Huff2 is the successor of the huff-rs compiler written in Rust. It comes with:
- better error handling
- clearer semantics (only 1 label def. per scope, checks for macro argument count match)
- push minimization (e.g. will use
PUSH1
instead ofPUSH2
if referencing a label withPC < 256
) - new builtins:
__codeoffset(macro_name: MacroIdentifier)
- Jump tables
- parsing
- builtins (
__tablestart
,__tablesize
)
- Code tables
- builtins (
__tablestart
,__tablesize
)
- builtins (
- ABI builtins (
__FUNC_SIG
,__EVEN_HASH
,__ERROR
) - Imports (
#include
statements)
The huff-rs
compiler was a passion project by pioneers
in the Huff community who aimed to create a better version of Zac's
original typescript implementation.
The initial developers of huff-rs
were relatively new to compilers, choosing to write some of the
components like the lexer, parser & assembler themselves while adding novel compiler features like
the Huff testing framework that unfortunately didn't see a lot of usage.
Combined with a lot of the tech debt that accrued from the early days made us decide that it was best to start fresh, using existing libraries to do as much of the heavy lifting as possible:
chumsky
for lexing & parsingariadne
for pretty errorsevm-glue
for EVM assemblyalloy
for ABI types & parsing
This new foundation will allow bugs to be fixed more easily as well as allowing us to experiment with our own novel compiler features. 😁
The -b
, --bytecode
, -r
, --bin-runtime
, -m
, --alt-main
, -t
, --alt-constructor
have
been replaced in favor of a required positional argument indicating what macro to compile and
optional -f
/ --default-constructor
flags to wrap the compiled result with a minimal default
constructor.
This was done to make the CLI simpler and clearer, you will always get a single output, the output you ask for and nothing extra will be added without you asking for it.
A lot of other flags were not reimplemented either because they were not widely used or because we just haven't gotten around to it. Raise an issue if you'd like to suggest a feature.
The compiler will not validatate certain things that were not checked/simply allowed in the previous compiler:
- duplicate label definitions in the same scope
- duplicate definitions (you could define multiple constants, macros, etc. with identical names)
- mismatching macro argument count
These errors were serious footguns that could easily go unnoticed when using the previous compiler.
- Up Only: Macros can only reference labels defined within them or the parents invoking them
- Shadowing: Label definitions deeper down in a chain of invocations shadows previous definitions
Examples:
References resolve to the labels defined within the same macro:
#define macro INNER() = {
target ✅ Resolves ─┐
│
0x1 0x1 │
add │
0x2 │
eq │
│
target: <────────────┘
}
Falls back to resolving to invoker's label:
#define macro MAIN() = takes(0) returns(0) {
INNER() ────┐
target: <───┘
}
#define macro INNER() = {
^
│
target ✅ Resolves
0x1 0x1
add
0x2
eq
}
Resolution does not go down into invoked macros:
#define macro MAIN() = takes(0) returns(0) {
INNER()
target ❌ Fails to Resolve
}
#define macro INNER() = {
target:
0x1 0x1
add
0x2
eq
}
As you go down an invocation chain label definitions are added to a stack where the highest most definition is resolved by references.
#define macro MAIN() = takes(0) returns(0) {
INNER()
target: 🟡 Shadowed by ────┐
} │
│
│
#define macro INNER() = { │
target ✅ Resolves ─┐ │
│ │
0x1 0x1 │ │
add │ │
0x2 │ │
eq │ │
│ │
target: <────────────┘ <───┘
}