Conversation
Switch the default Cranelift optimization level from `None` to `Speed`, improving WASM handler execution performance. The optimization level is configurable via the `GRAPH_WASM_OPT_LEVEL` environment variable (valid values: `none`, `speed`, `speed_and_size`; default: `speed`). This is safe because Cranelift adheres to the Wasm spec and NaN canonicalization is already enabled. The increased compilation time is a one-time cost per module, amortized over thousands of triggers.
Move the per-trigger GasCounter and deterministic_host_trap flag from standalone variables captured by linker closures into the Store data (WasmInstanceData). This is a pure refactoring with no behavior change. Previously, GasCounter and an Arc<AtomicBool> for deterministic_host_trap were created as separate variables and captured by each linker closure. Now they live on WasmInstanceData and are accessed through caller.data()/caller.data_mut() inside the closures. This decouples the linker closures from per-trigger state, which is a prerequisite for caching InstancePre on ValidModule (reusing the linker across triggers).
…ation Extract linker construction from per-trigger `from_valid_module_with_ctx` into a standalone `build_linker()` function called once at module validation time. The linker is pre-linked via `linker.instantiate_pre()` and the resulting `InstancePre` is stored on `ValidModule`. This eliminates the per-trigger cost of: - Rebuilding a Linker with ~60 func_wrap_async registrations - Resolving imports against the module Chain-specific host functions (e.g. ethereum.call) are now dispatched generically: the linker registers them by import name and looks up the actual HostFn from caller.data().ctx.host_fns at call time. Conditional functions (ipfs.getBlock, arweave.transactionData, box.profile) are now linked unconditionally since they already check feature flags or return errors internally.
Every WASM heap allocation (API version > 0.0.4) called asc_type_id(), which did a call_async into the WASM module for a trivial idof<T>() switch. The result is deterministic per compiled module, yet every allocation paid ~200-500ns of call_async overhead (fiber creation, epoch checking, context switching). Cache the results on ValidModule using a parking_lot::RwLock<HashMap>, so after the first trigger warms the cache, all subsequent lookups are cheap read-lock hits. Lock guards never span .await points.
…ic discovery Replace the dynamic loop that discovered chain-specific host functions by excluding BUILTIN_IMPORT_NAMES with explicit registration calls for each chain function (ethereum.call, ethereum.getBalance, ethereum.hasCode). Extract the dispatcher into a `link_chain_host_fn` helper that: - Takes a `&'static str` name, eliminating the `Box::leak` for name strings - Precomputes metrics strings once rather than on every call - Returns a proper error instead of panicking when a chain host function is not available for the current chain Delete the BUILTIN_IMPORT_NAMES constant (55 entries) which is no longer needed.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The common thread is making the WASM runtime faster and cleaner — the first four commits reduce per-trigger overhead, and the last one is a cleanup enabled by the linker extraction in commit 3.