[workspace]
members = [
    "rakc",
    "ide/src-tauri",
    "stdlib",
    "rakpkg",
    "rak-setup",
]
resolver = "2"

[workspace.package]
version = "0.8.4"
edition = "2021"
authors = ["Ryuzaki"]
license = "MIT OR Apache-2.0"
repository = "https://github.com/ryuzaki/rak"

[workspace.dependencies]
# Common dependencies across crates
thiserror = "1.0"
anyhow = "1.0"
serde = { version = "1.0", features = ["derive"] }
tokio = { version = "1", features = ["full"] }
tracing = "0.1"
tracing-subscriber = "0.3"

# Compiler specific
logos = "0.14"          # Lexer generator
# inkwell = { version = "0.2", features = ["llvm15-0"] } # LLVM backend (optional)

# IDE / Tauri
tauri = { version = "1.6", features = ["shell-open"] }
serde_json = "1.0"
tower-lsp = "0.20"       # LSP server implementation
tokio-util = "0.7"

# Release hardening. These are the compiler-level defenses Rak can actually
# make about itself; they do not replace the missing control-flow integrity
# support noted in docs/V8-ROADMAP.md.
#
#   lto + codegen-units=1  whole-program optimisation, so a bounds check in
#                          one crate can be folded against a caller in another
#                          instead of living behind an opaque call
#   overflow-checks=true   keep arithmetic overflow panicking in release. The
#                          default is `false`, which silently wraps and turns
#                          a length calculation into a heap overflow
#   panic=abort            no unwinding through the FFI trampolines
#   strip                  no symbols to attach a debugger to by accident
[profile.release]
lto = "fat"
codegen-units = 1
overflow-checks = true
panic = "abort"
strip = "symbols"

# The fuzzing crate and the IDE both want a usable panic message, so they get
# their own profile rather than inheriting `panic = "abort"`.
[profile.release.package."*"]
opt-level = 2