菜单
Use when a struct field must be transformed on the way to and from the database — JSON/gob columns, encrypting or compressing a field at rest — without changing its Go type. The codec:<name> tag + liteorm.org/codec.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Use when storing large or growing binary content (files, uploads, blobs) in SQLite as streamed io.ReaderAt/io.WriterAt instead of loading whole []byte — the orm.LOB field + liteorm.org/dialect/sqlite/lob.
Use when working with liteorm's declarative orm front-end — model structs and tags, AutoMigrate, the orm.Repo, associations (Load/Attach), hooks, or soft delete.
Use when liteorm code behaves unexpectedly, or proactively before writing it, to avoid the common gotchas around loading, soft delete, types, and dialect-gated operators.
Use when migrating a gorm codebase to liteorm — running models on gorm tags as-is, rewriting them to native orm tags, and adjusting for what differs.
Use when adding vector (sqlite-vec), full-text (FTS5), or hybrid RRF search to liteorm's SQLite backend.
Use when capturing, applying, inverting, or concatenating SQLite changesets (the SESSION extension) for audit logs, one-way replication, or undo — the liteorm.org/dialect/sqlite/changeset package. SQLite backend only.
| name | field-codecs |
| description | Use when a struct field must be transformed on the way to and from the database — JSON/gob columns, encrypting or compressing a field at rest — without changing its Go type. The codec:<name> tag + liteorm.org/codec. |
A field codec transparently encodes a field on write and decodes it on read, without widening the field's Go type and without wrapping every read/write. It lives in liteorm.org/codec and is attached by tag: orm:"col,codec:<name>" (or the gorm-compatible gorm:"column:col;serializer:<name>"). Codecs run in the shared scan layer, so the transform applies uniformly through BOTH the orm and query front-ends — no read path can return the raw (encoded) value by mistake.
Reach for a codec instead of a sql.Scanner/driver.Valuer when you do not want to change the field's type for a persistence concern (e.g. the field is a plain []byte you want stored encrypted).
| Name | Column | Use |
|---|---|---|
json | TEXT | a struct/map/slice field as JSON |
gob | BLOB | any gob-encodable value |
unixtime | INTEGER | a time.Time as Unix seconds |
type Document struct {
ID int64 `orm:"id,pk"`
Meta map[string]any `orm:"meta,codec:json"`
}
orm.AutoMigrate[Document](ctx, db) // `meta` provisioned as TEXT
orm.NewRepo[Document](db).Create(ctx, &Document{Meta: map[string]any{"k": "v"}})
The migrated column type follows the codec's storage (TEXT/BLOB/INTEGER), not the Go type — AutoMigrate handles it. An explicit type: override wins.
codec.Func builds a codec from two typed functions; register it once during init, before first query/migration. The stored type sets the column kind (string→TEXT, []byte→BLOB).
func init() {
codec.Register("secretbox", codec.Func(
func(plain []byte) ([]byte, error) { return seal(key, plain) },
func(stored []byte) ([]byte, error) { return open(key, stored) },
))
}
type Secret struct {
ID int64 `orm:"id,pk"`
Value []byte `orm:"value,codec:secretbox"` // []byte stays []byte; ciphertext at rest
}
Read/write Secret.Value as plaintext; the column holds ciphertext; both front-ends return plaintext.
init; an unregistered codec name fails AutoMigrate loudly, naming the field. Last registration for a name wins (override a built-in).query.Raw[struct{ Value []byte }]), which sees the stored/encoded value.query.Select, query.Raw[T] into the model, RETURNING, eager loads).