13 KiB
nim-serde
Easy-to-use json serialization capabilities, and a drop-in replacement for std/json
.
Quick examples
Opt-in serialization by default:
import pkg/serde/json
type MyType = object
field1 {.serialize.}: bool
field2: bool
assert MyType(field1: true, field2: true).toJson == """{"field1":true}"""
Opt-out deserialization by default:
import pkg/serde/json
# All fields deserialized, as none are ignored
type MyType1 = object
field1: bool
field2: bool
let jsn1 = """{
"field1": true,
"field2": true
}"""
assert !MyType1.fromJson(jsn1) == MyType1(field1: true, field2: true)
# Don't deserialize ignored fields in OptOut mode
type MyType2 = object
field1 {.deserialize(ignore=true).}: bool
field2: bool
let jsn2 = """{
"field1": true,
"field2": true,
"extra": "extra fields don't error in OptOut mode"
}"""
assert !MyType2.fromJson(jsn2) == MyType2(field1: false, field2: true)
# Note, the ! operator is part of https://github.com/codex-storage/questionable, which retrieves a value if set
Serialize all fields of a type (OptOut mode):
import pkg/serde/json
type MyType {.serialize.} = object
field1: int
field2: int
assert MyType(field1: 1, field2: 2).toJson == """{"field1":1,"field2":2}"""
Alias field names in both directions!
import pkg/serde/json
type MyType {.serialize.} = object
field1 {.serialize("othername"),deserialize("takesprecedence").}: int
field2: int
assert MyType(field1: 1, field2: 2).toJson == """{"othername":1,"field2":2}"""
let jsn = """{
"othername": 1,
"field2": 2,
"takesprecedence": 3
}"""
assert !MyType.fromJson(jsn) == MyType(field1: 3, field2: 2)
Supports strict mode, where type fields and json fields must match
import pkg/serde/json
type MyType {.deserialize(mode=Strict).} = object
field1: int
field2: int
let jsn = """{
"field1": 1,
"field2": 2,
"extra": 3
}"""
let res = MyType.fromJson(jsn)
assert res.isFailure
assert res.error of SerdeError
assert res.error.msg == "json field(s) missing in object: {\"extra\"}"
Serde modes
nim-serde
uses three different modes to control de/serialization:
OptIn
OptOut
Strict
Modes can be set in the {.serialize.}
and/or {.deserialize.}
pragmas on type
definitions. Each mode has a different meaning depending on if the type is being
serialized or deserialized. Modes can be set by setting mode
in the serialize
or
deserialize
pragma annotation, eg:
type MyType {.serialize(mode=Strict).} = object
field1: bool
field2: bool
Modes reference
serialize | deserialize | |
---|---|---|
SerdeMode.OptOut |
All object fields will be serialized, except fields marked with {.serialize(ignore=true).} . |
All json keys will be deserialized, except fields marked with {.deserialize(ignore=true).} . No error if extra json fields exist. |
SerdeMode.OptIn |
Only fields marked with {.serialize.} will be serialized. Fields marked with {.serialize(ignore=true).} will not be serialized. |
Only fields marked with {.deserialize.} will be deserialized. Fields marked with {.deserialize(ignore=true).} will not be deserialized. A SerdeError error is raised if the field is missing in json. |
SerdeMode.Strict |
All object fields will be serialized, regardless if the field is marked with {.serialize(ignore=true).} . |
Object fields and json fields must match exactly, otherwise a SerdeError is raised. |
Default modes
nim-serde
will de/serialize types if they are not annotated with serialize
or
deserialize
, but will assume a default mode. By default, with no pragmas specified,
serde
will always serialize in OptIn
mode, meaning any fields to b Additionally, if
the types are annotated, but a mode is not specified, serde
will assume a (possibly
different) default mode.
# Type is not annotated
# A default mode of OptIn (for serialize) and OptOut (for deserialize) is assumed.
type MyObj = object
field1: bool
field2: bool
# Type is annotated, but mode not specified
# A default mode of OptOut is assumed for both serialize and deserialize.
type MyObj {.serialize, deserialize.} = object
field1: bool
field2: bool
Default mode reference
serialize | deserialize | |
---|---|---|
Default (no pragma) | OptIn |
OptOut |
Default (pragma, but no mode) | OptOut |
OptOut |
Serde field options
Type fields can be annotated with {.serialize.}
and {.deserialize.}
and properties
can be set on these pragmas, determining de/serialization behavior.
For example,
import pkg/serde/json
type
Person {.serialize(mode=OptOut), deserialize(mode=OptIn).} = object
id {.serialize(ignore=true), deserialize(key="personid").}: int
name: string
birthYear: int
address: string
phone: string
let person = Person(
name: "Lloyd Christmas",
birthYear: 1970,
address: "123 Sesame Street, Providence, Rhode Island 12345",
phone: "555-905-justgivemethedamnnumber!⛽️🔥")
let createRequest = """{
"name": "Lloyd Christmas",
"birthYear": 1970,
"address": "123 Sesame Street, Providence, Rhode Island 12345",
"phone": "555-905-justgivemethedamnnumber!⛽️🔥"
}"""
assert person.toJson(pretty=true) == createRequest
let createResponse = """{
"personid": 1,
"name": "Lloyd Christmas",
"birthYear": 1970,
"address": "123 Sesame Street, Providence, Rhode Island 12345",
"phone": "555-905-justgivemethedamnnumber!⛽️🔥"
}"""
assert !Person.fromJson(createResponse) == Person(id: 1)
key
Specifying a key
, will alias the field name. When seriazlizing, json will be written
with key
instead of the field name. When deserializing, the json must contain key
for the field to be deserialized.
ignore
Specifying ignore
, will prevent de/serialization on the field.
Serde field options reference
serialize | deserialize | |
---|---|---|
key |
aliases the field name in json | deserializes the field if json contains key |
ignore |
Deserialization
serde
deserializes using fromJson
, and in all instances returns Result[T, CatchableError]
, where T
is the type being deserialized. For example:
type MyType = object
field1: bool
field2: bool
let jsn1 = """{
"field1": true,
"field2": true
}"""
assert !MyType.fromJson(jsn1) == MyType(field1: true, field2: true)
If there was an error during deserialization, the result of fromJson
will contain it:
import pkg/serde/json
type MyType {.deserialize(mode=Strict).} = object
field1: int
field2: int
let jsn = """{
"field1": 1,
"field2": 2,
"extra": 3
}"""
let res = MyType.fromJson(jsn)
assert res.isFailure
assert res.error of SerdeError
assert res.error.msg == "json field(s) missing in object: {\"extra\"}"
Custom types
If serde
can't de/serialize a custom type, de/serialization can be supported by
overloading %
and fromJson
. For example:
type
Address* = distinct array[20, byte]
SerializationError* = object of CatchableError
func `%`*(address: Address): JsonNode =
%($address)
func fromJson(_: type Address, json: JsonNode): ?!Address =
expectJsonKind(Address, JString, json)
without address =? Address.init(json.getStr), error:
return failure newException(SerializationError,
"Failed to convert '" & $json & "' to Address: " & error.msg)
success address
Serializing to string (toJson
)
toJson
is a shortcut for serializing an object into its serialized string
representation:
import pkg/serde/json
type MyType {.serialize.} = object
field1: string
field2: bool
let mt = MyType(field1: "hw", field2: true)
assert mt.toJson == """{"field1":"hw","field2":true}"""
This comes in handy, for example, when sending API responses:
let availability = getAvailability(...)
return RestApiResponse.response(availability.toJson,
contentType="application/json")
std/json
drop-in replacment
nim-serde
can be used as a drop-in replacement for the standard library's json
module, with a few notable improvements.
Instead of importing std/json
into your application, pkg/serde/json
can be imported
instead:
- import std/json
+ import pkg/serde/json
As with std/json
, %
can be used to serialize a type into a JsonNode
:
import pkg/serde/json
assert %"hello" == newJString("hello")
And %*
can be used to serialize objects:
import pkg/serde/json
let expected = newJObject()
expected["hello"] = newJString("world")
assert %*{"hello": "world"} == expected
As well, serialization of types can be overridden, and serialization of custom types can
be introduced. Here, we are overriding the serialization of int
:
import pkg/serde/json
func `%`(i: int): JsonNode =
newJInt(i + 1)
assert 1.toJson == "2"
parseJson
and exception tracking
Unfortunately, std/json
's parseJson
can raise an Exception
, so proper exception
tracking breaks, eg
## Fails to compile:
## Error: parseJson(me, false, false) can raise an unlisted exception: Exception
import std/json
{.push raises:[].}
type
MyAppError = object of CatchableError
proc parseMe(me: string): JsonNode =
try:
return me.parseJson
except CatchableError as error:
raise newException(MyAppError, error.msg)
assert """{"hello":"world"}""".parseMe == %* { "hello": "world" }
This is due to std/json
's parseJson
incorrectly raising Exception
. This can be
worked around by instead importing serde
and calling its JsonNode.parse
routine.
Note that serde
's JsonNode.parse
returns a Result[JsonNode, CatchableError]
instead of just a plain JsonNode
object as in std/json
's parseJson
:
import pkg/serde/json
{.push raises:[].}
type
MyAppError = object of CatchableError
proc parseMe(me: string): JsonNode {.raises: [MyAppError].} =
without parsed =? JsonNode.parse(me), error:
raise newException(MyAppError, error.msg)
parsed
assert """{"hello":"world"}""".parseMe == %* { "hello": "world" }
Known issues
There is a known issue when using mixins with generic overloaded procs like
fromJson
. At the time of mixin call, only the fromJson
overloads in scope of
the called mixin are available to be dispatched at runtime. There could be other
fromJson
overloads declared in other modules, but are not in scope at the time
the mixin was called. Therefore, anytime fromJson
is called targeting a
declared overload, it may or may not be dispatchable. This can be worked around
by forcing the fromJson
overload into scope at compile time. For example, in
your application where the fromJson
overload is defined, at the bottom of the
module add:
static: MyType.fromJson("")
This will ensure that the MyType.fromJson
overload is dispatchable.
The basic types that serde supports should already have their overloads forced
in scope in the deserializer
module.
For an illustration of the problem, please see this narrow example by @gmega.