feat: Starknet data source validation with improvements #5030
Conversation
|
Just finished implementing block and event trigger filter, which depends on the |
Currently, an array of block handlers is allowed, but only the first one is actually used. This can easily cause confusion for end users. While this is technically a breaking change, it should be fine as the support for Starknet isn't official yet.
Currently, the native `FieldElement` type is used for Starknet data sources, which uses the Montgomery representation internally for fast finite field mulitplication, at the cost of slower (de)serialization. However, in `graph-node`, this type is only used for equality comparison, which calls for a type that's backed by a `u8` array instead. In fact, the `graph::blockchain::DataSource` trait requires the return type of its `address()` method to be `Option<&[u8]>`, which necessitates the use of `u8`-backed types (currently the Starknet data source simply returnes `None`). This commit adds a new type `Felt` that's a wrapper around `[u8; 32]`, and replaces any `FieldElement` occurrence. The `starknet-ff` dependency is also removed.
xJonathanLEI
force-pushed
the
dev/starknet_ds_validation
branch
from
e0d0bd7 to
31777b0
Compare
|
Thanks @mangas for the review and approval. I've rebased the branch to the latest |
|
Hi @incrypto32, I noticed that the base branch has been changed to the |
|
Yes @xJonathanLEI, will be merged to master once the release goes out |
|
Thanks for confirming! |
Long overdue follow-up PR of #4895 to resolve this comment from @mangas.
Main changes:
block_handlerstoblock_handler(and type changed fromVec<_>toOption<_>) because it makes more sense.validate()(this function previously always returns an empty array).Also adds data source validation tests to
chain/starknet/src/chain.rs, similar to those found inchain/near/src/chain.rs. This does not include trigger filter tests inchain/near/src/chain.rsthough - trigger filter hasn't been implemented for Starknet, as noted in the original PR, and should be addressed with another follow-up PR.In the process of implementing validation, I noticed that the
DataSourcetrait has a methodaddress()that returns aOption<&[u8]>, which is impossible to implement with the previous use ofFieldElementin the data source, as this type doesn't have a[u8]backing. The type uses the Montgomery representation internally for fast finite field mulitplication, at the cost of slower (de)serialization. However, ingraph-node, this type is only used for equality comparison, which calls for a type that's backed by au8array instead.Therefore, this PR also adds a new type
Feltthat's a wrapper around[u8; 32], and replaces anyFieldElementoccurrence. Thestarknet-ffdependency is also removed.