Nightly Rust

Why Nighly?

The Rust framework currently requires nightly Rust to run. This often causes issues with the inherent instability and frequent changes in the Rust compiler.

While it would be ideal to have our infrastructure running on stable Rust, there are a number of features that we currently require.

This page is intended to provide a list of all these features that we use, and explain what it would take to renounce them and move to stable.

Nightly features

Feature	Why we need it	What it would take to discard
never_type and exhaustive_patterns	The codec is optimized to work within smart contracts, but also to be used outside them, in a standard environment. Encoding and decoding functions return Result, because we want to allow programs to recover from a failed encoding or decoding attempt. However, this feature is not needed in contracts, where we want failures to interrupt execution immediately. Our solution relies heavily on the never type (!), to guide the compiler to discard any error recovery when compiling contracts. Exhaustive patterns are a nice way to engage with code that is guaranteed at compile time to never return an error.	The never_type seems to be close to being stabilized. Exhaustive patterns, if not stabilized soon, could be replaced with regular patterns with an unreacheable clause on the else.
auto_traits and negative_impls	We have a feature that allows safe conversions from one type to another via the codec. For instance, a smaller integer can be safely cast to a larger one. This feature is central to proxies and transaction syntax. We are using auto-traits and negative implementations to overcome overlaps in the implementations of the CodecFrom trait.	Could be replaced by relying more heavily on proc-macros. We would need a way to generate the CodecFrom<Self> when deriving TopEncode and TopDecode. Alternately, we could wait for an overlapping marker trait impls.
try_trait_v2	Used in the deprecated SCResult type.	Remove SCResult.
maybe_uninit_uninit_array and maybe_uninit_array_assume_init	Used in ManagedVec to_array_of_refs.	There are less elegant ways to write that method, without using these features.
generic_const_exprs	Used everywhere in working with the ManagedVec payload, without using the heap. They allow us to allocate variable sized byte arrays on the stack.	Using universal fixed-sized arrays on heap for handling payloads. Might hurt performance, and might also occasionally run into overflows.
slice_partition_dedup	Used in ManagedVec dedup.	We could probably replace that one with an in-house de-duplication function.
is_sorted	Used in ManagedVec is_sorted, is_sorted_by, and is_sorted_by_key. These methods are currently mostly used in tests.	We could probably replace them with in-house equivalent functions.
panic_info_message	Used in the wasm adapter to retrieve panic messages, for debugging purposes.	Figuring out a way to use the format function provided by the stable version of the library, without performing memory allocation.

Deal breakers

Some of the nightly features that we use could be easily replaced with something from stable. However, there are a few essential ones that we have yet no solution for:

generic_const_exprs

This one is very difficult to get rid of without compromising the ManagedVec functionality. We have also recently added a new ManagedDecimal component that makes use of this feature. This is the main deal breaker right now for switching to stable Rust.

never_type

Our entire codec relies on the never type. There might be a workaround using void enums, not sure if it works. Fortunately, this feature is close to being stabilized.

auto_traits and negative_impls

This is not impossible to overcome, but would need a redesign of some of the codec traits.