pinocchio

Define an entrypoint with entrypoint!(process_instruction), route instructions by a single-byte discriminator, and validate accounts manually using AccountInfo checks. Use pinocchio-system and pinocchio-token crates for CPI helpers and bytemuck for zero-copy account data.

Pinocchio reduces compute units by 88-95% and binary size by about 40% compared to Anchor, suiting high-throughput programs like DEXs and orderbooks. Anchor is better for rapid prototyping, complex account relationships, and teams unfamiliar with low-level Rust.

Replace #[account] structs with bytemuck Pod types using manual one-byte discriminators, convert Anchor constraints into explicit is_signer and is_writable checks, and swap CpiContext calls for pinocchio-system invoke helpers. Use Shank to generate the IDL that Anchor produced automatically.

Yes, the pinocchio-token crate provides Token-2022 instructions including metadata pointers and token metadata initialization, though support is under active development. Standard CPIs use invoke and invoke_signed with AccountMeta and Instruction structs.

Bytemuck structs must be padded to 8-byte alignment, and account data cannot be borrowed mutably while an immutable borrow exists. Drop the first borrow before borrowing again, and add explicit padding fields after any field smaller than 8 bytes.

Pinocchio is currently unaudited, so use it with caution in production despite being maintained by Anza, the Solana Agave client developers. Thorough testing with solana-program-test and manual security validation of all accounts is essential.