Skip to content

Built-in Types Overview

We've now built and tested libraries in Temper. We've also already seen a lot of the built-in types and features. But to round out ourselves, let's look more at what comes built into Temper. If you want all the details, though, the full references on built-in functions and constants and also types are elsewhere. Here, we'll just be doing some overview.

We also diverge here from our example geometric shapes library, though it served us well for context to get started.

For the types below, some of them can be found as class or interface definitions in the file Implicits.temper in the Temper source code. The interpreter uses much of the code here, but classes here are also separately redefined for each backend.

Special types

Temper has a number of special types, some of which you'll see more than others. See also additional discussion and a diagram of some of these types:

  • Top - The supertype of all other types, equivalent to AnyValue throws Bubble.
  • Bubble - Unusable as an ordinary value. Handled via orelse. It's used in conjunction with other types as function return types, such as Int throws Bubble.
  • Void - Indicates that a function return value must be ignored, allowing for better compatibility with type expectations of some backends.
  • AnyValue - The supertype of all usable values. As for any type that isn't a sealed interface, you can't downcast from AnyValue. This allows type safety while still using appropriate backend values.
  • Null - Supports data interop with external systems. Can be used along with other types, such as String?. There is only one Null value, null. For functions that might fail, usually Bubble is a better choice than making the type nullable with ?. And for functions whose result should be ignored, use Void.
  • Never - The subtype of all other types, representing computations that never complete, such as infinite loops. Computations that exit abnormally use Bubble, not Never.
  • Invalid - A type for when a type can't be computed.

As suggested above, sometimes we can distinguish types at runtime, but this often depends on the specific backend. That's why downcasting is so constrained in Temper. But in addition to sealed interface subtypes, you can also downcast to check against null:

$ let maybeMultiply(x: Float64?, y: Float64?): Float64? {
    // The ` as T` macro has return type `T throws Bubble`, so use `orelse`.
    ((x as Float64) * (y as Float64)) orelse null
  }
interactive#0: void
$ maybeMultiply(1.5, 2.0)
interactive#1: 3.0
$ maybeMultiply(null, 2.0)
interactive#2: null
$ maybeMultiply(1.5, null)
interactive#3: null
$ maybeMultiply(null, null)
interactive#4: null

Function types

Function types are represented specially with the keyword fn. A higher order function reference supports positional arguments only. Some examples include:

Example Means
fn (): Void Type for a function that takes no arguments and returns the void value
fn (Int): Int Type for a function that takes one integer and returns an integer
fn<T> (T): List<T> Type for a generic function with a type parameter <T>
fn (...Int): Boolean Type for a function that takes any number of integers

Numeric types

As we saw earlier, Temper has the following numeric types built in:

  • Float64 - 64-bit IEEE 754 floating point number. These are widely supported across modern languages and platforms. Math operations might not be guaranteed to provide 100% consistent results across all backends, however.
  • Int32 - Signed 32-bit integers that wrap on overflow. Alias Int also exists for this type. Where at all possible, this type maps to a signed 32-bit int type on each backend. And for backends without that, where possible, this type maps to the simplest, most idiomatic numeric representation that can store 32-bit ints. Note that math done outside of Temper might have different semantics than math done in code built from Temper. For example, on Python, the arbitrarily sized int type is used, and on JS, the common number type is used.
  • Int64 - Signed 64-bit integers that wrap on overflow. Where possible, this type maps to a signed 64-bit int type on each backend. Or if necessary, it maps to the most appropriate type that can store 64-bit ints. For example, on JS, the bigint type is used, because number can't store 64 bits of integer precision. On some versions of Lua, a custom type is needed when no standard numeric type can store 64-bit ints.

These two types don't interop directly. For example:

$ 1 + 2.5
1: 1 + 2.5
   ┗━━━━━┛
[interactive#0:1+0-7]@G: No applicable variants in (fn (Int, Int): Int & fn (Int): Int & fn (Float64, Float64): Float64 & fn (Float64): Float64) for inputs (Int, Float64)
interactive#0: fail
$ 1.toFloat64() + 2.5
interactive#1: 3.5
$ 1 + (2.5).toInt()
interactive#2: 3
$ 1 + Infinity.toInt()
interactive#3: fail
$ 1 + NaN.toInt()
interactive#4: fail

The methods toFloat64 and toInt might fail with Bubble depending on the size of the values (or NaN) and the backend's Int size. Variants toFloat64Unsafe and toIntUnsafe always return a value, but they have backend-dependent behavior outside safe values. The unsafe methods should be used only for small values and/or with fuzz testing across backends.

We also plan to provide additional common mathematical operations, including rounding (see issue#27).

Boolean

For logical operations, Temper also has a distinct type:

  • Boolean - Either true or false.

Common Boolean operations exist, such as &&, ||, !, and so on. You can't use numbers as Booleans:

$ if (1) { console.log("hi"); }
1: if (1) { console.log("hi"
       ⇧
[interactive#1:1+4-5]@G: Expected value of type Boolean not Int
1: if (1) { console.log("hi"
       ⇧
[interactive#1:1+4-5]@R: Expected value of type Boolean not Int
interactive#1: fail
$ if (true) { console.log("hi"); }
hi
interactive#2: void

The error repeats in the bad usage example because it happens at multiple stages. The first report is from static type checking. When compiler errors (including static type errors) happen in Temper, you can still carry through to backends or the interpreter. This behavior assists testing portions of code that aren't in error. But then the behavior is backend-dependent. In this case, the interpreter sees the Int value and fails with the same error message as the type checker.

You shouldn't publish libraries with Temper compiler errors.

String

Temper has the following types for representing text:

  • String - Immutable text data.
  • StringIndex - An index into a string that allows efficient left-to-right traversal.

Here are some examples:

$ let what = "libraries";
interactive#0: void
$ let message = "I ❤ ${what}";
interactive#1: void
$ message.split(" ")
interactive#2: ["I","❤","libraries"]
$ message.isEmpty
interactive#3: false
$ message.length
1: message.length
           ┗━━━━┛
[interactive#4:1+8-14]@G: Expected function type, but got Invalid
1: message.length
   ┗━━━━━━━━━━━━┛
[interactive#4:1+0-14]@G: No member length in String__27
interactive#4: fail
$ message.countBetween(String.begin, message.end)
interactive#6: 13
$ message[message.next(message.next(String.begin))]
interactive#7: 10084

Different backends have different internal string representations, so simple random access into string data can be inefficient. StringIndex is an opaque type that connects to an integer type on each backend, but the exact integer differs depending on the "native string encoding."

  • In languages like Rust and C++, the integer is a byte offset. Supplemental code-points use four bytes, so myString.next(i) might return a number up to 4 greater than i.
  • In languages like C#, Java, and JavaScript, the integer is a UTF-16 code unit offet, and a supplemental code-point uses 2 UTF-16 surrogates, so myString.next(i) might return a number up to 2 greater than i.
  • Python3 presents its strings as like an array of code-point, so myString.next(i) is 1 greater than i, unless i is already at the end of the string.

Temper also has raw string syntax to treat backslashes as text data via the raw string tag, as well as multiline string syntax:

$ raw"\d+\.\d+"
interactive#0: "\\d+\\.\\d+"
$ """
    - An outline
      - With indentation
    - Final point
    """
interactive#1: "- An outline\n  - With indentation\n- Final point"

Note how the minimum indentation level denotes the beginning of each line of text. You can also use more than three quote to control the number of quotes needed to finish the string:

$ """"
    """
    """"
interactive#2: "\"\"\""

But you can't do single-line multi-quoted strings. For example, """abc""" is an error. And there are reasons for that. The current and future feature set of Temper string templates goes much deeper than this (see issue#18), but let's move on for now.

Temper has character syntax using a string tagged with char. Character values are simple integer code-point values.

$ char'👪'
interactive#3: 128106

List types

A list is a sequential, random access, sized collection, and built-in types include three list types:

  • Listed - An interface for read-only access to data. Extended by both List and ListBuilder. In the future, third party code might also be able to extend it (see issue#28), but that doesn't work correctly today.
  • List - A class for immutable listed data.
  • ListBuilder - A class for mutable building of listed data.

The classes List and ListBuilder, just as for other built-in types, are represented by core types in each backend to the extent possible, such as by IReadOnlyList or IList in C#, List in Java, Array in JS, or tuple or list in Python.

Here are some examples:

$ let nums = [10, 20, 30]; // inferred type: List<Int>
interactive#0: void
$ nums.length
interactive#1: 3
$ nums[1]
interactive#2: 20
$ nums.add(40);
1: nums.add(40);
        ┗━┛
[interactive#3:1+5-8]@G: Expected function type, but got Invalid
1: nums.add(40);
   ┗━━━━━━━━━━┛
[interactive#3:1+0-12]@G: No member add in List__10 | Listed__44
interactive#3: fail
$ let nums2 = do {
    let builder = new ListBuilder<Int>();
    builder.addAll(nums);
    builder.add(40);
    builder.toList()
  }
interactive#4: void
$ nums2
interactive#5: [10,20,30,40]
$ nums2.map { (n): Int => n / 10 }
interactive#6: [1,2,3,4]

For many common cases, functions should accept Listed views and return immutable List data:

let calculateSomething<T>(items: Listed<T>): List<T> {
  // ... meaningful calculations here ...
}

As mentioned earlier, Temper allowes for mutable data but prefers immutable.

Map types

Temper's map types are like its list types, but they associate keys more arbitrarily with values:

  • Mapped - An interface for read-only access to data, extended by both Map and MapBuilder.
  • Map - A class for immutable listed data.
  • MapBuilder - A class for mutable building of listed data.

As for list types, Map and MapBuilder use standard backend types where available, such as IReadOnlyDictionary or IDictionary in C#, Map in Java or JS, or dict in Python.

$ let messages = new Map([
    new Pair(200, "OK"),
    new Pair(404, "Not Found"),
    new Pair(500, "Internal Server Error"),
  ]);
interactive#0: void
$ messages[200]
interactive#1: "OK"
$ messages.toList().map { (entry): Int => entry.key }
interactive#2: [200,404,500]
$ let neighbors = do {
    let builder = new MapBuilder<String, List<String>>();
    builder["Honduras"] = ["El Salvador", "Guatemala", "Nicaragua"];
    builder["Nicaragua"] = ["Costa Rica", "Honduras"];
    builder.toMap()
  };
interactive#3: void
$ neighbors["Nicaragua"]
interactive#4: ["Costa Rica","Honduras"]

For now, you can use only Int or String keys. We hope to expand support to user-defined key types in the future (see issue#29).

Other types

Temper has additional built-in types, but above discussion highlights some of the most important ones. Also, some types currently in Implicits.temper, such as Deque<T> and DenseBitVector might move to support libraries in the future (see issue#30).

There's also currently a standard library called "std" with modules "std/regex" and "std/testing". These also are likely to see some reorgnization before 1.0 (see issue#26). We already saw a bit of testing earlier. And we'll take a look at regular expressions soon, but first let's understand how Temper handles memory management.