Kotlin and Convex type conversion
Custom data types
When receiving values from Convex, you aren't limited to primitive values. You
can create custom @Serializable
classes that will be automatically decoded
from response data.
Consider a Convex query function that returns results like this JavaScript object:
{
name: "Guardians",
uniformColors: ["blue", "white", "red"],
wins: 80n,
losses: 60n
}
That can be represented in Kotlin using:
@Serializable
data class BaseballTeam(
val name: String,
val uniformColors: List<String>,
val wins: @ConvexNum Int,
val losses: @ConvexNum Int)
Then you can pass it as the type argument in your subscribe
call:
convex.subscribe<Team>("mlb:first_place_team", args = mapOf("division" to "AL Central"))
The data from the remote function will be deserialized to your custom class.
Numerical types
Your Convex backend code is written in JavaScript, which has two relatively
common types for numerical data: number
and BigInt
.
number
is used whenever a value is assigned a literal numeric value, whether
42
or 3.14
. BigInt
can be used by adding a trailing n
, like 42n
.
Despite the two types, is very common to use number
for holding either integer
or floating point values in JavaScript.
Because of this, Convex takes extra care to encode values so they won't lose
precision. Since technically the number
type is an IEEE 754 floating point
value, anytime you get a plain number
from Convex it will be represented as
floating point in Kotlin. You can choose to use Double
or Float
, depending
on your needs but be aware that Float
might lose precision from the original.
It also means that Kotlin's Long
type (64 bit) can't be safely stored in a
number
(only 53 bits are available to encode integers) and requires a
BigInt
.
That's a long lead up to explain that in order to represent numerical values in responses from Convex, you need to hint to Kotlin that they should use custom decoding.
You can do this in three ways. Use whichever seems most useful to your project.
-
Annotate the plain Kotlin type (
Int
,Long
,Float
,Double
) with@ConvexNum
-
Use a provided type alias for those types (
Int32
,Int64
,Float32
,Float64
) -
Include a special annotation at the top of any file that defines
@Serializable
classes and just use the plain types with no annotation@file:UseSerializers(
Int64ToIntDecoder::class,
Int64ToLongDecoder::class,
Float64ToFloatDecoder::class,
Float64ToDoubleDecoder::class
)
package com.example.convexapp
import kotlinx.serialization.UseSerializers
// @Serializable classes and things.
In the example, JavaScript's BigInt
type is used by adding a trailing n
to
the wins
and losses
values which lets the Kotlin code use Int
. If instead
the code used regular JavaScript number
types, on the Kotlin side those would
be received as floating point values and deserialization would fail.
If you have a situation like that where number
is used but by convention only
contains integer values, you can handle that in your @Serializable
class.
@Serializable
data class BaseballTeam(
val name: String,
val uniformColors: List<String>,
@SerialName("wins") private val internalWins: Double,
@SerialName("losses") private val internalLosses: Double) {
// Expose the JavaScript number values as Ints.
val wins get() = internalWins.toInt()
val losses get() = internalLosses.toInt()
}
The pattern is to store the Double
values privately and with different names
that the value from the backend. Then add accessors to provide the Int
values.
Field name conversion
This pattern was used above, but it bears describing on its own. Sometimes a
value will be produced on the backend with a key that matches a Kotlin keyword
({fun: true}
) or doesn't conform to Kotlin naming conventions (e.g. starts
with an underscore). You can use @SerialName
to handle those cases.
For example, here's how you can ingest the Convex document ID from a backend response and convert it to a field name that won't trigger Kotlin lint warnings:
@Serializable
data class ConvexDocument(@SerialName("_id") val id: String)