compile instruction accounts via mock_compile_message

[buffalojoec]

In Agave, top-level instruction accounts are queried from the compiled transaction Message, which compiles its account keys here (as CompiledKeys). When a message doesn't properly compile its account keys, duplicate entries are caught by Bank through this check. You can see this being tested here.

This means that a sanitized message in SVM's message processor is guaranteed to have no duplicate account keys. Due to the protocol-enforced layout of a Solana Message, no two instruction accounts sharing the same index can have different roles (writable/signer). This is because a Message uses an account key's index in its account_keys list to query its role. See is_signer and is_writable in the SDK.

In CPI, we don't have a Message layout to enforce uniqueness, but we know the transaction accounts have already been deduped (above). So, the program-runtime will merge a CPI instruction's instruction accounts to the highest role manually via prepare_next_cpi_instruction.

SolFuzz-Agave incorrectly builds the list of instruction accounts because it does not merge instruction accounts to their highest role. Instead, it assigns them whatever role was specified in the AccountMeta, meaning two accounts sharing the same index can have two different roles.

solfuzz-agave/src/instr.rs

Lines 233 to 242 in 4cb2fbc

	for account_meta in acct_metas.iter() {
	let index_in_transaction = txn_context
	.find_index_of_account(&account_meta.pubkey)
	.expect("invariant violation: account not found in transaction context")
	as IndexOfAccount;
	instruction_accounts.push(InstructionAccount::new(
	index_in_transaction,
	account_meta.is_signer,
	account_meta.is_writable,
	));

This PR patches this discrepancy and rectifies all input instruction accounts to resemble a valid message by actually using a real Message via mock_compile_message, a DCOU function from solana-program-runtime.

The issue is: this will break 100% of the instruction fixtures and all CPI syscall fixtures, since the syscall harness shares the same invoke context setup.

[Copilot]

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

Updates instruction and syscall execution setup to use Solana’s SanitizedMessage-based path (prepare_top_level_instructions) instead of the older test-only configure_top_level_instruction_for_tests, removing reliance on stable-layout instruction/account snapshots.

Changes:

• Replace StableInstruction/stable-layout snapshots with solana_instruction::Instruction and SanitizedMessage.
• Build a SanitizedMessage + transaction accounts via mock_compile_message and use prepare_top_level_instructions.
• Extend VM syscall cleanup to drop a leaked SanitizedMessage.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 4 comments.

File	Description
src/vm_syscalls.rs	Switches VM syscall execution to prepare_top_level_instructions and manages SanitizedMessage lifetime via leak + explicit cleanup.
src/vm_serialization.rs	Updates VM serialization execution to use SanitizedMessage and prepare_top_level_instructions.
src/instr.rs	Replaces stable-layout instruction types with Instruction, compiles a SanitizedMessage, and wires it into invoke context setup.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.