Open athei opened 5 days ago
/cmd prdoc --audience runtime_dev
/cmd fmt
Command "fmt" has started 🚀 See logs here
Command "fmt" has finished ✅ See logs here
Okay great. I will finish up this PR by adding a migration to reset the contracts and disabling the failing tests ASAP then.
All GitHub workflows were cancelled due to failure one of the required jobs. Failed workflow url: https://github.com/paritytech/polkadot-sdk/actions/runs/12016268142 Failed job name: test-linux-stable-int
This PR updates pallet_revive to the newest PolkaVM version and adapts the test fixtures and syscall interface to work under 64bit.
Please note that after this PR no 32bit contracts can be deployed (they will be rejected at deploy time). Pre-deployed 32bit contracts are now considered defunct since we changes how parameters are passed for functions with more than 6 arguments.
Fixtures
The fixtures are now built for the 64bit target. I also removed the temporary directory mechanism that triggered a full rebuild every time. It also makes it easier to find the compiled fixtures since they are now always in
target/pallet-revive-fixtures
.Syscall interface
Passing pointer
Registers and pointers are now 64bit wide. This allows us to pass u64 arguments in a single register. Before we needed two registers to pass them. This means that just as before we need one register per pointer we pass. We keep pointers as
u32
argument by truncating the register. This is done since the memory space of PolkaVM is 32bit.Functions with more than 6 arguments
We only have 6 registers to pass arguments. This is why we pass a pointer to a struct when we need more than 6. Before this PR we expected a packed struct and interpreted it as SCALE encoded tuple. However, this was buggy because the
MaxEncodedLen
returned something that was larger than the packed size of the structure. This wasn't a problem before. But now the memory space changed in a way that things were placed at the edges of the memory space and those extra bytes lead to an out of bound access.This is why this PR drops SCALE and expects the arguments to be passed as a pointer to a
C
aligned struct. This avoids unaligned accesses. However, revive needs to adapt its codegen to properly align the structure fields.TODO