }
    }

    scope.Err(rows.Scan(values...))

    for index, field := range resetFields {
        if v := reflect.ValueOf(values[index]).Elem().Elem(); v.IsValid() {
            field.Field.Set(v)
        }
    }
}

就和它的名字暗示的那样, 实际上就是调用了 rows.Scan(values...), 将查询到的数据复制到对应的字段中.

由此, 我们就了解了查询时的主要流程了.

前面专注于流程, 略过了构建 SQL 语句的细节, 来仔细看看 prepareQuerySQL 方法.

构建查询 SQL 语句

func (scope *Scope) prepareQuerySQL() {
    if scope.Search.raw {
        scope.Raw(scope.CombinedConditionSql())
    } else {
        scope.Raw(fmt.Sprintf("SELECT %v FROM %v %v", scope.selectSQL(), scope.QuotedTableName(), scope.CombinedConditionSql()))
    }
    return
}

内部分支中都使用到了 scope.Raw, 看一下它的实现:

// Raw set raw sql
func (scope *Scope) Raw(sql string) *Scope {
    scope.SQL = strings.Replace(sql, "$$$", "?", -1)
    return scope
}

它的作用是将获取到的 sql 语句赋值到 scope.SQL 字段上, 其中替换了所有的 $$$ 为 ?.

回到 prepareQuerySQL 上来, 重要的部分是其实是 Raw 的参数.
if 的后半部分更好理解点, 就是构建了 SELECT 表达式.

SELECT 表达式需要三个变量, 字段名, 表名, 条件.

将每个都看一下吧.

func (scope *Scope) selectSQL() string {
    if len(scope.Search.selects) == 0 {
        if len(scope.Search.joinConditions) > 0 {
            return fmt.Sprintf("%v.*", scope.QuotedTableName())
        }
        return "*"
    }
    return scope.buildSelectQuery(scope.Search.selects)
}

func (scope *Scope) buildSelectQuery(clause map[string]interface{}) (str string) {
    switch value := clause["query"].(type) {
    case string:
        str = value
    case []string:
        str = strings.Join(value, ", ")
    }

    args := clause["args"].([]interface{})
    replacements := []string{}
    for _, arg := range args {
        switch reflect.ValueOf(arg).Kind() {
        case reflect.Slice:
            values := reflect.ValueOf(arg)
            var tempMarks []string
            for i := 0; i < values.Len(); i++ {
                tempMarks = append(tempMarks, scope.AddToVars(values.Index(i).Interface()))
            }
            replacements = append(replacements, strings.Join(tempMarks, ","))
        default:
            if valuer, ok := interface{}(arg).(driver.Valuer); ok {
                arg, _ = valuer.Value()
            }
            replacements = append(replacements, scope.AddToVars(arg))
        }
    }

    buff := bytes.NewBuffer([]byte{})
    i := 0
    for pos, char := range str {
        if str[pos] == '?' {
            buff.WriteString(replacements[i])
            i++
        } else {
            buff.WriteRune(char)
        }
    }

    str = buff.String()

    return
}

当 scope.Search.selects 为空的时候, 比较简单.
只要根据是否有连表查询, 返回 table.* 或 *.

buildSelectQuery 就是根据 scope.Search.selects 构建查询字段名.

前面半部分一看就明白.

switch value := clause["query"].(type) {
case string:
  str = value
case []string:
  str = strings.Join(value, ", ")
}

重点是遇到参数时如何处理, 也就是后半段代码.

args := clause["args"].([]interface{})
replacements := []string{}
for _, arg := range args {
  switch reflect.ValueOf(arg).Kind() {
  case reflect.Slice:
    values := reflect.ValueOf(arg)
    var tempMarks []string
    for i := 0; i < values.Len(); i++ {
      tempMarks = append(tempMarks, scope.AddToVars(values.Index(i).Interface()))
    }
    replacements = append(replacements, strings.Join(tempMarks, ","))
  default:
    if valuer, ok := interface{}(arg).(driver.Valuer); ok {
      arg, _ = valuer.Value()
    }
    replacements = append(replacements, scope.AddToVars(arg))
  }
}

buff := bytes.NewBuffer([]byte{})
i := 0
for pos, char := range str {
  if str[pos] == '?' {
    buff.WriteString(replacements[i])
    i++
  } else {
    buff.WriteRune(char)
  }
}

主要的过程是遍历 args := clause["args"].([]interface{}),
创建了一个 replacements 切片. 然后将 str 中所有的 ?,
替换为了对应的字段.

到此, 构建 SELECT 字段的过程就结束了.

获取表名的过程相对简单, 直接展示代码吧:

// QuotedTableName return quoted table name
func (scope *Scope) QuotedTableName() (name string) {
    if scope.search != nil && len(scope.Search.tableName) > 0 {
        if strings.Contains(scope.Search.tableName, " ") {
            return scope.Search.tableName
        }
        return scope.Quote(scope.Search.tableName)
    }

    return scope.Quote(scope.TableName())
}

条件语句

更多的关注点在于如何构建筛选条件, 即 CombinedConditionSql 方法.

// CombinedConditionSql return combined condition sql
func (scope *Scope) CombinedConditionSql() string {
    joinSQL := scope.joinsSQL()
    whereSQL := scope.whereSQL()
    if scope.Search.raw {
        whereSQL = strings.TrimSuffix(strings.TrimPrefix(whereSQL, "WHERE ("), ")")
    }
    return joinSQL + whereSQL + scope.groupSQL() +
        scope.havingSQL() + scope.orderSQL() + scope.limitAndOffsetSQL()
}

短小的代码中是精简的逻辑, 条件语句有很多模块, 这里总共有 6 个子句.
都看一遍吧, 看完之后应该对如何构建条件语句不会陌生了.

func (scope *Scope) joinsSQL() string {
    var joinConditions []string
    for _, clause := range scope.Search.joinConditions {
        if sql := scope.buildCondition(clause, true); sql != "" {
            joinConditions = append(joinConditions, strings.TrimSuffix(strings.TrimPrefix(sql, "("), ")"))
        }
    }

    return strings.Join(joinConditions, " ") + " "
}

创建 joinSQL 的过程中主要用到了 buildCondition, 继续深入:

func (scope *Scope) buildCondition(clause map[string]interface{}, include bool) (str string) {
    var (
        quotedTableName  = scope.QuotedTableName()
        quotedPrimaryKey = scope.Quote(scope.PrimaryKey())
        equalSQL         = "="
        inSQL            = "IN"
    )

    // If building not conditions
    if !include {
        equalSQL = "<>"
        inSQL = "NOT IN"
    }

    switch value := clause["query"].(type) {
    case sql.NullInt64:
        return fmt.Sprintf("(%v.%v %s %v)", quotedTableName, quotedPrimaryKey, equalSQL, value.Int64)
    case int, int8, int16, int32, int64, uint, uint8, uint16, uint32, uint64:
        return fmt.Sprintf("(%v.%v %s %v)", quotedTableName, quotedPrimaryKey, equalSQL, value)
    case []int, []int8, []int16, []int32, []int64, []uint, []uint8, []uint16, []uint32, []uint64, []string, []interface{}:
        if !include && reflect.ValueOf(value).Len() == 0 {
            return
        }
        str = fmt.Sprintf("(%v.%v %s (?))", quotedTableName, quotedPrimaryKey, inSQL)
        clause["args"] = []interface{}{value}
    case string:
        if isNumberRegexp.MatchString(value) {
            return fmt.Sprintf("(%v.%v %s %v)", quotedTableName, quotedPrimaryKey, equalSQL, scope.AddToVars(value))
        }

        if value != "" {
            if !include {
                if comparisonRegexp.MatchString(value) {
                    str = fmt.Sprintf("NOT (%v)", value)
                } else {
                    str = fmt.Sprintf("(%v.%v NOT IN (?))", quotedTableName, scope.Quote(value))
                }
            } else {
                str = fmt.Sprintf("(%v)", value)
            }
        }
    case map[string]interface{}:
        var sqls []string
        for key, value := range value {
            if value != nil {
                sqls = append(sqls, fmt.Sprintf("(%v.%v %s %v)", quotedTableName, scope.Quote(key), equalSQL, scope.AddToVars(value)))
            } else {
                if !include {
                    sqls = append(sqls, fmt.Sprintf("(%v.%v IS NOT NULL)", quotedTableName, scope.Quote(key)))
                } else {
                    sqls = append(sqls, fmt.Sprintf("(%v.%v IS NULL)", quotedTableName, scope.Quote(key)))
                }
            }
        }
        return strings.Join(sqls, " AND ")
    case interface{}:
        var sqls []string
        newScope := scope.New(value)

        if len(newScope.Fields()) == 0 {
            scope.Err(fmt.Errorf("invalid query condition: %v", value))
            return
        }
        scopeQuotedTableName := newScope.QuotedTableName()
        for _, field := range newScope.Fields() {
            if !field.IsIgnored && !field.IsBlank {
                sqls = append(sqls, fmt.Sprintf("(%v.%v %s %v)", scopeQuotedTableName, scope.Quote(field.DBName), equalSQL, scope.AddToVars(field.Field.Interface())))
            }
        }
        return strings.Join(sqls, " AND ")
    default:
        scope.Err(fmt.Errorf("invalid query condition: %v", value))
        return
    }

    replacements := []string{}
    args := clause["args"].([]interface{})
    for _, arg := range args {
        var err error
        switch reflect.ValueOf(arg).Kind() {
        case reflect.Slice: // For where("id in (?)", []int64{1,2})
            if scanner, ok := interface{}(arg).(driver.Valuer); ok {
                arg, err = scanner.Value()
                replacements = append(replacements, scope.AddToVars(arg))
            } else if b, ok := arg.([]byte); ok {
                replacements = append(replacements, scope.AddToVars(b))
            } else if as, ok := arg.([][]interface{}); ok {
                var tempMarks []string
                for _, a := range as {
                    var arrayMarks []string
                    for _, v := range a {
                        arrayMarks = append(arrayMarks, scope.AddToVars(v))
                    }

                    if len(arrayMarks) > 0 {
                        tempMarks = append(tempMarks, fmt.Sprintf("(%v)", strings.Join(arrayMarks, ",")))
                    }
                }

                if len(tempMarks) > 0 {
                    replacements = append(replacements, strings.Join(tempMarks, ","))
                }
            } else if values := reflect.ValueOf(arg); values.Len() > 0 {
                var tempMarks []string
                for i := 0; i < values.Len(); i++ {
                    tempMarks = append(tempMarks, scope.AddToVars(values.Index(i).Interface()))
                }
                replacements = append(replacements, strings.Join(tempMarks, ","))
            } else {
                replacements = append(replacements, scope.AddToVars(Expr("NULL")))
            }
        default:
            if valuer, ok := interface{}(arg).(driver.Valuer); ok {
                arg, err = valuer.Value()
            }

            replacements = append(replacements, scope.AddToVars(arg))
        }

        if err != nil {
            scope.Err(err)
        }
    }

    buff := bytes.NewBuffer([]byte{})
    i := 0
    for _, s := range str {
        if s == '?' && len(replacements) > i {
            buff.WriteString(replacements[i])
            i++
        } else {
            buff.WriteRune(s)
        }
    }

    str = buff.String()

    return
}

开头是一个精妙的选择, 基于 include, 实现了 not 条件.

var (
  quotedTableName  = scope.QuotedTableName()
  quotedPrimaryKey = scope.Quote(scope.PrimaryKey())
  equalSQL         = "="
  inSQL            = "IN"
)

// If building not conditions
if !include {
  equalSQL = "<>"
  inSQL = "NOT IN"
}

中间是一个 switch value := clause["query"].(type) 选择.
在这个 switch 选择中, 大部分的条件都会直接返回.
剩余的部分, 则会构建 str 字符串变量.

而这会继续进入到结尾部分, 这部分的代码和我们上面看过的非常类似,
就是根据 clause["args"] 构建 replacements 切片,
用来替换 str 变量中的 ?.

接着看下一个 whereSQL 方法.

func (scope *Scope) whereSQL() (sql string) {
    var (
        quotedTableName                                = scope.QuotedTableName()
        deletedAtField, hasDeletedAtField              = scope.FieldByName("DeletedAt")
        primaryConditions, andConditions, orConditions []string
    )

    if !scope.Search.Unscoped && hasDeletedAtField {
        sql := fmt.Sprintf("%v.%v IS NULL", quotedTableName, scope.Quote(deletedAtField.DBName))
        primaryConditions = append(primaryConditions, sql)
    }

    if !scope.PrimaryKeyZero() {
        for _, field := range scope.PrimaryFields() {
            sql := fmt.Sprintf("%v.%v = %v", quotedTableName, scope.Quote(field.DBName), scope.AddToVars(field.Field.Interface()))
            primaryConditions = append(primaryConditions, sql)
        }
    }

    for _, clause := range scope.Search.whereConditions {
        if sql := scope.buildCondition(clause, true); sql != "" {
            andConditions = append(andConditions, sql)
        }
    }

    for _, clause := range scope.Search.orConditions {
        if sql := scope.buildCondition(clause, true); sql != "" {
            orConditions = append(orConditions, sql)
        }
    }

    for _, clause := range scope.Search.notConditions {
        if sql := scope.buildCondition(clause, false); sql != "" {
            andConditions = append(andConditions, sql)
        }
    }

    orSQL := strings.Join(orConditions, " OR ")
    combinedSQL := strings.Join(andConditions, " AND ")
    if len(combinedSQL) > 0 {
        if len(orSQL) > 0 {
            combinedSQL = combinedSQL + " OR " + orSQL
        }
    } else {
        combinedSQL = orSQL
    }

    if len(primaryConditions) > 0 {
        sql = "WHERE " + strings.Join(primaryConditions, " AND ")
        if len(combinedSQL) > 0 {
            sql = sql + " AND (" + combinedSQL + ")"
        }
    } else if len(combinedSQL) > 0 {
        sql = "WHERE " + combinedSQL
    }
    return
}

主要构建了三个部分, primaryConditions, andConditions, orConditions.

if !scope.Search.Unscoped && hasDeletedAtField {
  sql := fmt.Sprintf("%v.%v IS NULL", quotedTableName, scope.Quote(deletedAtField.DBName))
  primaryConditions = append(primaryConditions, sql)
}

if !scope.PrimaryKeyZero() {
  for _, field := range scope.PrimaryFields() {
    sql := fmt.Sprintf("%v.%v = %v", quotedTableName, scope.Quote(field.DBName), scope.AddToVars(field.Field.Interface()))
    primaryConditions = append(primaryConditions, sql)
  }
}

前面两个 if 构建了 primaryConditions 条件.

for _, clause := range scope.Search.whereConditions {
  if sql := scope.buildCondition(clause, true); sql != "" {
    andConditions = append(andConditions, sql)
  }
}

for _, clause := range scope.Search.orConditions {
  if sql := scope.buildCondition(clause, true); sql != "" {
    orConditions = append(orConditions, sql)
  }
}

for _, clause := range scope.Search.notConditions {
  if sql := scope.buildCondition(clause, false); sql != "" {
    andConditions = append(andConditions, sql)
  }
}

然后三个 for 循环都使用了 buildCondition 方法.
注意到 scope.Search.notConditions 是算在 andConditions 中的.

orSQL := strings.Join(orConditions, " OR ")
combinedSQL := strings.Join(andConditions, " AND ")
if len(combinedSQL) > 0 {
  if len(orSQL) > 0 {
    combinedSQL = combinedSQL + " OR " + orSQL
  }
} else {
  combinedSQL = orSQL
}

结合 orConditions 和 andConditions 生成了条件语句.

if len(primaryConditions) > 0 {
  sql = "WHERE " + strings.Join(primaryConditions, " AND ")
  if len(combinedSQL) > 0 {
    sql = sql + " AND (" + combinedSQL + ")"
  }
} else if len(combinedSQL) > 0 {
  sql = "WHERE " + combinedSQL
}
return

最后, 结合 primaryConditions 生成最终的 WHERE 子句.

接着看另一个:

func (scope *Scope) groupSQL() string {
    if len(scope.Search.group) == 0 {
        return ""
    }
    return " GROUP BY " + scope.Search.group
}

GROUP BY 子句比较简单, 直接就能构建.

继续:

func (scope *Scope) havingSQL() string {
    if len(scope.Search.havingConditions) == 0 {
        return ""
    }

    var andConditions []string
    for _, clause := range scope.Search.havingConditions {
        if sql := scope.buildCondition(clause, true); sql != "" {
            andConditions = append(andConditions, sql)
        }
    }

    combinedSQL := strings.Join(andConditions, " AND ")
    if len(combinedSQL) == 0 {
        return ""
    }

    return " HAVING " + combinedSQL
}

HAVING 子句也不算难, 构建完条件之后用 AND 连接, 然后在最前面加上 HAVING 就行了.

继续:

func (scope *Scope) orderSQL() string {
    if len(scope.Search.orders) == 0 || scope.Search.ignoreOrderQuery {
        return ""
    }

    var orders []string
    for _, order := range scope.Search.orders {
        if str, ok := order.(string); ok {
            orders = append(orders, scope.quoteIfPossible(str))
        } else if expr, ok := order.(*expr); ok {
            exp := expr.expr
            for _, arg := range expr.args {
                exp = strings.Replace(exp, "?", scope.AddToVars(arg), 1)
            }
            orders = append(orders, exp)
        }
    }
    return " ORDER BY " + strings.Join(orders, ",")
}

结构也是类似, 遍历 scope.Search.orders 切片, order 有两种不同的类型, 字符串或者 expr 结构体.
后者用于处理带参数的情况.

最后还有一个 limitAndOffsetSQL 方法:

func (scope *Scope) limitAndOffsetSQL() string {
    return scope.Dialect().LimitAndOffsetSQL(scope.Search.limit, scope.Search.offset)
}

这直接调用了具体数据库驱动中的 LimitAndOffsetSQL 方法.

看两个具体的实现, 一个是通用中的实现, 另一个是 mysql 中的实现.

func (commonDialect) LimitAndOffsetSQL(limit, offset interface{}) (sql string) {
    if limit != nil {
        if parsedLimit, err := strconv.ParseInt(fmt.Sprint(limit), 0, 0); err == nil && parsedLimit >= 0 {
            sql += fmt.Sprintf(" LIMIT %d", parsedLimit)
        }
    }
    if offset != nil {
        if parsedOffset, err := strconv.ParseInt(fmt.Sprint(offset), 0, 0); err == nil && parsedOffset >= 0 {
            sql += fmt.Sprintf(" OFFSET %d", parsedOffset)
        }
    }
    return
}

直接将 limit 和 offset 解析为 int 类型, 然后连接对应的关键字即可.

接着看一下 mysql 中的实现:

func (s mysql) LimitAndOffsetSQL(limit, offset interface{}) (sql string) {
    if limit != nil {
        if parsedLimit, err := strconv.ParseInt(fmt.Sprint(limit), 0, 0); err == nil && parsedLimit >= 0 {
            sql += fmt.Sprintf(" LIMIT %d", parsedLimit)

            if offset != nil {
                if parsedOffset, err := strconv.ParseInt(fmt.Sprint(offset), 0, 0); err == nil && parsedOffset >= 0 {
                    sql += fmt.Sprintf(" OFFSET %d", parsedOffset)
                }
            }
        }
    }
    return
}

两者的区别在于 offset 的嵌套, mysql 中 offset 必须和 limit 一起使用.

就这样, CombinedConditionSql 中的所有子句都看完了.
说到底其实也没什么魔法, 不过是根据不同的条件, 构建不同的 SQL 语句.

小结

一路从 First 深入到查询的内部细节. 在了解了底层细节之后, 其他类似的方法也就不难理解了.

// Take return a record that match given conditions, the order will depend on the database implementation
func (s *DB) Take(out interface{}, where ...interface{}) *DB {
    newScope := s.NewScope(out)
    newScope.Search.Limit(1)
    return newScope.inlineCondition(where...).callCallbacks(s.parent.callbacks.queries).db
}

// Last find last record that match given conditions, order by primary key
func (s *DB) Last(out interface{}, where ...interface{}) *DB {
    newScope := s.NewScope(out)
    newScope.Search.Limit(1)
    return newScope.Set("gorm:order_by_primary_key", "DESC").
        inlineCondition(where...).callCallbacks(s.parent.callbacks.queries).db
}

// Find find records that match given conditions
func (s *DB) Find(out interface{}, where ...interface{}) *DB {
    return s.NewScope(out).inlineCondition(where...).callCallbacks(s.parent.callbacks.queries).db
}

search 结构体

前面的过程中, 我们只看到了最简单的查询是如何产生的.
在这个过程中, 没有仔细研究查询条件是如何存储的.

看一下如何使用 Where 方法添加查询条件.

// Get first matched record
db.Where("name = ?", "jinzhu").First(&user)
//// SELECT * FROM users WHERE name = 'jinzhu' limit 1;

// Get all matched records
db.Where("name = ?", "jinzhu").Find(&users)
//// SELECT * FROM users WHERE name = 'jinzhu';

上面的例子来自于官方文档. GORM 使用链式调用的风格, 可以串联多个 Where 方法, 或是其他的查询条件.

// Where return a new relation, filter records with given conditions, accepts `map`, `struct` or `string` as conditions, refer http://jinzhu.github.io/gorm/crud.html#query
func (s *DB) Where(query interface{}, args ...interface{}) *DB {
    return s.clone().search.Where(query, args...).db
}

上面是 Where 方法的代码, 在它的源码附近有很多类似的的方法.

// Or filter records that match before conditions or this one, similar to `Where`
func (s *DB) Or(query interface{}, args ...interface{}) *DB {
    return s.clone().search.Or(query, args...).db
}

// Not filter records that don't match current conditions, similar to `Where`
func (s *DB) Not(query interface{}, args ...interface{}) *DB {
    return s.clone().search.Not(query, args...).db
}

可以很容易的发现, 这一切的源头都是 search 对象.

结构体 DB 定义的时候, 有个字段就是 search:

search            *search

search 的定义

这就是用于存储查询条件的地方. 它的定义如下:

type search struct {
    db               *DB
    whereConditions  []map[string]interface{}
    orConditions     []map[string]interface{}
    notConditions    []map[string]interface{}
    havingConditions []map[string]interface{}
    joinConditions   []map[string]interface{}
    initAttrs        []interface{}
    assignAttrs      []interface{}
    selects          map[string]interface{}
    omits            []string
    orders           []interface{}
    preload          []searchPreload
    offset           interface{}
    limit            interface{}
    group            string
    tableName        string
    raw              bool
    Unscoped         bool
    ignoreOrderQuery bool
}

type searchPreload struct {
    schema     string
    conditions []interface{}
}

这里有很多类型为 []map[string]interface{} 的字段, 结合前面关于条件查询的代码, 就能回忆起这就是存储各种条件的地方.

另一些字段比如 offset 和 limit 也很容易明白它的作用.

search 的方法

search 下有很多方法, 虽然方法数量比较多, 但基本都很短, 总共也就一百行出头.

func (s *search) clone() *search {
    clone := *s
    return &clone
}

这个克隆方法有点独特, 似乎什么也没做, 也可能是我见识少.

func (s *search) Where(query interface{}, values ...interface{}) *search {
    s.whereConditions = append(s.whereConditions, map[string]interface{}{"query": query, "args": values})
    return s
}

func (s *search) Not(query interface{}, values ...interface{}) *search {
    s.notConditions = append(s.notConditions, map[string]interface{}{"query": query, "args": values})
    return s
}

func (s *search) Or(query interface{}, values ...interface{}) *search {
    s.orConditions = append(s.orConditions, map[string]interface{}{"query": query, "args": values})
    return s
}

上面这些方法都是用参数构建成一个 map 然后推入对应的切片中, 考虑到链式调用, 返回了本身.

func (s *search) Attrs(attrs ...interface{}) *search {
    s.initAttrs = append(s.initAttrs, toSearchableMap(attrs...))
    return s
}

func (s *search) Assign(attrs ...interface{}) *search {
    s.assignAttrs = append(s.assignAttrs, toSearchableMap(attrs...))
    return s
}

func toSearchableMap(attrs ...interface{}) (result interface{}) {
    if len(attrs) > 1 {
        if str, ok := attrs[0].(string); ok {
            result = map[string]interface{}{str: attrs[1]}
        }
    } else if len(attrs) == 1 {
        if attr, ok := attrs[0].(map[string]interface{}); ok {
            result = attr
        }

        if attr, ok := attrs[0].(interface{}); ok {
            result = attr
        }
    }
    return
}

这两个方法也是类似, 并使用了 toSearchableMap 转换参数.

func (s *search) Order(value interface{}, reorder ...bool) *search {
    if len(reorder) > 0 && reorder[0] {
        s.orders = []interface{}{}
    }

    if value != nil && value != "" {
        s.orders = append(s.orders, value)
    }
    return s
}

看到这个可能有点疑惑, 可以从文档和注释中获取解释.

// Order specify order when retrieve records from database, set reorder to `true` to overwrite defined conditions
//     db.Order("name DESC")
//     db.Order("name DESC", true) // reorder
//     db.Order(gorm.Expr("name = ? DESC", "first")) // sql expression
func (s *DB) Order(value interface{}, reorder ...bool) *DB {
    return s.clone().search.Order(value, reorder...).db
}

第二个参数用于判断是否覆盖前面的排序条件.

可能有点奇怪的是为什么 reorder 是可变参数, 不知为了兼容或者是历史遗留.

另一点是不能理解 []interface{}{}, 这其实可以分为两部分, []interface{} 是类型, {} 构造了一个空的该类型实例.

func (s *search) Select(query interface{}, args ...interface{}) *search {
    s.selects = map[string]interface{}{"query": query, "args": args}
    return s
}

func (s *search) Omit(columns ...string) *search {
    s.omits = columns
    return s
}

func (s *search) Limit(limit interface{}) *search {
    s.limit = limit
    return s
}

func (s *search) Offset(offset interface{}) *search {
    s.offset = offset
    return s
}

这几个就是替换型的了, 每次调用都只会保存最新值.

func (s *search) Group(query string) *search {
    s.group = s.getInterfaceAsSQL(query)
    return s
}

func (s *search) getInterfaceAsSQL(value interface{}) (str string) {
    switch value.(type) {
    case string, int, int8, int16, int32, int64, uint, uint8, uint16, uint32, uint64:
        str = fmt.Sprintf("%v", value)
    default:
        s.db.AddError(ErrInvalidSQL)
    }

    if str == "-1" {
        return ""
    }
    return
}

getInterfaceAsSQL 的一个特性是使用 -1 会重置.

func (s *search) Having(query interface{}, values ...interface{}) *search {
    if val, ok := query.(*expr); ok {
        s.havingConditions = append(s.havingConditions, map[string]interface{}{"query": val.expr, "args": val.args})
    } else {
        s.havingConditions = append(s.havingConditions, map[string]interface{}{"query": query, "args": values})
    }
    return s
}

func (s *search) Joins(query string, values ...interface{}) *search {
    s.joinConditions = append(s.joinConditions, map[string]interface{}{"query": query, "args": values})
    return s
}

这其实也比较类似前面看过的, 就不多解释了.

func (s *search) Preload(schema string, values ...interface{}) *search {
    var preloads []searchPreload
    for _, preload := range s.preload {
        if preload.schema != schema {
            preloads = append(preloads, preload)
        }
    }
    preloads = append(preloads, searchPreload{schema, values})
    s.preload = preloads
    return s
}

Preload 需要防止重复, 所以开头会重新遍历一遍已经存在的 schema.

func (s *search) Raw(b bool) *search {
    s.raw = b
    return s
}

func (s *search) unscoped() *search {
    s.Unscoped = true
    return s
}

func (s *search) Table(name string) *search {
    s.tableName = name
    return s
}

最后几个方法也没什么特殊的.

小结

search 结构体还是挺简单的, 定义加方法总共也就一百多行.
但用处却不小, 查询相关的条件都是存储在这里的.

总结

这部分主要查看了 SQL 查询是如何发生的, 并在这个过程中探索了各种查询子句是如何实现的. 同时, 也研究了一下 search 结构体和它的作用.