/conv2

Conversion traits for Rust

Primary LanguageRustOtherNOASSERTION

conv2 is a fork of conv, written by Daniel Keep. It's in the progress of being updated to modern idiomatic Rust.

This crate provides a number of conversion traits with more specific semantics than those provided by as or From/Into.

The goal with the traits provided here is to be more specific about what generic code can rely on, as well as provide reasonably self-describing alternatives to the standard From/Into traits. For example, the although T: From<U> might be satisfied, it imposes no restrictions on the kind of conversion being implemented. As such, the traits in this crate try to be very specific about what conversions are allowed. This makes them less generally applicable, but more useful where they do apply.

In addition, From/Into requires all conversions to succeed or panic. All conversion traits in this crate define an associated error type, allowing code to react to failed conversions as appropriate.

Compatibility

conv2 is compatible with Rust 1.61 and higher.

Overview

The following traits are used to define various conversion semantics:

  • [ApproxFrom] - approximate conversions, with selectable approximation scheme (see [ApproxScheme]).
  • [ValueFrom] - exact, value-preserving conversions.

When defining a conversion, try to implement the *From trait variant where possible. When using a conversion, try to depend on the *Into trait variant where possible. This is because the *Into traits automatically use *From implementations, but not the reverse. Implementing *From and using *Into ensures conversions work in as many contexts as possible.

These extension methods are provided to help with some common cases:

  • [ConvUtil::approx_as] - approximates to Dst with the DefaultApprox scheme.
  • [ConvUtil::approx_as_by] - approximates to Dst with the scheme S.
  • [ConvUtil::into_as<Dst>] - converts to Dst using Into::into.
  • [ConvUtil::try_as<Dst>] - converts to Dst using TryInto::try_into.
  • [ConvUtil::value_as<Dst>] - converts to Dst using ValueInto::value_into.
  • [ConvAsUtil::approx] - approximates to an inferred destination type with the DefaultApprox scheme.
  • [ConvAsUtil::approx_by] - approximates to an inferred destination type with the scheme S.
  • [Saturate::saturate]- saturates on overflow.
  • [UnwrapOk::unwrap_ok] - unwraps results from conversions that cannot fail.
  • [UnwrapOrInf::unwrap_or_inf] - saturates to ±∞ on failure.
  • [UnwrapOrInvalid::unwrap_or_invalid] - substitutes the target type's "invalid" sentinel value on failure.
  • [UnwrapOrSaturate::unwrap_or_saturate] - saturates to the maximum or minimum value of the target type on failure.

Provided Implementations

The crate provides several blanket implementations:

  • *From<A> for A (all types can be converted from and into themselves).
  • *Into<Dst> for Src where Dst: *From<Src> (*From implementations imply a matching *Into implementation).

Conversions for the builtin numeric (integer and floating point) types are provided. In general, ValueFrom conversions exist for all pairs except for float → integer (since such a conversion is generally unlikely to exactly succeed) and f64 → f32 (for the same reason). ApproxFrom conversions with the DefaultApprox scheme exist between all pairs. ApproxFrom with the Wrapping scheme exist between integers.

Errors

A number of error types are defined in the [errors] module. Generally, conversions use whichever error type most narrowly defines the kinds of failures that can occur. For example:

  • ValueFrom<u8> for u16 cannot possibly fail, and as such it uses NoError.
  • ValueFrom<i8> for u16 can only fail with a negative overflow, thus it uses the NegOverflow type.
  • ValueFrom<i32> for u16 can overflow in either direction, hence it uses RangeError.
  • Finally, ApproxFrom<f32> for u16 can overflow (positive or negative), or attempt to convert NaN; FloatError covers those three cases.

Because there are numerous error types, the GeneralError enum is provided. From<E, T> for GeneralError<T> exists for each error type E<T> defined by this crate (even for NoError!), allowing errors to be translated automatically by the ? operator. In fact, all errors can be "expanded" to all more general forms (e.g. NoError → NegOverflow, PosOverflow → RangeError → FloatError).

Aside from NoError, the various error types wrap the input value that you attempted to convert. This is so that non-Copy types do not need to be pre-emptively cloned prior to conversion, just in case the conversion fails. A downside is that this means there are many, many incompatible error types.

To help alleviate this, there is also GeneralErrorKind, which is simply GeneralError<T> without the payload, and all errors can be converted into it directly.

The reason for not just using GeneralErrorKind in the first place is to statically reduce the number of potential error cases you need to deal with. It also allows the Unwrap* extension traits to be defined without the possibility for runtime failure (e.g. you cannot use unwrap_or_saturate with a FloatError, because what do you do if the error is NotANumber; saturate to max or to min? Or panic?).

Examples

// This *cannot* fail, so we can use `unwrap_ok` to discard the `Result`.
assert_eq!(u8::value_from(0u8).unwrap_ok(), 0u8);

// This *can* fail. Specifically, it can overflow toward negative infinity.
assert_eq!(u8::value_from(0i8),     Ok(0u8));
assert_eq!(u8::value_from(-1i8),    Err(NegOverflow(-1)));

// This can overflow in *either* direction; hence the change to `RangeError`.
assert_eq!(u8::value_from(-1i16),   Err(RangeError::NegOverflow(-1)));
assert_eq!(u8::value_from(0i16),    Ok(0u8));
assert_eq!(u8::value_from(256i16),  Err(RangeError::PosOverflow(256)));

// We can use the extension traits to simplify this a little.
assert_eq!(u8::value_from(-1i16).unwrap_or_saturate(),  0u8);
assert_eq!(u8::value_from(0i16).unwrap_or_saturate(),   0u8);
assert_eq!(u8::value_from(256i16).unwrap_or_saturate(), 255u8);

// Obviously, all integers can be "approximated" using the default scheme (it
// doesn't *do* anything), but they can *also* be approximated with the
// `Wrapping` scheme.
assert_eq!(
    <u8 as ApproxFrom<_, DefaultApprox>>::approx_from(400u16),
    Err(PosOverflow(400)));
assert_eq!(
    <u8 as ApproxFrom<_, Wrapping>>::approx_from(400u16),
    Ok(144u8));

// This is rather inconvenient; as such, there are a number of convenience
// extension methods available via `ConvUtil` and `ConvAsUtil`.
assert_eq!(400u16.approx(),                       Err::<u8, _>(PosOverflow(400)));
assert_eq!(400u16.approx_by::<Wrapping>(),        Ok::<u8, _>(144u8));
assert_eq!(400u16.approx_as::<u8>(),              Err(PosOverflow(400)));
assert_eq!(400u16.approx_as_by::<u8, Wrapping>(), Ok(144));

// Integer -> float conversions *can* fail due to limited precision.
// Once the continuous range of exactly representable integers is exceeded, the
// provided implementations fail with overflow errors.
assert_eq!(f32::value_from(16_777_216i32), Ok(16_777_216.0f32));
assert_eq!(f32::value_from(16_777_217i32), Err(RangeError::PosOverflow(16_777_217)));

// Float -> integer conversions have to be done using approximations. Although
// exact conversions are *possible*, "advertising" this with an implementation
// is misleading.
//
// Note that `DefaultApprox` for float -> integer uses whatever rounding
// mode is currently active (*i.e.* whatever `as` would do).
assert_eq!(41.0f32.approx(), Ok(41u8));
assert_eq!(41.3f32.approx(), Ok(41u8));
assert_eq!(41.5f32.approx(), Ok(41u8));
assert_eq!(41.8f32.approx(), Ok(41u8));
assert_eq!(42.0f32.approx(), Ok(42u8));

assert_eq!(255.0f32.approx(), Ok(255u8));
assert_eq!(256.0f32.approx(), Err::<u8, _>(FloatError::PosOverflow(256.0)));

// Sometimes, it can be useful to saturate the conversion from float to
// integer directly, then account for NaN as input separately. The `Saturate`
// extension trait exists for this reason.
assert_eq!((-23.0f32).approx_as::<u8>().saturate(), Ok(0));
assert_eq!(302.0f32.approx_as::<u8>().saturate(), Ok(255u8));
assert!(std::f32::NAN.approx_as::<u8>().saturate().is_err());

// If you really don't care about the specific kind of error, you can just rely
// on automatic conversion to `GeneralErrorKind`.
fn too_many_errors() -> Result<(), GeneralErrorKind> {
    let x: u8 = 0u8.value_into()?;
    assert_eq!(x, 0);
    let y: i8 = 0u8.value_into()?;
    assert_eq!(y, 0);
    let z: i16 = 0u8.value_into()?;
    assert_eq!(z, 0);

    let x: u8 = 0.0f32.approx()?;
    assert_eq!(x, 0u8);
    Ok(())
}
too_many_errors().unwrap();