Decorators & Attributes

Defined in stdlib/sequre/attributes.codon, stdlib/sequre/decorators.codon, and stdlib/sequre/runtime.codon

Sequre uses compile-time attributes (processed by the Sequre IR plugin) and runtime decorators to control how functions are transformed, optimized, and executed.

---

Compiler attributes

These decorators are processed by the Sequre compiler plugin during IR transformation passes. They do not exist at runtime — they instruct the compiler to rewrite annotated functions.

@sequre

The primary attribute. Marks a function for Sequre's IR rewriting pipeline. The compiler:

1. Expressiveness transformations — rewrites operator overloads (e.g., a > b on encrypted types dispatches to secure comparison)
2. MPC optimizations (@mpc_poly_opt) — polynomial-evaluation optimizations for MPC operations
3. MHE optimizations (@mhe_mat_opt, @mhe_cipher_opt, @mhe_enc_opt) — HE-specific optimizations

from sequre.attributes import sequre

@sequre
def my_secure_function(mpc, x, y):
    return x @ y + x * y  # operators are rewritten to secure versions

Note

Every function that operates on Sequre types (Sharetensor, Ciphertensor, MPU, etc.) should be annotated with @sequre.

---

Runtime decorators

@flatten(idx)

Defined in stdlib/sequre/decorators.codon

A runtime decorator that automatically flattens the first idx tensor arguments before calling the function, and reshapes the result back to the original shape. Useful for functions that operate on 1-D vectors but should accept matrices transparently.

from sequre.decorators import flatten

@flatten(1)
def fp_div(mpc, a, b):
    # a and b are flattened to 1-D here
    ...
    # result is reshaped back to original shape of args[1]

@local

Defined in stdlib/sequre/runtime.codon

A runtime decorator that forks the current process into N parties (using fork()), each running the decorated function as a separate MPC party with its own MPCEnv. Used for local testing where all parties run on a single machine.

from sequre.runtime import local

@local
def my_protocol(mpc):
    # Each forked process gets its own mpc with a unique pid
    X = MPU(mpc, local_data, "partition")
    result = X @ X.T
    print(f"CP{mpc.pid}: done")

Command-line flags (e.g., --ring, --skip-mhe-setup) are parsed from sys.argv and passed as control toggles.

---

Runtime initialization

Defined in stdlib/sequre/runtime.codon

These functions set up the MPC environment for distributed (non-local) execution:

Function	Description
mpc()	Parse command-line args, create an MPCEnv for the current party, run MHE setup. Returns the initialized environment.

Typical distributed entry point

from sequre.runtime import mpc as init_mpc

mpc = init_mpc()
# mpc.pid is set from sys.argv
# mpc.mhe is initialized with default_setup()

Typical local entry point

from sequre.runtime import local

@local
def main(mpc):
    ...  # protocol logic

main()  # forks N parties automatically --- no need to pass mpc instance

---

Compiler IR passes

The Sequre compiler plugin processes @sequre-annotated functions through these IR passes (in order):

Pass	Description
ExpressivenessTransformations	Rewrites standard operators on secure types to their secure equivalents
MPCOptimizations	Optimizes polynomial evaluations and MPC-specific patterns
MHEOptimizations	Optimizes HE expression ordering, encoding modes, and matrix strategies