Upgrade

View Source

This document is intended to simplify upgrading to newer versions by extending the changelog.

0.34 -> 0.35

rustler_sys as a standalone library has been replaced by an embedded rustler::sys submodule. Due to how the rustler_/::sys initialisation works, it is not possible to use the new rustler in conjunction with rustler_sys. A simple textual replacement is enough, though.

0.33 -> 0.34

  1. NIF implementations are now discovered automatically, the respective argument in the rustler::init! macro should be removed. If a NIF implementation should not be exported, it must be disabled with a #[cfg] marker.
  2. The functionality related to the derive feature is now unconditionally active. The feature flag is kept for compatibility for now but will be removed in the future.
  3. To use a type as a resource, the Resource trait should now be implemented on the type, which also allows for specifying a destructor (taking an Env argument) or a down callback for process monitoring. If the recommended resource_impl attribute is used on the impl block, the type will by default be automatically registered and the IMPLEMENTS_... constants will be set for implemented callbacks.

0.32 -> 0.33

The macro changes that where already carried out in version 0.22 are now mandatory, the deprecated macros have been removed. Please see below for documentation on how to convert from the old to the new set of macros.

0.31 -> 0.32

  1. The functionality of rustler_bigint has moved into rustler. The library will still work, but it can now also be replaced by activating the new big_integer feature on rustler. The new rustler::BigInt is a re-export of num_bigint::BigInt in contrast to rustler_bigint::BigInt, which was a wrapper. For most codebases, it will be enough to activate the feature and replace all rustler_bigint::BigInt usages by rustler::BigInt (or num_bigint::BigInt).
  2. serde_rustler has been integrated into rustler behind the feature flag serde. Arbitrary, serde-compatible objects (i.e. with Deserialize or Serialize impls) can be wrapped in SerdeTerm to use them in place of Encoder or Decoder. The API is for now considered experimental.

0.29 -> 0.30

  1. rustler_crates configuration is deprecated in favor of explicitly passing options on use Rustler or configuring the module in your config/*.exs files.

  2. Env::send and OwnedEnv::send_and_clear will now return a Result. Updating will thus introduce warnings about unused Results. To remove the warnings without changing behaviour, the Results can be "used" as

    let _ = env.send(...)

    Neither the Ok nor the Err case carry additional information so far. An error is returned if either the receiving or the sending process is dead. See also enif_send.

  3. As Term::get_type is now implemented using enif_get_type on all non-Windows systems, some cases of the TermType enum are changed, removed, or added (on all systems):

    1. EmptyList is dropped, List is returned for both empty and non-empty lists
    2. Exception is dropped
    3. Number is split into Integer and Float (if NIF 2.14 support is explicitly enforced, only Float is returned)

  4. The default NIF version is raised to 2.15 to make use of enif_get_type. To use a compiled NIF with an older version than OTP22, disable the default features and expliictly use the nif_version_2_14 feature in the library's Cargo.toml:

    rustler = { version = "0.30", default-features = false, features = ["derive", "nif_version_2_14"] }

  5. As noted for the 0.28 -> 0.29 transition below, the environment variable RUSTLER_NIF_VERSION will not be considered anymore from 0.30 onwards.

0.28 -> 0.29

RUSTLER_NIF_VERSION is deprecated and will not be considered anymore for 0.30. The NIF version will also not be guessed anymore from a potentially available installed Erlang version. By default, NIF libraries will now be compiled against NIF version 2.14 which is compatible down to OTP21. The default will be adjusted along with the supported OTP versions.

If additional features are required that use newer NIF versions, these can be included explicitly in the project's Cargo.toml as, e.g.

[dependencies]
rustler = { version = "0.30", features = ["nif_version_2_17"] }

With this configuration, the resulting NIF library will only work from OTP26 onwards, but will also have access to the largest feature set.

0.26 -> 0.27

MIX_ENV is no longer considered for determining the build profile. Now, the profile defaults to :release. Use the :mode option to pick another profile explicitly. See #496.

0.21 -> 0.22

0.22 changes how to define NIFs. Users upgrading to 0.22 should to do these things:

  1. Replace rustler_atoms! with rustler::atoms!
  2. Replace resource_struct_init! with rustler::resource!
  3. Replace rustler::rustler_export_nifs! with rustler::init!
  4. Use the new rustler::nif proc_macro to declare NIFs

Replacing rustler_atoms! with rustler::atoms! is fairly simple and already sufficiently described in CHANGELOG.md. Similarly, replacing resource_struct_init! with rustler::resource! is a simple rename, so this does not need additional examples here.

Replace rustler::rustler_export_nifs! with rustler::init!

rustler::init! in combination with the new rustler::nif proc_macro simplifies exporting NIFs. Before, the NIFs and their arity needed to be specified using tuple syntax:

rustler::rustler_export_nifs! {
    "Elixir.Math",
    [
        ("add", 2, add),
        ("long_running_operation", 0, long_running_operation, SchedulerFlags::DirtyCpu)
    ],
    None
}

Now, listing the NIFs directly is sufficient:

rustler::init!("Elixir.Math", [add, long_running_operation]);

With this new macro, defining an on_load function (e.g. to set up a resource with rustler::resource!), is done like this:

rustler::init!("Elixir.Math", [add, long_running_operation], load = a_function);

Note that NIF flags such as SchedulerFlags::DirtyCpu are not declared in rustler::init!, but using the proc_macro rustler::nif. See further below for information on migration NIF flags.

Use the new rustler::nif proc_macro to declare NIFs

0.22 introduces a new proc_macro allowing to spell out the parameter of a NIF directly instead of using an args: &[Term<'a>]. Lets consider an example add(), where the Elixir function looks like this:

def add(left, right), do: :erlang.nif_error(:not_loaded)

Previously, the signature of the corresponding NIF might have looked like this:

fn add<'a>(env: Env<'a>, args: &[Term<'a>]) -> Result<Term<'a>, Error>

When calling the NIF from Elixir as add(1, 2), args would then contain two Term, one for 1, and one for 2. With 0.22, this becomes more obvious, as the NIFs signature resembles the Elixir function's signature:

#[rustler::nif]
fn add(a: i64, b: i64) -> i64

Under the hood, this is implemented by the rustler::nif proc_macro. For the new form to work, the parameters' types need to implement Decoder, and the return type needs to implement Encoder.

What if Env is required in the function?

Sometimes, we still need the environment Env for the NIF. For example, if work with Binary and OwnedBinary, the environment would be needed to create a Binary from an OwnedBinary. To allow this, env: Env<'a> can be added explicitly as well:

#[rustler::nif]
pub fn map_entries_sorted<'a>(env: Env<'a>, iter: MapIterator<'a>) -> NifResult<Vec<Term<'a>>>

env can then be used the same way as before.

Migrating Flags and Rename

The rustler::nif proc_macro allows setting options directly on a NIF. Assume that we have a NIF called _long_running_operation, which used to be declared prior to Rustler v0.22 like this:

// Before

rustler::rustler_export_nifs! {
    "Elixir.SomeNif",
    [
        // Note that the function in Rust is long_running_operation, but the NIF is exported as
        // _long_running_operation!
        ("_long_running_operation", 0, long_running_operation, SchedulerFlags::DirtyCpu)
    ],
    None
}

fn long_running_operation<'a>(env: Env<'a>, _args: &[Term<'a>]) -> Result<Term<'a>, Error> {
  // ...
}

This definition declares that a function _long_running_operation with arity zero is to be exported, and that this function should be schedules on the DirtyCpu scheduler. With the changes in Rustler v0.22, the function would be declared like this:

// Now

rustler::init!("Elixir.SomeNif", [long_running_operation]);

#[rustler::nif(
    rename = "_long_running_operation",
    schedule = "DirtyCpu"
  )]
fn long_running_operation() -> TheProperReturnType {
 // ..
}