test

Signed-off-by: Cyrill Leutwiler <bigcyrill@hotmail.com>

.github/workflows/test.yml

@@ -18,9 +18,28 @@ env:
   POLKADOT_VERSION: polkadot-stable2506-2
 
 jobs:
+  fmt:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout repo
+        uses: actions/checkout@v4
+
+      - name: Setup Rust toolchain
+        uses: actions-rust-lang/setup-rust-toolchain@v1
+
+      - name: Install nightly toolchain
+        run: rustup toolchain install nightly
+
+      - name: Install rustfmt for nightly
+        run: rustup component add --toolchain nightly rustfmt
+
+      - name: Cargo fmt
+        run: cargo +nightly fmt --all -- --check
+
   cache-polkadot:
     name: Build and cache Polkadot binaries on ${{ matrix.os }}
     runs-on: ${{ matrix.os }}
+    needs: [fmt]
     strategy:
       matrix:
         os: [ubuntu-24.04, macos-14]

.gitignore

@@ -13,4 +13,3 @@ resolc-compiler-tests
 workdir
 
 !/schema.json
-!/dev-genesis.json

Cargo.lock

@@ -4526,6 +4526,27 @@ dependencies = [
  "windows-sys 0.59.0",
 ]
 
+[[package]]
+name = "ml-test-runner"
+version = "0.1.0"
+dependencies = [
+ "alloy",
+ "anyhow",
+ "clap",
+ "revive-dt-common",
+ "revive-dt-compiler",
+ "revive-dt-config",
+ "revive-dt-core",
+ "revive-dt-format",
+ "revive-dt-node",
+ "revive-dt-node-interaction",
+ "revive-dt-report",
+ "temp-dir",
+ "tokio",
+ "tracing",
+ "tracing-subscriber",
+]
+
 [[package]]
 name = "moka"
 version = "0.12.10"

crates/common/src/cached_fs/mod.rs

@@ -1,9 +1,11 @@
 //! This module implements a cached file system allowing for results to be stored in-memory rather
 //! rather being queried from the file system again.
 
-use std::fs;
-use std::io::{Error, Result};
-use std::path::{Path, PathBuf};
+use std::{
+    fs,
+    io::{Error, Result},
+    path::{Path, PathBuf},
+};
 
 use moka::sync::Cache;
 use once_cell::sync::Lazy;

crates/common/src/futures/poll.rs

@@ -1,5 +1,4 @@
-use std::ops::ControlFlow;
-use std::time::Duration;
+use std::{ops::ControlFlow, time::Duration};
 
 use anyhow::{Context as _, Result, anyhow};
 

crates/common/src/macros/define_wrapper_type.rs

@@ -135,6 +135,6 @@ macro_rules! define_wrapper_type {
     };
 }
 
-/// Technically not needed but this allows for the macro to be found in the `macros` module of the
-/// crate in addition to being found in the root of the crate.
+/// Technically not needed but this allows for the macro to be found in the `macros` module of
+/// the crate in addition to being found in the root of the crate.
 pub use {define_wrapper_type, impl_for_wrapper};

crates/common/src/types/mode.rs

@@ -1,9 +1,7 @@
 use crate::types::VersionOrRequirement;
 use semver::Version;
 use serde::{Deserialize, Serialize};
-use std::fmt::Display;
-use std::str::FromStr;
-use std::sync::LazyLock;
+use std::{fmt::Display, str::FromStr, sync::LazyLock};
 
 /// This represents a mode that a given test should be run with, if possible.
 ///

crates/common/src/types/private_key_allocator.rs

@@ -1,5 +1,4 @@
-use alloy::primitives::U256;
-use alloy::signers::local::PrivateKeySigner;
+use alloy::{primitives::U256, signers::local::PrivateKeySigner};
 use anyhow::{Context, Result, bail};
 
 /// This is a sequential private key allocator. When instantiated, it allocated private keys in

crates/compiler/src/lib.rs

@@ -10,8 +10,7 @@ use std::{
     pin::Pin,
 };
 
-use alloy::json_abi::JsonAbi;
-use alloy::primitives::Address;
+use alloy::{json_abi::JsonAbi, primitives::Address};
 use anyhow::{Context as _, Result};
 use semver::Version;
 use serde::{Deserialize, Serialize};

crates/compiler/src/revive_resolc.rs

@@ -253,45 +253,44 @@ impl SolidityCompiler for Resolc {
                         .evm
                         .and_then(|evm| evm.bytecode.clone())
                         .context("Unexpected - Contract compiled with resolc has no bytecode")?;
-                    let abi = {
-                        let metadata = contract_information
-                            .metadata
-                            .as_ref()
-                            .context("No metadata found for the contract")?;
-                        let solc_metadata_str = match metadata {
-                            serde_json::Value::String(solc_metadata_str) => {
-                                solc_metadata_str.as_str()
-                            }
-                            serde_json::Value::Object(metadata_object) => {
-                                let solc_metadata_value = metadata_object
-                                    .get("solc_metadata")
-                                    .context("Contract doesn't have a 'solc_metadata' field")?;
-                                solc_metadata_value
-                                    .as_str()
-                                    .context("The 'solc_metadata' field is not a string")?
-                            }
-                            serde_json::Value::Null
-                            | serde_json::Value::Bool(_)
-                            | serde_json::Value::Number(_)
-                            | serde_json::Value::Array(_) => {
-                                anyhow::bail!("Unsupported type of metadata {metadata:?}")
-                            }
+                    let abi =
+                        {
+                            let metadata = contract_information
+                                .metadata
+                                .as_ref()
+                                .context("No metadata found for the contract")?;
+                            let solc_metadata_str = match metadata {
+                                serde_json::Value::String(solc_metadata_str) => {
+                                    solc_metadata_str.as_str()
+                                }
+                                serde_json::Value::Object(metadata_object) => {
+                                    let solc_metadata_value = metadata_object
+                                        .get("solc_metadata")
+                                        .context("Contract doesn't have a 'solc_metadata' field")?;
+                                    solc_metadata_value
+                                        .as_str()
+                                        .context("The 'solc_metadata' field is not a string")?
+                                }
+                                serde_json::Value::Null
+                                | serde_json::Value::Bool(_)
+                                | serde_json::Value::Number(_)
+                                | serde_json::Value::Array(_) => {
+                                    anyhow::bail!("Unsupported type of metadata {metadata:?}")
+                                }
+                            };
+                            let solc_metadata =
+							serde_json::from_str::<serde_json::Value>(solc_metadata_str).context(
+								"Failed to deserialize the solc_metadata as a serde_json generic value",
+							)?;
+                            let output_value = solc_metadata
+                                .get("output")
+                                .context("solc_metadata doesn't have an output field")?;
+                            let abi_value = output_value
+                                .get("abi")
+                                .context("solc_metadata output doesn't contain an abi field")?;
+                            serde_json::from_value::<JsonAbi>(abi_value.clone())
+                                .context("ABI found in solc_metadata output is not valid ABI")?
                         };
-                        let solc_metadata = serde_json::from_str::<serde_json::Value>(
-                            solc_metadata_str,
-                        )
-                        .context(
-                            "Failed to deserialize the solc_metadata as a serde_json generic value",
-                        )?;
-                        let output_value = solc_metadata
-                            .get("output")
-                            .context("solc_metadata doesn't have an output field")?;
-                        let abi_value = output_value
-                            .get("abi")
-                            .context("solc_metadata output doesn't contain an abi field")?;
-                        serde_json::from_value::<JsonAbi>(abi_value.clone())
-                            .context("ABI found in solc_metadata output is not valid ABI")?
-                    };
                     map.insert(contract_name, (bytecode.object, abi));
                 }
             }

crates/compiler/src/solc.rs

@@ -21,8 +21,7 @@ use foundry_compilers_artifacts::{
     output_selection::{
         BytecodeOutputSelection, ContractOutputSelection, EvmOutputSelection, OutputSelection,
     },
-    solc::CompilerOutput as SolcOutput,
-    solc::*,
+    solc::{CompilerOutput as SolcOutput, *},
 };
 use semver::Version;
 use tokio::{io::AsyncWriteExt, process::Command as AsyncCommand};
@@ -284,8 +283,9 @@ impl SolidityCompiler for Solc {
         _optimize_setting: ModeOptimizerSetting,
         pipeline: ModePipeline,
     ) -> bool {
-        // solc 0.8.13 and above supports --via-ir, and less than that does not. Thus, we support mode E
-        // (ie no Yul IR) in either case, but only support Y (via Yul IR) if the compiler is new enough.
+        // solc 0.8.13 and above supports --via-ir, and less than that does not. Thus, we support
+        // mode E (ie no Yul IR) in either case, but only support Y (via Yul IR) if the compiler
+        // is new enough.
         pipeline == ModePipeline::ViaEVMAssembly
             || (pipeline == ModePipeline::ViaYulIR && self.compiler_supports_yul())
     }

crates/config/src/lib.rs

@@ -202,11 +202,11 @@ impl AsRef<ReportConfiguration> for Context {
 
 #[derive(Clone, Debug, Parser, Serialize)]
 pub struct TestExecutionContext {
-    /// The working directory that the program will use for all of the temporary artifacts needed at
-    /// runtime.
+    /// The working directory that the program will use for all of the temporary artifacts needed
+    /// at runtime.
     ///
-    /// If not specified, then a temporary directory will be created and used by the program for all
-    /// temporary artifacts.
+    /// If not specified, then a temporary directory will be created and used by the program for
+    /// all temporary artifacts.
     #[clap(
         short,
         long,
@@ -282,11 +282,11 @@ pub struct TestExecutionContext {
 
 #[derive(Clone, Debug, Parser, Serialize)]
 pub struct BenchmarkingContext {
-    /// The working directory that the program will use for all of the temporary artifacts needed at
-    /// runtime.
+    /// The working directory that the program will use for all of the temporary artifacts needed
+    /// at runtime.
     ///
-    /// If not specified, then a temporary directory will be created and used by the program for all
-    /// temporary artifacts.
+    /// If not specified, then a temporary directory will be created and used by the program for
+    /// all temporary artifacts.
     #[clap(
         short,
         long,
@@ -580,8 +580,8 @@ pub struct ResolcConfiguration {
 pub struct PolkadotParachainConfiguration {
     /// Specifies the path of the polkadot-parachain node to be used by the tool.
     ///
-    /// If this is not specified, then the tool assumes that it should use the polkadot-parachain binary
-    /// that's provided in the user's $PATH.
+    /// If this is not specified, then the tool assumes that it should use the polkadot-parachain
+    /// binary that's provided in the user's $PATH.
     #[clap(
         id = "polkadot-parachain.path",
         long = "polkadot-parachain.path",
@@ -624,8 +624,8 @@ pub struct GethConfiguration {
 pub struct KurtosisConfiguration {
     /// Specifies the path of the kurtosis node to be used by the tool.
     ///
-    /// If this is not specified, then the tool assumes that it should use the kurtosis binary that's
-    /// provided in the user's $PATH.
+    /// If this is not specified, then the tool assumes that it should use the kurtosis binary
+    /// that's provided in the user's $PATH.
     #[clap(
         id = "kurtosis.path",
         long = "kurtosis.path",
@@ -663,8 +663,8 @@ pub struct KitchensinkConfiguration {
 pub struct ReviveDevNodeConfiguration {
     /// Specifies the path of the revive dev node to be used by the tool.
     ///
-    /// If this is not specified, then the tool assumes that it should use the revive dev node binary
-    /// that's provided in the user's $PATH.
+    /// If this is not specified, then the tool assumes that it should use the revive dev node
+    /// binary that's provided in the user's $PATH.
     #[clap(
         id = "revive-dev-node.path",
         long = "revive-dev-node.path",

crates/core/src/differential_benchmarks/driver.rs

@@ -59,8 +59,8 @@ pub struct Driver<'a, I> {
     /// The definition of the test that the driver is instructed to execute.
     test_definition: &'a TestDefinition<'a>,
 
-    /// The private key allocator used by this driver and other drivers when account allocations are
-    /// needed.
+    /// The private key allocator used by this driver and other drivers when account allocations
+    /// are needed.
     private_key_allocator: Arc<Mutex<PrivateKeyAllocator>>,
 
     /// The execution state associated with the platform.

crates/core/src/differential_benchmarks/execution_state.rs

@@ -10,14 +10,15 @@ use revive_dt_format::metadata::{ContractIdent, ContractInstance};
 #[derive(Clone)]
 /// The state associated with the test execution of one of the workloads.
 pub struct ExecutionState {
-    /// The compiled contracts, these contracts have been compiled and have had the libraries linked
-    /// against them and therefore they're ready to be deployed on-demand.
+    /// The compiled contracts, these contracts have been compiled and have had the libraries
+    /// linked against them and therefore they're ready to be deployed on-demand.
     pub compiled_contracts: HashMap<PathBuf, HashMap<String, (String, JsonAbi)>>,
 
     /// A map of all of the deployed contracts and information about them.
     pub deployed_contracts: HashMap<ContractInstance, (ContractIdent, Address, JsonAbi)>,
 
-    /// This map stores the variables used for each one of the cases contained in the metadata file.
+    /// This map stores the variables used for each one of the cases contained in the metadata
+    /// file.
     pub variables: HashMap<String, U256>,
 }
 

crates/core/src/differential_tests/driver.rs

@@ -131,8 +131,8 @@ pub struct PlatformDriver<'a, I> {
     /// The definition of the test that the driver is instructed to execute.
     test_definition: &'a TestDefinition<'a>,
 
-    /// The private key allocator used by this driver and other drivers when account allocations are
-    /// needed.
+    /// The private key allocator used by this driver and other drivers when account allocations
+    /// are needed.
     private_key_allocator: Arc<Mutex<PrivateKeyAllocator>>,
 
     /// The execution state associated with the platform.
@@ -415,9 +415,13 @@ where
             let caller = {
                 let context = self.default_resolution_context();
                 let resolver = self.platform_information.node.resolver().await?;
-                step.caller
+                let resolved = step
+                    .caller
                     .resolve_address(resolver.as_ref(), context)
-                    .await?
+                    .await?;
+                self.platform_information
+                    .node
+                    .resolve_signer_or_default(resolved)
             };
             if let (_, _, Some(receipt)) = self
                 .get_or_deploy_contract_instance(&instance, caller, calldata, value)
@@ -445,7 +449,7 @@ where
                 .context("Failed to find deployment receipt for constructor call"),
             Method::Fallback | Method::FunctionName(_) => {
                 let resolver = self.platform_information.node.resolver().await?;
-                let tx = match step
+                let mut tx = match step
                     .as_transaction(resolver.as_ref(), self.default_resolution_context())
                     .await
                 {
@@ -455,6 +459,15 @@ where
                     }
                 };
 
+                // Resolve the signer to ensure we use an address that has keys
+                if let Some(from) = tx.from {
+                    tx.from = Some(
+                        self.platform_information
+                            .node
+                            .resolve_signer_or_default(from),
+                    );
+                }
+
                 self.platform_information.node.execute_transaction(tx).await
             }
         }
@@ -954,6 +967,10 @@ where
         }
 
         let tx = {
+            let deployer = self
+                .platform_information
+                .node
+                .resolve_signer_or_default(deployer);
             let tx = TransactionRequest::default().from(deployer);
             let tx = match value {
                 Some(ref value) => tx.value(value.into_inner()),

crates/core/src/differential_tests/entry_point.rs

@@ -7,10 +7,10 @@ use std::{
     time::{Duration, Instant},
 };
 
+use crate::Platform;
 use anyhow::Context as _;
 use futures::{FutureExt, StreamExt};
 use revive_dt_common::types::PrivateKeyAllocator;
-use revive_dt_core::Platform;
 use tokio::sync::{Mutex, RwLock, Semaphore};
 use tracing::{Instrument, error, info, info_span, instrument};
 

crates/core/src/differential_tests/execution_state.rs

@@ -10,14 +10,15 @@ use revive_dt_format::metadata::{ContractIdent, ContractInstance};
 #[derive(Clone)]
 /// The state associated with the test execution of one of the tests.
 pub struct ExecutionState {
-    /// The compiled contracts, these contracts have been compiled and have had the libraries linked
-    /// against them and therefore they're ready to be deployed on-demand.
+    /// The compiled contracts, these contracts have been compiled and have had the libraries
+    /// linked against them and therefore they're ready to be deployed on-demand.
     pub compiled_contracts: HashMap<PathBuf, HashMap<String, (String, JsonAbi)>>,
 
     /// A map of all of the deployed contracts and information about them.
     pub deployed_contracts: HashMap<ContractInstance, (ContractIdent, Address, JsonAbi)>,
 
-    /// This map stores the variables used for each one of the cases contained in the metadata file.
+    /// This map stores the variables used for each one of the cases contained in the metadata
+    /// file.
     pub variables: HashMap<String, U256>,
 }
 

crates/core/src/helpers/cached_compiler.rs

@@ -8,10 +8,10 @@ use std::{
     sync::{Arc, LazyLock},
 };
 
+use crate::Platform;
 use futures::FutureExt;
 use revive_dt_common::{iterators::FilesWithExtensionIterator, types::CompilerIdentifier};
 use revive_dt_compiler::{Compiler, CompilerOutput, Mode, SolidityCompiler};
-use revive_dt_core::Platform;
 use revive_dt_format::metadata::{ContractIdent, ContractInstance, Metadata};
 
 use alloy::{hex::ToHexExt, json_abi::JsonAbi, primitives::Address};

crates/core/src/helpers/pool.rs

@@ -2,9 +2,9 @@
 
 use std::sync::atomic::{AtomicUsize, Ordering};
 
+use crate::Platform;
 use anyhow::Context as _;
 use revive_dt_config::*;
-use revive_dt_core::Platform;
 use revive_dt_node_interaction::EthereumNode;
 
 /// The node pool starts one or more [Node] which then can be accessed

crates/core/src/helpers/test.rs

@@ -1,28 +1,22 @@
-use std::collections::BTreeMap;
-use std::sync::Arc;
-use std::{borrow::Cow, path::Path};
+use std::{borrow::Cow, collections::BTreeMap, path::Path, sync::Arc};
 
 use futures::{Stream, StreamExt, stream};
 use indexmap::{IndexMap, indexmap};
-use revive_dt_common::iterators::EitherIter;
-use revive_dt_common::types::PlatformIdentifier;
+use revive_dt_common::{iterators::EitherIter, types::PlatformIdentifier};
 use revive_dt_config::Context;
 use revive_dt_format::mode::ParsedMode;
 use serde_json::{Value, json};
 
-use revive_dt_compiler::Mode;
-use revive_dt_compiler::SolidityCompiler;
+use revive_dt_compiler::{Mode, SolidityCompiler};
 use revive_dt_format::{
     case::{Case, CaseIdx},
     metadata::MetadataFile,
 };
 use revive_dt_node_interaction::EthereumNode;
-use revive_dt_report::{ExecutionSpecificReporter, Reporter};
-use revive_dt_report::{TestSpecificReporter, TestSpecifier};
+use revive_dt_report::{ExecutionSpecificReporter, Reporter, TestSpecificReporter, TestSpecifier};
 use tracing::{debug, error, info};
 
-use crate::Platform;
-use crate::helpers::NodePool;
+use crate::{Platform, helpers::NodePool};
 
 pub async fn create_test_definitions_stream<'a>(
     // This is only required for creating the compiler objects and is not used anywhere else in the
@@ -69,8 +63,8 @@ pub async fn create_test_definitions_stream<'a>(
                     )
                 })
             })
-            // Inform the reporter of each one of the test cases that were discovered which we expect to
-            // run.
+            // Inform the reporter of each one of the test cases that were discovered which we
+            // expect to run.
             .inspect(|(_, _, _, _, reporter)| {
                 reporter
                     .report_test_case_discovery_event()

crates/core/src/lib.rs

@@ -3,6 +3,9 @@
 //! This crate defines the testing configuration and
 //! provides a helper utility to execute tests.
 
+pub mod differential_tests;
+pub mod helpers;
+
 use std::{
     pin::Pin,
     thread::{self, JoinHandle},
@@ -14,13 +17,17 @@ use revive_dt_common::types::*;
 use revive_dt_compiler::{SolidityCompiler, revive_resolc::Resolc, solc::Solc};
 use revive_dt_config::*;
 use revive_dt_node::{
-    Node, node_implementations::geth::GethNode,
-    node_implementations::lighthouse_geth::LighthouseGethNode,
-    node_implementations::substrate::SubstrateNode, node_implementations::zombienet::ZombieNode,
+    Node,
+    node_implementations::{
+        geth::GethNode, lighthouse_geth::LighthouseGethNode, substrate::SubstrateNode,
+        zombienet::ZombieNode,
+    },
 };
 use revive_dt_node_interaction::EthereumNode;
 use tracing::info;
 
+pub use helpers::CachedCompiler;
+
 /// A trait that describes the interface for the platforms that are supported by the tool.
 #[allow(clippy::type_complexity)]
 pub trait Platform {

crates/format/src/case.rs

@@ -45,8 +45,8 @@ pub struct Case {
     #[serde(skip_serializing_if = "Option::is_none")]
     pub expected: Option<Expected>,
 
-    /// An optional boolean which defines if the case as a whole should be ignored. If null then the
-    /// case will not be ignored.
+    /// An optional boolean which defines if the case as a whole should be ignored. If null then
+    /// the case will not be ignored.
     #[serde(skip_serializing_if = "Option::is_none")]
     pub ignore: Option<bool>,
 }

crates/format/src/metadata.rs

@@ -31,8 +31,8 @@ pub struct MetadataFile {
     /// The path of the metadata file. This will either be a JSON or solidity file.
     pub metadata_file_path: PathBuf,
 
-    /// This is the path contained within the corpus file. This could either be the path of some dir
-    /// or could be the actual metadata file path.
+    /// This is the path contained within the corpus file. This could either be the path of some
+    /// dir or could be the actual metadata file path.
     pub corpus_file_path: PathBuf,
 
     /// The metadata contained within the file.
@@ -69,13 +69,13 @@ impl Deref for MetadataFile {
 /// of steps and assertions that should be performed as part of the test case.
 #[derive(Debug, Default, Serialize, Deserialize, JsonSchema, Clone, Eq, PartialEq)]
 pub struct Metadata {
-    /// This is an optional comment on the metadata file which has no impact on the execution in any
-    /// way.
+    /// This is an optional comment on the metadata file which has no impact on the execution in
+    /// any way.
     #[serde(skip_serializing_if = "Option::is_none")]
     pub comment: Option<String>,
 
-    /// An optional boolean which defines if the metadata file as a whole should be ignored. If null
-    /// then the metadata file will not be ignored.
+    /// An optional boolean which defines if the metadata file as a whole should be ignored. If
+    /// null then the metadata file will not be ignored.
     #[serde(skip_serializing_if = "Option::is_none")]
     pub ignore: Option<bool>,
 
@@ -94,8 +94,8 @@ pub struct Metadata {
     /// This is a map where the key is the name of the contract instance and the value is the
     /// contract's path and ident in the file.
     ///
-    /// If any contract is to be used by the test then it must be included in here first so that the
-    /// framework is aware of its path, compiles it, and prepares it.
+    /// If any contract is to be used by the test then it must be included in here first so that
+    /// the framework is aware of its path, compiles it, and prepares it.
     #[serde(skip_serializing_if = "Option::is_none")]
     pub contracts: Option<BTreeMap<ContractInstance, ContractPathAndIdent>>,
 
@@ -123,8 +123,9 @@ pub struct Metadata {
     pub required_evm_version: Option<EvmVersionRequirement>,
 
     /// A set of compilation directives that will be passed to the compiler whenever the contracts
-    /// for the test are being compiled. Note that this differs from the [`Mode`]s in that a [`Mode`]
-    /// is just a filter for when a test can run whereas this is an instruction to the compiler.
+    /// for the test are being compiled. Note that this differs from the [`Mode`]s in that a
+    /// [`Mode`] is just a filter for when a test can run whereas this is an instruction to the
+    /// compiler.
     #[serde(skip_serializing_if = "Option::is_none")]
     pub compiler_directives: Option<CompilationDirectives>,
 }
@@ -326,7 +327,8 @@ define_wrapper_type!(
 )]
 #[serde(try_from = "String", into = "String")]
 pub struct ContractPathAndIdent {
-    /// The path of the contract source code relative to the directory containing the metadata file.
+    /// The path of the contract source code relative to the directory containing the metadata
+    /// file.
     pub contract_source_path: PathBuf,
 
     /// The identifier of the contract.

crates/format/src/mode.rs

@@ -1,13 +1,12 @@
 use anyhow::Context as _;
 use regex::Regex;
-use revive_dt_common::iterators::EitherIter;
-use revive_dt_common::types::{Mode, ModeOptimizerSetting, ModePipeline};
+use revive_dt_common::{
+    iterators::EitherIter,
+    types::{Mode, ModeOptimizerSetting, ModePipeline},
+};
 use schemars::JsonSchema;
 use serde::{Deserialize, Serialize};
-use std::collections::HashSet;
-use std::fmt::Display;
-use std::str::FromStr;
-use std::sync::LazyLock;
+use std::{collections::HashSet, fmt::Display, str::FromStr, sync::LazyLock};
 
 /// This represents a mode that has been parsed from test metadata.
 ///

crates/format/src/steps.rs

@@ -1,11 +1,10 @@
 use std::{collections::HashMap, fmt::Display, str::FromStr};
 
-use alloy::primitives::{FixedBytes, utils::parse_units};
 use alloy::{
     eips::BlockNumberOrTag,
     json_abi::Function,
     network::TransactionBuilder,
-    primitives::{Address, Bytes, U256},
+    primitives::{Address, Bytes, FixedBytes, U256, utils::parse_units},
     rpc::types::TransactionRequest,
 };
 use anyhow::Context as _;
@@ -17,8 +16,10 @@ use serde::{Deserialize, Serialize};
 use revive_dt_common::macros::define_wrapper_type;
 use tracing::{Instrument, info_span, instrument};
 
-use crate::traits::ResolverApi;
-use crate::{metadata::ContractInstance, traits::ResolutionContext};
+use crate::{
+    metadata::ContractInstance,
+    traits::{ResolutionContext, ResolverApi},
+};
 
 /// A test step.
 ///
@@ -147,8 +148,8 @@ pub struct FunctionCallStep {
     #[schemars(skip)]
     pub storage: Option<HashMap<String, Calldata>>,
 
-    /// Variable assignment to perform in the framework allowing us to reference them again later on
-    /// during the execution.
+    /// Variable assignment to perform in the framework allowing us to reference them again later
+    /// on during the execution.
     #[serde(skip_serializing_if = "Option::is_none")]
     pub variable_assignments: Option<VariableAssignments>,
 }
@@ -488,21 +489,20 @@ impl FunctionCallStep {
                     anyhow::bail!("ABI for instance '{}' not found", self.instance.as_ref());
                 };
 
-                // We follow the same logic that's implemented in the matter-labs-tester where they resolve
-                // the function name into a function selector and they assume that he function doesn't have
-                // any existing overloads.
+                // We follow the same logic that's implemented in the matter-labs-tester where they
+                // resolve the function name into a function selector and they assume that he
+                // function doesn't have any existing overloads.
                 // Overloads are handled by providing the full function signature in the "function
                 // name".
                 // https://github.com/matter-labs/era-compiler-tester/blob/1dfa7d07cba0734ca97e24704f12dd57f6990c2c/compiler_tester/src/test/case/input/mod.rs#L158-L190
-                let selector =
-                    if function_name.contains('(') && function_name.contains(')') {
-                        Function::parse(function_name)
+                let selector = if function_name.contains('(') && function_name.contains(')') {
+                    Function::parse(function_name)
                         .context(
                             "Failed to parse the provided function name into a function signature",
                         )?
                         .selector()
-                    } else {
-                        abi.functions()
+                } else {
+                    abi.functions()
                         .find(|function| function.signature().starts_with(function_name))
                         .ok_or_else(|| {
                             anyhow::anyhow!(
@@ -511,19 +511,21 @@ impl FunctionCallStep {
                                 &self.instance
                             )
                         })
-                        .with_context(|| format!(
-                            "Failed to resolve function selector for {:?} on instance {:?}",
-                            function_name, &self.instance
-                        ))?
+                        .with_context(|| {
+                            format!(
+                                "Failed to resolve function selector for {:?} on instance {:?}",
+                                function_name, &self.instance
+                            )
+                        })?
                         .selector()
-                    };
+                };
 
-                // Allocating a vector that we will be using for the calldata. The vector size will be:
-                // 4 bytes for the function selector.
+                // Allocating a vector that we will be using for the calldata. The vector size will
+                // be: 4 bytes for the function selector.
                 // function.inputs.len() * 32 bytes for the arguments (each argument is a U256).
                 //
-                // We're using indices in the following code in order to avoid the need for us to allocate
-                // a new buffer for each one of the resolved arguments.
+                // We're using indices in the following code in order to avoid the need for us to
+                // allocate a new buffer for each one of the resolved arguments.
                 let mut calldata = Vec::<u8>::with_capacity(4 + self.calldata.size_requirement());
                 calldata.extend(selector.0);
                 self.calldata
@@ -959,9 +961,12 @@ impl<'de> Deserialize<'de> for EtherValue {
 #[cfg(test)]
 mod tests {
 
-    use alloy::primitives::{BlockHash, BlockNumber, BlockTimestamp, ChainId, TxHash, address};
-    use alloy::sol_types::SolValue;
-    use alloy::{eips::BlockNumberOrTag, json_abi::JsonAbi};
+    use alloy::{
+        eips::BlockNumberOrTag,
+        json_abi::JsonAbi,
+        primitives::{BlockHash, BlockNumber, BlockTimestamp, ChainId, TxHash, address},
+        sol_types::SolValue,
+    };
     use std::{collections::HashMap, pin::Pin};
 
     use super::*;

crates/format/src/traits.rs

@@ -1,10 +1,10 @@
-use std::collections::HashMap;
-use std::pin::Pin;
+use std::{collections::HashMap, pin::Pin};
 
-use alloy::eips::BlockNumberOrTag;
-use alloy::json_abi::JsonAbi;
-use alloy::primitives::TxHash;
-use alloy::primitives::{Address, BlockHash, BlockNumber, BlockTimestamp, ChainId, U256};
+use alloy::{
+    eips::BlockNumberOrTag,
+    json_abi::JsonAbi,
+    primitives::{Address, BlockHash, BlockNumber, BlockTimestamp, ChainId, TxHash, U256},
+};
 use anyhow::Result;
 
 use crate::metadata::{ContractIdent, ContractInstance};

crates/ml-test-runner/Cargo.toml

@@ -0,0 +1,34 @@
+[package]
+name = "ml-test-runner"
+description = "ML-based test runner for executing differential tests file by file"
+version.workspace = true
+authors.workspace = true
+license.workspace = true
+edition.workspace = true
+repository.workspace = true
+rust-version.workspace = true
+
+[[bin]]
+name = "ml-test-runner"
+path = "src/main.rs"
+
+[dependencies]
+revive-dt-common = { workspace = true }
+revive-dt-compiler = { workspace = true }
+revive-dt-config = { workspace = true }
+revive-dt-core = { workspace = true }
+revive-dt-format = { workspace = true }
+revive-dt-node = { workspace = true }
+revive-dt-node-interaction = { workspace = true }
+revive-dt-report = { workspace = true }
+
+alloy = { workspace = true }
+anyhow = { workspace = true }
+clap = { workspace = true }
+tokio = { workspace = true }
+temp-dir = { workspace = true }
+tracing = { workspace = true }
+tracing-subscriber = { workspace = true }
+
+[lints]
+workspace = true

crates/ml-test-runner/README.md

@@ -0,0 +1,74 @@
+# ML Test Runner
+
+A test runner for executing Revive differential tests file-by-file with cargo-test-style output.
+
+This is similar to the `retester` binary but designed for ML-based test execution with a focus on:
+- Running tests file-by-file (rather than in bulk)
+- Caching passed tests to skip them in future runs
+- Providing cargo-test-style output for easy integration with ML pipelines
+- Single platform testing (rather than differential testing)
+
+## Features
+
+- **File-by-file execution**: Run tests on individual `.sol` files, corpus files (`.json`), or recursively walk directories
+- **Cached results**: Skip tests that have already passed using `--cached-passed`
+- **Fail fast**: Stop on first failure with `--bail`
+- **Cargo-like output**: Familiar test output format with colored pass/fail indicators
+- **Platform support**: Test against `geth` or `kitchensink` platforms
+
+## Usage
+
+```bash
+# Run a single .sol file (compile-only mode, default)
+./ml-test-runner path/to/test.sol --platform geth
+
+# Run all tests in a corpus file
+./ml-test-runner path/to/corpus.json --platform kitchensink
+
+# Walk a directory recursively for .sol files
+./ml-test-runner path/to/tests/ --platform geth
+
+# Use cached results and bail on first failure
+./ml-test-runner path/to/tests/ --cached-passed ./cache.txt --bail
+
+# Start the platform and execute tests (full mode)
+./ml-test-runner path/to/tests/ --platform geth --start-platform
+
+# Enable verbose logging (info, debug, or trace level)
+RUST_LOG=info ./ml-test-runner path/to/tests/
+RUST_LOG=debug ./ml-test-runner path/to/tests/ --start-platform
+RUST_LOG=trace ./ml-test-runner path/to/tests/ --start-platform
+```
+
+## Arguments
+
+- `<PATH>` - Path to test file (`.sol`), corpus file (`.json`), or folder of `.sol` files
+- `--cached-passed <FILE>` - File to track tests that have already passed
+- `--bail` - Stop after the first file failure
+- `--platform <PLATFORM>` - Platform to test against (`geth`, `kitchensink`, or `zombienet`, default: `geth`)
+- `--start-platform` - Start the platform and execute tests (default: `false`, compile-only mode)
+
+## Output Format
+
+The runner produces cargo-test-style output:
+
+```
+test path/to/test1.sol ... ok
+test path/to/test2.sol ... FAILED
+test path/to/test3.sol ... cached
+
+failures:
+
+---- path/to/test2.sol ----
+Error: ...
+
+test result: FAILED. 1 passed; 1 failed; 1 cached; finished in 2.34s
+```
+
+## Building
+
+```bash
+cargo build --release -p ml-test-runner
+```
+
+The binary will be available at `target/release/ml-test-runner`.

crates/ml-test-runner/src/main.rs

@@ -0,0 +1,541 @@
+use anyhow::Context;
+use clap::Parser;
+use revive_dt_common::{
+    iterators::FilesWithExtensionIterator,
+    types::{PlatformIdentifier, PrivateKeyAllocator},
+};
+use revive_dt_config::TestExecutionContext;
+use revive_dt_core::{
+    CachedCompiler, Platform,
+    helpers::{TestDefinition, TestPlatformInformation},
+};
+use revive_dt_format::{
+    case::CaseIdx,
+    corpus::Corpus,
+    metadata::{Metadata, MetadataFile},
+};
+use std::{
+    borrow::Cow,
+    collections::{BTreeMap, HashSet},
+    fs::File,
+    io::{BufRead, BufReader, BufWriter, Write},
+    path::{Path, PathBuf},
+    sync::Arc,
+    time::Instant,
+};
+use temp_dir::TempDir;
+use tokio::sync::Mutex;
+use tracing::info;
+use tracing_subscriber::{EnvFilter, FmtSubscriber};
+
+/// ML-based test runner for executing differential tests file by file
+#[derive(Debug, Parser)]
+#[command(name = "ml-test-runner")]
+struct MlTestRunnerArgs {
+    /// Path to test file (.sol), corpus file (.json), or folder containing .sol files
+    #[arg(value_name = "PATH")]
+    path: PathBuf,
+
+    /// File to cache tests that have already passed
+    #[arg(long = "cached-passed")]
+    cached_passed: Option<PathBuf>,
+
+    /// Stop after the first file failure
+    #[arg(long = "bail")]
+    bail: bool,
+
+    /// Platform to test against (e.g., geth-evm-solc, kitchensink-polkavm-resolc)
+    #[arg(long = "platform", default_value = "geth-evm-solc")]
+    platform: PlatformIdentifier,
+
+    /// Start the platform and wait for RPC readiness
+    #[arg(long = "start-platform", default_value = "false")]
+    start_platform: bool,
+
+    /// Private key to use for wallet initialization (hex string with or without 0x prefix)
+    #[arg(
+        long = "private-key",
+        default_value = "0x5fb92d6e98884f76de468fa3f6278f8807c48bebc13595d45af5bdc4da702133"
+    )]
+    private_key: String,
+
+    /// RPC port to connect to when using existing node
+    #[arg(long = "rpc-port", default_value = "8545")]
+    rpc_port: u16,
+}
+
+fn main() -> anyhow::Result<()> {
+    let subscriber = FmtSubscriber::builder()
+        .with_env_filter(EnvFilter::from_default_env())
+        .with_writer(std::io::stderr)
+        .finish();
+    tracing::subscriber::set_global_default(subscriber).expect("Failed to set tracing subscriber");
+
+    let args = MlTestRunnerArgs::parse();
+
+    info!("ML test runner starting");
+    info!("Platform: {:?}", args.platform);
+    info!("Start platform: {}", args.start_platform);
+
+    tokio::runtime::Builder::new_multi_thread()
+        .enable_all()
+        .build()
+        .expect("Failed building the Runtime")
+        .block_on(run(args))
+}
+
+async fn run(args: MlTestRunnerArgs) -> anyhow::Result<()> {
+    let start_time = Instant::now();
+
+    info!("Discovering test files from: {}", args.path.display());
+    let test_files = discover_test_files(&args.path)?;
+    info!("Found {} test file(s)", test_files.len());
+
+    let cached_passed = if let Some(cache_file) = &args.cached_passed {
+        let cached = load_cached_passed(cache_file)?;
+        info!("Loaded {} cached passed test(s)", cached.len());
+        cached
+    } else {
+        HashSet::new()
+    };
+
+    let cached_passed = Arc::new(Mutex::new(cached_passed));
+
+    let mut passed_files = 0;
+    let mut failed_files = 0;
+    let mut skipped_files = 0;
+    let mut failures = Vec::new();
+
+    const GREEN: &str = "\x1B[32m";
+    const RED: &str = "\x1B[31m";
+    const YELLOW: &str = "\x1B[33m";
+    const COLOUR_RESET: &str = "\x1B[0m";
+    const BOLD: &str = "\x1B[1m";
+    const BOLD_RESET: &str = "\x1B[22m";
+
+    for test_file in test_files {
+        let file_display = test_file.display().to_string();
+
+        // Check if already passed
+        {
+            let cache = cached_passed.lock().await;
+            if cache.contains(&file_display) {
+                println!("test {} ... {YELLOW}cached{COLOUR_RESET}", file_display);
+                skipped_files += 1;
+                continue;
+            }
+        }
+
+        info!("Loading metadata from: {}", test_file.display());
+        let metadata_file = match load_metadata_file(&test_file) {
+            Ok(mf) => {
+                info!("Loaded metadata with {} case(s)", mf.cases.len());
+                mf
+            }
+            Err(e) => {
+                println!("test {} ... {RED}FAILED{COLOUR_RESET}", file_display);
+                println!("    Error loading metadata: {}", e);
+                failed_files += 1;
+                failures.push((
+                    file_display.clone(),
+                    format!("Error loading metadata: {}", e),
+                ));
+                if args.bail {
+                    break;
+                }
+                continue;
+            }
+        };
+
+        info!("Executing test file: {}", file_display);
+        match execute_test_file(&args, &metadata_file).await {
+            Ok(_) => {
+                println!("test {} ... {GREEN}ok{COLOUR_RESET}", file_display);
+                info!("Test file passed: {}", file_display);
+                passed_files += 1;
+
+                {
+                    let mut cache = cached_passed.lock().await;
+                    cache.insert(file_display);
+                }
+            }
+            Err(e) => {
+                println!("test {} ... {RED}FAILED{COLOUR_RESET}", file_display);
+                failed_files += 1;
+                failures.push((file_display, format!("{:?}", e)));
+
+                if args.bail {
+                    info!("Bailing after first failure");
+                    break;
+                }
+            }
+        }
+    }
+
+    if let Some(cache_file) = &args.cached_passed {
+        let cache = cached_passed.lock().await;
+        info!("Saving {} cached passed test(s)", cache.len());
+        save_cached_passed(cache_file, &cache)?;
+    }
+
+    // Print summary
+    println!();
+    if !failures.is_empty() {
+        println!("{BOLD}failures:{BOLD_RESET}");
+        println!();
+        for (file, error) in &failures {
+            println!("---- {} ----", file);
+            println!("{}", error);
+            println!();
+        }
+    }
+
+    let elapsed = start_time.elapsed();
+    println!(
+        "test result: {}. {} passed; {} failed; {} cached; finished in {:.2}s",
+        if failed_files == 0 {
+            format!("{GREEN}ok{COLOUR_RESET}")
+        } else {
+            format!("{RED}FAILED{COLOUR_RESET}")
+        },
+        passed_files,
+        failed_files,
+        skipped_files,
+        elapsed.as_secs_f64()
+    );
+
+    if failed_files > 0 {
+        std::process::exit(1);
+    }
+
+    Ok(())
+}
+
+/// Discover test files from the given path
+fn discover_test_files(path: &Path) -> anyhow::Result<Vec<PathBuf>> {
+    if !path.exists() {
+        anyhow::bail!("Path does not exist: {}", path.display());
+    }
+
+    let mut files = Vec::new();
+
+    if path.is_file() {
+        let extension = path.extension().and_then(|s| s.to_str()).unwrap_or("");
+
+        match extension {
+            "sol" => {
+                // Single .sol file
+                files.push(path.to_path_buf());
+            }
+            "json" => {
+                // Corpus file - enumerate its tests
+                let corpus = Corpus::try_from_path(path)?;
+                let metadata_files = corpus.enumerate_tests();
+                for metadata in metadata_files {
+                    files.push(metadata.metadata_file_path);
+                }
+            }
+            _ => anyhow::bail!(
+                "Unsupported file extension: {}. Expected .sol or .json",
+                extension
+            ),
+        }
+    } else if path.is_dir() {
+        // Walk directory recursively for .sol files
+        for entry in FilesWithExtensionIterator::new(path)
+            .with_allowed_extension("sol")
+            .with_use_cached_fs(true)
+        {
+            files.push(entry);
+        }
+    } else {
+        anyhow::bail!("Path is neither a file nor a directory: {}", path.display());
+    }
+
+    Ok(files)
+}
+
+/// Load metadata from a test file
+fn load_metadata_file(path: &Path) -> anyhow::Result<MetadataFile> {
+    let metadata = Metadata::try_from_file(path)
+        .ok_or_else(|| anyhow::anyhow!("Failed to load metadata from {}", path.display()))?;
+
+    Ok(MetadataFile {
+        metadata_file_path: path.to_path_buf(),
+        corpus_file_path: path.to_path_buf(),
+        content: metadata,
+    })
+}
+
+/// Execute all test cases in a metadata file
+async fn execute_test_file(
+    args: &MlTestRunnerArgs,
+    metadata_file: &MetadataFile,
+) -> anyhow::Result<()> {
+    if metadata_file.cases.is_empty() {
+        anyhow::bail!("No test cases found in file");
+    }
+
+    info!("Processing {} test case(s)", metadata_file.cases.len());
+
+    // Get the platform based on CLI args
+    let platform: &dyn Platform = match args.platform {
+        PlatformIdentifier::GethEvmSolc => &revive_dt_core::GethEvmSolcPlatform,
+        PlatformIdentifier::LighthouseGethEvmSolc => &revive_dt_core::LighthouseGethEvmSolcPlatform,
+        PlatformIdentifier::KitchensinkPolkavmResolc => {
+            &revive_dt_core::KitchensinkPolkavmResolcPlatform
+        }
+        PlatformIdentifier::KitchensinkRevmSolc => &revive_dt_core::KitchensinkRevmSolcPlatform,
+        PlatformIdentifier::ReviveDevNodePolkavmResolc => {
+            &revive_dt_core::ReviveDevNodePolkavmResolcPlatform
+        }
+        PlatformIdentifier::ReviveDevNodeRevmSolc => &revive_dt_core::ReviveDevNodeRevmSolcPlatform,
+        PlatformIdentifier::ZombienetPolkavmResolc => {
+            &revive_dt_core::ZombienetPolkavmResolcPlatform
+        }
+        PlatformIdentifier::ZombienetRevmSolc => &revive_dt_core::ZombienetRevmSolcPlatform,
+    };
+
+    let temp_dir = TempDir::new()?;
+    info!("Created temporary directory: {}", temp_dir.path().display());
+
+    let test_context = TestExecutionContext::default();
+    let context = revive_dt_config::Context::Test(Box::new(test_context));
+
+    let node: &'static dyn revive_dt_node_interaction::EthereumNode = if args.start_platform {
+        info!("Starting blockchain node...");
+        let node_handle = platform
+            .new_node(context.clone())
+            .context("Failed to spawn node thread")?;
+
+        info!("Waiting for node to start...");
+        let node = node_handle
+            .join()
+            .map_err(|e| anyhow::anyhow!("Node thread panicked: {:?}", e))?
+            .context("Failed to start node")?;
+
+        info!(
+            "Node started with ID: {}, connection: {}",
+            node.id(),
+            node.connection_string()
+        );
+        let node = Box::leak(node);
+
+        info!("Running pre-transactions...");
+        node.pre_transactions()
+            .await
+            .context("Failed to run pre-transactions")?;
+        info!("Pre-transactions completed");
+
+        node
+    } else {
+        info!("Using existing node");
+        let existing_node: Box<dyn revive_dt_node_interaction::EthereumNode> = match args.platform {
+            PlatformIdentifier::GethEvmSolc | PlatformIdentifier::LighthouseGethEvmSolc => {
+                Box::new(
+                    revive_dt_node::node_implementations::geth::GethNode::new_existing(
+                        &args.private_key,
+                        args.rpc_port,
+                    )
+                    .await?,
+                )
+            }
+            PlatformIdentifier::KitchensinkPolkavmResolc
+            | PlatformIdentifier::KitchensinkRevmSolc
+            | PlatformIdentifier::ReviveDevNodePolkavmResolc
+            | PlatformIdentifier::ReviveDevNodeRevmSolc
+            | PlatformIdentifier::ZombienetPolkavmResolc
+            | PlatformIdentifier::ZombienetRevmSolc => Box::new(
+                revive_dt_node::node_implementations::substrate::SubstrateNode::new_existing(
+                    &args.private_key,
+                    args.rpc_port,
+                )
+                .await?,
+            ),
+        };
+        Box::leak(existing_node)
+    };
+
+    info!("Initializing cached compiler");
+    let cached_compiler = CachedCompiler::new(temp_dir.path().join("compilation_cache"), false)
+        .await
+        .map(Arc::new)
+        .context("Failed to create cached compiler")?;
+
+    let private_key_allocator = Arc::new(Mutex::new(PrivateKeyAllocator::new(
+        alloy::primitives::U256::from(100),
+    )));
+
+    let (reporter, report_task) =
+        revive_dt_report::ReportAggregator::new(context.clone()).into_task();
+
+    tokio::spawn(report_task);
+
+    info!(
+        "Building test definitions for {} case(s)",
+        metadata_file.cases.len()
+    );
+    let mut test_definitions = Vec::new();
+    for (case_idx, case) in metadata_file.cases.iter().enumerate() {
+        info!("Building test definition for case {}", case_idx);
+        let test_def = build_test_definition(
+            metadata_file,
+            case,
+            case_idx,
+            platform,
+            node,
+            &context,
+            &reporter,
+        )
+        .await?;
+
+        if let Some(test_def) = test_def {
+            info!("Test definition for case {} created successfully", case_idx);
+            test_definitions.push(test_def);
+        }
+    }
+
+    info!("Executing {} test definition(s)", test_definitions.len());
+    for (idx, test_definition) in test_definitions.iter().enumerate() {
+        info!("─────────────────────────────────────────────────────────────────");
+        info!(
+            "Executing case {}/{}: case_idx={}, mode={}, steps={}",
+            idx + 1,
+            test_definitions.len(),
+            test_definition.case_idx,
+            test_definition.mode,
+            test_definition.case.steps.len()
+        );
+
+        info!("Creating driver for case {}", test_definition.case_idx);
+        let driver = revive_dt_core::differential_tests::Driver::new_root(
+            test_definition,
+            private_key_allocator.clone(),
+            &cached_compiler,
+        )
+        .await
+        .context("Failed to create driver")?;
+
+        info!(
+            "Running {} step(s) for case {}",
+            test_definition.case.steps.len(),
+            test_definition.case_idx
+        );
+        let steps_executed = driver.execute_all().await.context(format!(
+            "Failed to execute case {}",
+            test_definition.case_idx
+        ))?;
+        info!(
+            "✓ Case {} completed successfully, executed {} step(s)",
+            test_definition.case_idx, steps_executed
+        );
+    }
+    info!("─────────────────────────────────────────────────────────────────");
+    info!(
+        "All {} test case(s) executed successfully",
+        test_definitions.len()
+    );
+
+    Ok(())
+}
+
+/// Build a test definition for a single test case
+async fn build_test_definition<'a>(
+    metadata_file: &'a MetadataFile,
+    case: &'a revive_dt_format::case::Case,
+    case_idx: usize,
+    platform: &'a dyn Platform,
+    node: &'a dyn revive_dt_node_interaction::EthereumNode,
+    context: &revive_dt_config::Context,
+    reporter: &revive_dt_report::Reporter,
+) -> anyhow::Result<Option<TestDefinition<'a>>> {
+    let mode = case
+        .modes
+        .as_ref()
+        .or(metadata_file.modes.as_ref())
+        .and_then(|modes| modes.first())
+        .and_then(|parsed_mode| parsed_mode.to_modes().next())
+        .map(Cow::Owned)
+        .or_else(|| revive_dt_compiler::Mode::all().next().map(Cow::Borrowed))
+        .unwrap();
+
+    let compiler = platform
+        .new_compiler(context.clone(), mode.version.clone().map(Into::into))
+        .await
+        .context("Failed to create compiler")?;
+
+    let test_reporter =
+        reporter.test_specific_reporter(Arc::new(revive_dt_report::TestSpecifier {
+            solc_mode: mode.as_ref().clone(),
+            metadata_file_path: metadata_file.metadata_file_path.clone(),
+            case_idx: CaseIdx::new(case_idx),
+        }));
+
+    let execution_reporter =
+        test_reporter.execution_specific_reporter(node.id(), platform.platform_identifier());
+
+    let mut platforms = BTreeMap::new();
+    platforms.insert(
+        platform.platform_identifier(),
+        TestPlatformInformation {
+            platform,
+            node,
+            compiler,
+            reporter: execution_reporter,
+        },
+    );
+
+    let test_definition = TestDefinition {
+        metadata: metadata_file,
+        metadata_file_path: &metadata_file.metadata_file_path,
+        mode,
+        case_idx: CaseIdx::new(case_idx),
+        case,
+        platforms,
+        reporter: test_reporter,
+    };
+
+    if let Err((reason, _)) = test_definition.check_compatibility() {
+        println!("    Skipping case {}: {}", case_idx, reason);
+        return Ok(None);
+    }
+
+    Ok(Some(test_definition))
+}
+
+/// Load cached passed tests from file
+fn load_cached_passed(path: &Path) -> anyhow::Result<HashSet<String>> {
+    if !path.exists() {
+        return Ok(HashSet::new());
+    }
+
+    let file = File::open(path).context("Failed to open cached-passed file")?;
+    let reader = BufReader::new(file);
+
+    let mut cache = HashSet::new();
+    for line in reader.lines() {
+        let line = line?;
+        let trimmed = line.trim();
+        if !trimmed.is_empty() {
+            cache.insert(trimmed.to_string());
+        }
+    }
+
+    Ok(cache)
+}
+
+/// Save cached passed tests to file
+fn save_cached_passed(path: &Path, cache: &HashSet<String>) -> anyhow::Result<()> {
+    let file = File::create(path).context("Failed to create cached-passed file")?;
+    let mut writer = BufWriter::new(file);
+
+    let mut entries: Vec<_> = cache.iter().collect();
+    entries.sort();
+
+    for entry in entries {
+        writeln!(writer, "{}", entry)?;
+    }
+
+    writer.flush()?;
+    Ok(())
+}

crates/node-interaction/src/lib.rs

@@ -1,11 +1,14 @@
 //! This crate implements all node interactions.
 
-use std::pin::Pin;
-use std::sync::Arc;
+use std::{pin::Pin, sync::Arc};
 
-use alloy::primitives::{Address, BlockNumber, BlockTimestamp, StorageKey, TxHash, U256};
-use alloy::rpc::types::trace::geth::{DiffMode, GethDebugTracingOptions, GethTrace};
-use alloy::rpc::types::{EIP1186AccountProofResponse, TransactionReceipt, TransactionRequest};
+use alloy::{
+    primitives::{Address, BlockNumber, BlockTimestamp, StorageKey, TxHash, U256},
+    rpc::types::{
+        EIP1186AccountProofResponse, TransactionReceipt, TransactionRequest,
+        trace::geth::{DiffMode, GethDebugTracingOptions, GethTrace},
+    },
+};
 use anyhow::Result;
 
 use futures::Stream;
@@ -74,6 +77,10 @@ pub trait EthereumNode {
                 + '_,
         >,
     >;
+
+    /// Checks if the provided address is in the wallet. If it is, returns the address.
+    /// Otherwise, returns the default signer's address.
+    fn resolve_signer_or_default(&self, address: Address) -> Address;
 }
 
 #[derive(Clone, Debug, PartialEq, Eq, PartialOrd, Ord, Hash)]

crates/node/src/node_implementations/geth.rs

@@ -18,7 +18,9 @@ use alloy::{
     eips::BlockNumberOrTag,
     genesis::{Genesis, GenesisAccount},
     network::{Ethereum, EthereumWallet, NetworkWallet},
-    primitives::{Address, BlockHash, BlockNumber, BlockTimestamp, StorageKey, TxHash, U256},
+    primitives::{
+        Address, BlockHash, BlockNumber, BlockTimestamp, ChainId, StorageKey, TxHash, U256,
+    },
     providers::{
         Provider,
         ext::DebugApi,
@@ -75,6 +77,7 @@ pub struct GethNode {
     wallet: Arc<EthereumWallet>,
     nonce_manager: CachedNonceManager,
     provider: OnceCell<ConcreteProvider<Ethereum, Arc<EthereumWallet>>>,
+    chain_id: ChainId,
 }
 
 impl GethNode {
@@ -125,9 +128,120 @@ impl GethNode {
             wallet: wallet.clone(),
             nonce_manager: Default::default(),
             provider: Default::default(),
+            chain_id: CHAIN_ID,
         }
     }
 
+    pub async fn new_existing(private_key: &str, rpc_port: u16) -> anyhow::Result<Self> {
+        use alloy::{
+            primitives::FixedBytes,
+            providers::{Provider, ProviderBuilder},
+            signers::local::PrivateKeySigner,
+        };
+
+        let key_str = private_key
+            .trim()
+            .strip_prefix("0x")
+            .unwrap_or(private_key.trim());
+        let key_bytes = alloy::hex::decode(key_str)
+            .map_err(|e| anyhow::anyhow!("Failed to decode private key hex: {}", e))?;
+
+        if key_bytes.len() != 32 {
+            anyhow::bail!(
+                "Private key must be 32 bytes (64 hex characters), got {}",
+                key_bytes.len()
+            );
+        }
+
+        let mut bytes = [0u8; 32];
+        bytes.copy_from_slice(&key_bytes);
+
+        let signer = PrivateKeySigner::from_bytes(&FixedBytes(bytes))
+            .map_err(|e| anyhow::anyhow!("Failed to create signer from private key: {}", e))?;
+
+        let address = signer.address();
+        let wallet = Arc::new(EthereumWallet::new(signer));
+        let connection_string = format!("http://localhost:{}", rpc_port);
+
+        let chain_id = ProviderBuilder::new()
+            .connect_http(connection_string.parse()?)
+            .get_chain_id()
+            .await
+            .context("Failed to query chain ID from RPC")?;
+
+        let node = Self {
+            connection_string: format!("http://localhost:{}", rpc_port),
+            base_directory: PathBuf::new(),
+            data_directory: PathBuf::new(),
+            logs_directory: PathBuf::new(),
+            geth: PathBuf::new(),
+            id: 0,
+            chain_id,
+            handle: None,
+            start_timeout: Duration::from_secs(0),
+            wallet,
+            nonce_manager: Default::default(),
+            provider: Default::default(),
+        };
+
+        // Check balance and fund if needed
+        node.ensure_funded(address).await?;
+
+        Ok(node)
+    }
+
+    /// Ensure that the given address has at least 1000 ETH, funding it from the node's managed
+    /// account if necessary.
+    async fn ensure_funded(&self, address: Address) -> anyhow::Result<()> {
+        use alloy::{
+            primitives::utils::{format_ether, parse_ether},
+            providers::{Provider, ProviderBuilder},
+        };
+
+        let provider = ProviderBuilder::new().connect_http(self.connection_string.parse()?);
+        let balance = provider.get_balance(address).await?;
+        let min_balance = parse_ether("1000")?;
+
+        if balance >= min_balance {
+            tracing::info!(
+                "Wallet {} already has sufficient balance: {} ETH",
+                address,
+                format_ether(balance)
+            );
+            return Ok(());
+        }
+
+        tracing::info!(
+            "Funding wallet {} (current: {} ETH, target: 1000 ETH)",
+            address,
+            format_ether(balance)
+        );
+
+        // Get the node's managed account
+        let accounts = provider.get_accounts().await?;
+        if accounts.is_empty() {
+            anyhow::bail!("No managed accounts available on the node to fund wallet");
+        }
+
+        let from_account = accounts[0];
+
+        let funding_amount = min_balance - balance;
+        let tx = TransactionRequest::default()
+            .from(from_account)
+            .to(address)
+            .value(funding_amount);
+
+        provider
+            .send_transaction(tx)
+            .await?
+            .get_receipt()
+            .await
+            .context("Failed to get receipt for funding transaction")?;
+
+        tracing::info!("Successfully funded wallet {}", address);
+        Ok(())
+    }
+
     /// Create the node directory and call `geth init` to configure the genesis.
     #[instrument(level = "info", skip_all, fields(geth_node_id = self.id))]
     fn init(&mut self, mut genesis: Genesis) -> anyhow::Result<&mut Self> {
@@ -255,7 +369,7 @@ impl GethNode {
                 construct_concurrency_limited_provider::<Ethereum, _>(
                     self.connection_string.as_str(),
                     FallbackGasFiller::default(),
-                    ChainIdFiller::new(Some(CHAIN_ID)),
+                    ChainIdFiller::new(Some(self.chain_id)),
                     NonceFiller::new(self.nonce_manager.clone()),
                     self.wallet.clone(),
                 )
@@ -349,23 +463,25 @@ impl EthereumNode for GethNode {
                 .context("Failed to submit transaction to geth node")?;
             let transaction_hash = *pending_transaction.tx_hash();
 
-            // The following is a fix for the "transaction indexing is in progress" error that we used
-            // to get. You can find more information on this in the following GH issue in geth
-            // https://github.com/ethereum/go-ethereum/issues/28877. To summarize what's going on,
+            // The following is a fix for the "transaction indexing is in progress" error that we
+            // used to get. You can find more information on this in the following GH issue in
+            // geth https://github.com/ethereum/go-ethereum/issues/28877. To summarize what's going on,
             // before we can get the receipt of the transaction it needs to have been indexed by the
-            // node's indexer. Just because the transaction has been confirmed it doesn't mean that it
-            // has been indexed. When we call alloy's `get_receipt` it checks if the transaction was
-            // confirmed. If it has been, then it will call `eth_getTransactionReceipt` method which
-            // _might_ return the above error if the tx has not yet been indexed yet. So, we need to
-            // implement a retry mechanism for the receipt to keep retrying to get it until it
-            // eventually works, but we only do that if the error we get back is the "transaction
-            // indexing is in progress" error or if the receipt is None.
+            // node's indexer. Just because the transaction has been confirmed it doesn't mean that
+            // it has been indexed. When we call alloy's `get_receipt` it checks if the
+            // transaction was confirmed. If it has been, then it will call
+            // `eth_getTransactionReceipt` method which _might_ return the above error if the tx
+            // has not yet been indexed yet. So, we need to implement a retry mechanism for the
+            // receipt to keep retrying to get it until it eventually works, but we only do that
+            // if the error we get back is the "transaction indexing is in progress" error or if
+            // the receipt is None.
             //
-            // Getting the transaction indexed and taking a receipt can take a long time especially when
-            // a lot of transactions are being submitted to the node. Thus, while initially we only
-            // allowed for 60 seconds of waiting with a 1 second delay in polling, we need to allow for
-            // a larger wait time. Therefore, in here we allow for 5 minutes of waiting with exponential
-            // backoff each time we attempt to get the receipt and find that it's not available.
+            // Getting the transaction indexed and taking a receipt can take a long time especially
+            // when a lot of transactions are being submitted to the node. Thus, while initially
+            // we only allowed for 60 seconds of waiting with a 1 second delay in polling, we
+            // need to allow for a larger wait time. Therefore, in here we allow for 5 minutes of
+            // waiting with exponential backoff each time we attempt to get the receipt and find
+            // that it's not available.
             poll(
                 Self::RECEIPT_POLLING_DURATION,
                 PollingWaitBehavior::Constant(Duration::from_millis(200)),
@@ -542,6 +658,16 @@ impl EthereumNode for GethNode {
                 as Pin<Box<dyn Stream<Item = MinedBlockInformation>>>)
         })
     }
+
+    fn resolve_signer_or_default(&self, address: Address) -> Address {
+        let signer_addresses: Vec<_> =
+            <EthereumWallet as NetworkWallet<Ethereum>>::signer_addresses(&self.wallet).collect();
+        if signer_addresses.contains(&address) {
+            address
+        } else {
+            self.wallet.default_signer().address()
+        }
+    }
 }
 
 pub struct GethNodeResolver {

crates/node/src/node_implementations/lighthouse_geth.rs

@@ -761,6 +761,16 @@ impl EthereumNode for LighthouseGethNode {
                 as Pin<Box<dyn Stream<Item = MinedBlockInformation>>>)
         })
     }
+
+    fn resolve_signer_or_default(&self, address: Address) -> Address {
+        let signer_addresses: Vec<_> =
+            <EthereumWallet as NetworkWallet<Ethereum>>::signer_addresses(&self.wallet).collect();
+        if signer_addresses.contains(&address) {
+            address
+        } else {
+            self.wallet.default_signer().address()
+        }
+    }
 }
 
 pub struct LighthouseGethNodeResolver<F: TxFiller<Ethereum>, P: Provider<Ethereum>> {

crates/node/src/node_implementations/substrate.rs

@@ -80,6 +80,7 @@ pub struct SubstrateNode {
     wallet: Arc<EthereumWallet>,
     nonce_manager: CachedNonceManager,
     provider: OnceCell<ConcreteProvider<ReviveNetwork, Arc<EthereumWallet>>>,
+    chain_id: alloy::primitives::ChainId,
 }
 
 impl SubstrateNode {
@@ -131,9 +132,64 @@ impl SubstrateNode {
             wallet: wallet.clone(),
             nonce_manager: Default::default(),
             provider: Default::default(),
+            chain_id: CHAIN_ID,
         }
     }
 
+    pub async fn new_existing(private_key: &str, rpc_port: u16) -> anyhow::Result<Self> {
+        use alloy::{
+            primitives::FixedBytes,
+            providers::{Provider, ProviderBuilder},
+            signers::local::PrivateKeySigner,
+        };
+
+        let key_str = private_key
+            .trim()
+            .strip_prefix("0x")
+            .unwrap_or(private_key.trim());
+        let key_bytes = alloy::hex::decode(key_str)
+            .map_err(|e| anyhow::anyhow!("Failed to decode private key hex: {}", e))?;
+
+        if key_bytes.len() != 32 {
+            anyhow::bail!(
+                "Private key must be 32 bytes (64 hex characters), got {}",
+                key_bytes.len()
+            );
+        }
+
+        let mut bytes = [0u8; 32];
+        bytes.copy_from_slice(&key_bytes);
+
+        let signer = PrivateKeySigner::from_bytes(&FixedBytes(bytes))
+            .map_err(|e| anyhow::anyhow!("Failed to create signer from private key: {}", e))?;
+
+        let wallet = Arc::new(EthereumWallet::new(signer));
+        let rpc_url = format!("http://localhost:{}", rpc_port);
+
+        // Query the chain ID from the RPC
+        let chain_id = ProviderBuilder::new()
+            .connect_http(rpc_url.parse()?)
+            .get_chain_id()
+            .await
+            .context("Failed to query chain ID from RPC")?;
+
+        Ok(Self {
+            id: 0,
+            node_binary: PathBuf::new(),
+            eth_proxy_binary: PathBuf::new(),
+            export_chainspec_command: String::new(),
+            rpc_url,
+            base_directory: PathBuf::new(),
+            logs_directory: PathBuf::new(),
+            substrate_process: None,
+            eth_proxy_process: None,
+            wallet,
+            nonce_manager: Default::default(),
+            provider: Default::default(),
+            chain_id,
+        })
+    }
+
     fn init(&mut self, mut genesis: Genesis) -> anyhow::Result<&mut Self> {
         let _ = remove_dir_all(self.base_directory.as_path());
         let _ = clear_directory(&self.base_directory);
@@ -350,7 +406,7 @@ impl SubstrateNode {
                 construct_concurrency_limited_provider::<ReviveNetwork, _>(
                     self.rpc_url.as_str(),
                     FallbackGasFiller::new(u64::MAX, 5_000_000_000, 1_000_000_000),
-                    ChainIdFiller::new(Some(CHAIN_ID)),
+                    ChainIdFiller::new(Some(self.chain_id)),
                     NonceFiller::new(self.nonce_manager.clone()),
                     self.wallet.clone(),
                 )
@@ -541,6 +597,16 @@ impl EthereumNode for SubstrateNode {
                 as Pin<Box<dyn Stream<Item = MinedBlockInformation>>>)
         })
     }
+
+    fn resolve_signer_or_default(&self, address: Address) -> Address {
+        let signer_addresses: Vec<_> =
+            <EthereumWallet as NetworkWallet<Ethereum>>::signer_addresses(&self.wallet).collect();
+        if signer_addresses.contains(&address) {
+            address
+        } else {
+            self.wallet.default_signer().address()
+        }
+    }
 }
 
 pub struct SubstrateNodeResolver {

crates/node/src/node_implementations/zombienet.rs

@@ -3,25 +3,17 @@
 //! ## Required Binaries
 //! This module requires the following binaries to be compiled and available in your PATH:
 //!
-//! 1. **polkadot-parachain**:
-//!    ```bash
-//!    git clone https://github.com/paritytech/polkadot-sdk.git
-//!    cd polkadot-sdk
-//!    cargo build --release --locked -p polkadot-parachain-bin --bin polkadot-parachain
-//!    ```
+//! 1. **polkadot-parachain**: ```bash git clone https://github.com/paritytech/polkadot-sdk.git cd
+//!    polkadot-sdk cargo build --release --locked -p polkadot-parachain-bin --bin
+//!    polkadot-parachain ```
 //!
-//! 2. **eth-rpc** (Revive EVM RPC server):
-//!    ```bash
-//!    git clone https://github.com/paritytech/polkadot-sdk.git
-//!    cd polkadot-sdk
-//!    cargo build --locked --profile production -p pallet-revive-eth-rpc --bin eth-rpc
-//!    ```
+//! 2. **eth-rpc** (Revive EVM RPC server): ```bash git clone https://github.com/paritytech/polkadot-sdk.git
+//!    cd polkadot-sdk cargo build --locked --profile production -p pallet-revive-eth-rpc --bin
+//!    eth-rpc ```
 //!
-//! 3. **polkadot** (for the relay chain):
-//!    ```bash
-//!    # In polkadot-sdk directory
-//!    cargo build --locked --profile testnet --features fast-runtime --bin polkadot --bin polkadot-prepare-worker --bin polkadot-execute-worker
-//!    ```
+//! 3. **polkadot** (for the relay chain): ```bash # In polkadot-sdk directory cargo build --locked
+//!    --profile testnet --features fast-runtime --bin polkadot --bin polkadot-prepare-worker --bin
+//!    polkadot-execute-worker ```
 //!
 //! Make sure to add the build output directories to your PATH or provide
 //! the full paths in your configuration.
@@ -585,6 +577,16 @@ impl EthereumNode for ZombieNode {
                 as Pin<Box<dyn Stream<Item = MinedBlockInformation>>>)
         })
     }
+
+    fn resolve_signer_or_default(&self, address: Address) -> Address {
+        let signer_addresses: Vec<_> =
+            <EthereumWallet as NetworkWallet<Ethereum>>::signer_addresses(&self.wallet).collect();
+        if signer_addresses.contains(&address) {
+            address
+        } else {
+            self.wallet.default_signer().address()
+        }
+    }
 }
 
 pub struct ZombieNodeResolver<F: TxFiller<ReviveNetwork>, P: Provider<ReviveNetwork>> {

crates/report/src/aggregator.rs

@@ -442,7 +442,8 @@ impl Report {
 
 #[derive(Clone, Debug, Serialize, Default)]
 pub struct TestCaseReport {
-    /// Information on the status of the test case and whether it succeeded, failed, or was ignored.
+    /// Information on the status of the test case and whether it succeeded, failed, or was
+    /// ignored.
     #[serde(skip_serializing_if = "Option::is_none")]
     pub status: Option<TestCaseStatus>,
     /// Information related to the execution on one of the platforms.

crates/report/src/runner_event.rs

@@ -8,8 +8,10 @@ use anyhow::Context as _;
 use indexmap::IndexMap;
 use revive_dt_common::types::PlatformIdentifier;
 use revive_dt_compiler::{CompilerInput, CompilerOutput};
-use revive_dt_format::metadata::Metadata;
-use revive_dt_format::{corpus::Corpus, metadata::ContractInstance};
+use revive_dt_format::{
+    corpus::Corpus,
+    metadata::{ContractInstance, Metadata},
+};
 use semver::Version;
 use tokio::sync::{broadcast, oneshot};