Skip to content
Unverified Commit 559fa1db authored by Alexander Theißen's avatar Alexander Theißen Committed by GitHub
Browse files

Add initial version of `pallet_revive` (#5293)

This is a heavily modified and stripped down version of
`pallet_contracts`. We decided to fork instead of extend the old pallet.
Reasons for that are:

- There is no benefit of supporting both on the same pallet as the
intended payload for the new pallet (recompiled YUL) will be using a
different ABI.
- It is much easier since it allows us to remove all the code that was
necessary to support Wasm and focus fully on running cross compiled YUL
contracts.

**The code is reviewable but can't be merged because it depends on an
unreleased version of PolkaVM via git.**

## Current state

All tests are passing and the code is not quick and dirty but written to
last. The work is not finished, though. It is included in the
`kitchensink-runtime` and a node can be built. However, we merge early
in order to be able to start testing other components as early as
possible.

Outstanding changes are tracked here and will be merged separately:
https://github.com/paritytech/polkadot-sdk/issues/5308

## Syscall Interface

The syscall interface is best explored by generating the docs of this
crate and looking at the `SyscallDoc` trait. Arguments are passed in
registers a0-a5 in the order they are listed. If there are more than 6
arguments (call, instantiate) a pointer to a packed struct of the
arguments is expected as the only argument. I plan to create variants of
those syscalls with less arguments specifically for YUL.

Functions are just referenced by their name as ASCII within the PolkaVM
container. Rather than by a syscall number as it was the case in the
last implementation.
 

## Changes vs. `pallet_contracts`

The changes are too numerous to list them all here. This is an
incomplete list:

- Use PolkaVM instead of wasmi to execute contracts
- Made Runtime generic over a new `Memory` trait as we can't map memory
directly on PolkaVM anymore
- No static verification on code upload. Everything is a determinstic
runtime failure
- Removed all migrations and reset the pallet version
- Removed the nonce storage item and instead use the deployers account
nonce to generate a unique trie
- We now bump the deployers account nonce on contract instantiation to
they are bumped even within a batch transaction
- Removed the instantiation nonce host function: We should add a new
`instantiate` variant as a replacement for thos
- ContractInfoOf of uses the indentity hasher now
- Remove the determinism feature: User of that feature should switch to
soft floats
- The `unstable` attribute has been replaced by a `api_version`
attribute to declare at which version an API became available
	- leaving out that attribute makes the API effectively unstable
- a new `api_version` field on the CodeInfo makes sure that old
contracts can't access new APIs (necessary due to lack of static
verification.
- Added a `behaviour_version` field to CodeInfo that can used if we need
to introduce breaking changes and keep the old behaviour for existing
contracts
- Unified storage vs. transient and fixed vs. variable sized keys all
into one set of multiplexing host functions
- Removed all contract observeable limits from the `Config` trait and
instead hardcode them
- Removed the Schedule
- Removed all deprecated host functions
- Simplify chain extension as preperation for making it a pre-compile

---------

Co-authored-by: command-bot <>
parent 4ffccac4
Pipeline #492414 waiting for manual action with stages
in 18 minutes and 56 seconds
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment