Cargo.toml

[package]
name = "raze-router"
version = "0.1.0"
edition = "2021"
description = "Generic CPI swap router — executes client-built hops under an on-chain safety envelope."
license = "Apache-2.0"

[lib]
crate-type = ["cdylib", "lib"]
name = "raze_router"

[features]
no-entrypoint = []
no-idl = []
no-log-ix-name = []
cpi = ["no-entrypoint"]
# The deployed .so ships with both of these on.
#
# `no-idl` drops the seven anchor IDL handlers and the `anchor:idl`
# create_with_seed path (which pulls in sol_sha256) — dead weight for a program
# whose clients encode instructions directly rather than from an on-chain IDL.
# `no-log-ix-name` silences the per-transaction `Program log: Instruction: …`
# sol_log, worth ~100 CU on every route. `idl-build` below is a separate compile
# pass, so generating IDL JSON for humans still works.
#
# ⚠️ `no-idl` STUBS IdlCreateAccount/Resize/Close. Before deploying with it,
# verify no IDL account already exists at the program's `anchor:idl` seed — once
# this build is live, that account can only be managed again by redeploying an
# IDL-enabled build.
default = ["no-idl", "no-log-ix-name"]
idl-build = ["anchor-lang/idl-build", "anchor-spl/idl-build"]

[dependencies]
anchor-lang = { version = "0.32.1", features = ["init-if-needed"] }
# Minimal feature set: we only use token_interface (Mint/TokenAccount) + the
# token_2022 transfer_checked CPI. Dropping `associated_token` and
# `token_2022_extensions` (spl-pod, token-group/metadata interfaces) trims the
# .so → lower deploy rent.
anchor-spl = { version = "0.32.1", default-features = false, features = ["mint", "token", "token_2022"] }
bytemuck = { version = "1", features = ["min_const_generics", "derive"] }

[dev-dependencies]
solana-program-test = "2.3"
solana-sdk = "2.3"
tokio = { version = "1", features = ["macros", "rt-multi-thread"] }

[profile.release]
overflow-checks = true   # keep: financial program, cheap safety (~2.9k CU)
# opt-level 3, deliberately not "z". The size-optimized build cost ~29% of this
# program's per-route compute (~42k CU, measured A/B in an SVM replay with the
# venue CPI byte-identical) and bought back only ~12% of the .so size. For a
# program on a hot path, runtime compute beats deploy rent — but check the
# resulting .so still fits the programdata account already provisioned on-chain,
# or the upgrade needs an `extend` first.
opt-level = 3
lto = "fat"
codegen-units = 1
strip = true             # drop symbols from the .so

[profile.release.build-override]
opt-level = 3
incremental = false
codegen-units = 1