crates/starknet-devnet-core/src/account.rs (1)

171-201: 🛠️ Refactor suggestion

Balance-initialisation loop silently double-counts when both fee-token addresses coincide

for fee_token_address in [self.eth_fee_token_address, self.strk_fee_token_address] { … } will iterate twice over the same address whenever the caller passes identical ETH & STRK addresses (as done in the tests).
That inflates ERC20_total_supply and rewrites ERC20_balances twice, masking potential bugs in real-world scenarios and making the helper dangerous if a future caller accidentally reuses an address.

Minimal fix—deduplicate before the loop:

-        for fee_token_address in [self.eth_fee_token_address, self.strk_fee_token_address] {
+        use std::collections::HashSet;
+        let unique_addresses: HashSet<_> =
+            [self.eth_fee_token_address, self.strk_fee_token_address].into_iter().collect();
+
+        for fee_token_address in unique_addresses {

This avoids unintended double mutation while keeping complexity trivial.

website/docs/balance.md (1)

49-56: Specify language in fenced block to satisfy markdown-lint

Add json after the opening fence to silence MD040 and enable syntax highlighting.

-```
+```json
 {
     "tx_hash": "0x123..."
     "revert_reason": "Something happened"
 }
-```
+```

🧰 Tools

🪛 markdownlint-cli2 (0.17.2)

51-51: Fenced code blocks should have a language specified
null

(MD040, fenced-code-language)

tests/integration/test_subscription_to_tx_status.rs (1)

23-26: Hard-coded tx hash may require frequent maintenance

The deterministic hash assertion is valuable, but every tweak to minting logic or class hash will force a test update.
Consider deriving the expected hash via the same hashing routine used by Devnet instead of embedding a literal.

crates/starknet-devnet-core/src/account.rs (1)

315-316: Test only deploys STRK ERC20 class – ETH path left un-exercised

The test pre-deploys STRK_ERC20_CONTRACT_CLASS_HASH for both fee tokens, leaving the ETH implementation untested. Consider pre-deploying the ETH class hash to improve coverage of the split-token logic.

crates/starknet-devnet-core/src/starknet/mod.rs (1)

833-837: Selector literal updated, but calldata length comment unchanged

permissioned_mint takes 3 arguments (recipient, low, high).
The preceding comment "// target address" is still accurate, but consider adding a short note that
the selector was switched from "transfer" to "permissioned_mint" to avoid confusion when
walking through git history.

crates/starknet-devnet-server/src/api/http/endpoints/mint_token.rs (1)

66-86: Hold lock only as long as necessary

The tokio::Mutex guarding starknet remains locked during the entire await chain, including the
call to get_balance. This blocks every concurrent JSON-RPC request that needs starknet, even
though balance retrieval is read-only.

-    let mut starknet = api.starknet.lock().await;
-    ...
-    let tx_hash = starknet.mint(...).await?;
-    let tx = starknet.get_transaction_execution_and_finality_status(tx_hash)?;
-    match tx.execution_status {
+    let tx_hash = {
+        let mut starknet = api.starknet.lock().await;
+        starknet.mint(request.address, request.amount, erc20_address).await?
+    };
+
+    let execution_status = {
+        let starknet = api.starknet.lock().await;
+        starknet.get_transaction_execution_and_finality_status(tx_hash)?
+    };
+
+    match execution_status.execution_status {

Releasing the lock earlier improves concurrency without altering semantics.

crates/starknet-devnet-server/src/api/json_rpc/error.rs (1)

80-82: Consider exposing the revert reason directly in the error message

JSON-RPC clients often surface only the top-level message field. Putting the primary revert reason
there (in addition to data) greatly improves UX:

-    #[error("Minting reverted")]
+    #[error("Minting reverted{revert_reason_msg}")]
+    MintingReverted {
+        tx_hash: Felt,
+        #[source]
+        revert_reason: Option<String>,
+    },

where revert_reason_msg is formatted conditionally. Optional but worth considering.

crates/starknet-devnet-core/src/starknet/starknet_config.rs (2)

12-17: Import list updated – keep ordering consistent

The added ETH/STRK constants are correct, but the list is now mixed between “general defaults” and “token-specific” constants.
For readability, group the ETH/STRK lines together (or in a dedicated // ERC-20 block) to mirror the layout in constants.rs.

---

171-175: Large Sierra blobs are cloned on every Default instantiation

ETH_ERC20_CONTRACT_CLASS.to_string() (and the STRK variant) allocates and copies the entire Sierra source into a String each time StarknetConfig::default() is called. These files are > 100 KB and the default is created frequently in tests.

If the field does not need mutability, store &'static str (or Cow<'static, str>) instead to share the memory:

-    pub eth_erc20_contract_class: String,
+    pub eth_erc20_contract_class: &'static str,
 ...
-            eth_erc20_contract_class: ETH_ERC20_CONTRACT_CLASS.to_string(),
+            eth_erc20_contract_class: ETH_ERC20_CONTRACT_CLASS,

Same for strk_erc20_contract_class.

crates/starknet-devnet-core/src/constants.rs (1)

28-36: Consistent naming & path constants

Previous patterns expose both _SIERRA_PATH and _SIERRA (in-file string). For the new ETH/STRK contracts only the latter is provided. Aligning the naming keeps the public API predictable:

+pub const ETH_ERC20_CONTRACT_SIERRA_PATH: &str = concat!(
+    env!("CARGO_MANIFEST_DIR"),
+    "/contracts/system_artifacts/erc20_eth.sierra"
+);
+pub const ETH_ERC20_CONTRACT_CLASS: &str = include_str!(ETH_ERC20_CONTRACT_SIERRA_PATH);
-
-pub const ETH_ERC20_CONTRACT_CLASS: &str = include_str!(concat!(
-    env!("CARGO_MANIFEST_DIR"),
-    "/contracts/system_artifacts/erc20_eth.sierra"
-));

Apply the same pattern to STRK.
This reduces duplication and matches the existing CAIRO_1_ACCOUNT_CONTRACT_SIERRA[_PATH] naming.

Configuration used: .coderabbit.yaml
Review profile: CHILL
Plan: Pro