precompile/GO_MODERNIZATION.md

Go 1.26 Modernization Analysis — luxfi/precompile

Analysis date: 2026-04-10 Go version: 1.26.1 Codebase: 72,578 lines across 137 Go files Test baseline: 5 failures (ed25519 panic, mldsa/mlkem build, go vet warnings)

1. Go Version Compatibility Matrix

Repo	Current	Recommended	Action
lux/precompile	1.26.1	1.26.1	None
lux/evm	1.26.1	1.26.1	None
lux/node	1.26.1	1.26.1	None
liquidity/evm	1.26.1	1.26.1	None
liquidity/dex	1.26	1.26.1	Bump to 1.26.1
liquidity/fhe	1.26	1.26.1	Bump to 1.26.1

Two repos are on 1.26 (no patch). Go 1.26.1 includes security fixes [1]. zoo/chains and hanzo/chains directories do not exist.

2. Critical: Consensus Safety Bug in pqcrypto

File: pqcrypto/contract.go:339 Severity: CRITICAL (chain-splitting)

// CURRENT — NON-DETERMINISTIC:
ciphertext, sharedSecret, err := pubKey.Encapsulate(rand.Reader)

rand.Reader produces different bytes on every validator. Different validators will produce different ciphertexts for the same input, causing state divergence.

The hpke/contract.go handles this correctly with a deterministicReader that derives pseudorandom bytes from a caller-provided seed via iterated SHA-256. The mlkem/contract.go also avoids this by not passing an explicit reader.

Fix: Replace rand.Reader with a deterministic reader seeded from calldata, identical to the pattern in hpke/contract.go:100-114. Or require a 32-byte seed in the encapsulate calldata format and use newDeterministicReader(seed).

3. Actionable Improvements

3a. interface{} to any (30+ occurrences)

Go 1.18 introduced any as an alias for interface{}. The go fix tool's any modernizer handles this automatically.

Files affected (partial list):

• dex/pool_manager.go:858 — args ...interface{}
• attestation/attestation.go:463,467,546 — decode/encode helpers
• graph/graphvm/types.go:31,37,48 — all map/interface types
• graph/graph.go:27,30,175,259,320,351 — query interfaces
• graph/graphvm_client.go:48,84 — client query methods
• blake3/contract_test.go:18-19 — mock state methods

Command: go fix -fix any ./...

3b. sort.Sort to slices.SortFunc (modules/registerer.go)

The moduleArray type in modules/module.go:25-37 exists solely to satisfy sort.Interface. Replace with slices.SortFunc and delete the type.

Current (modules/module.go:25-37 + modules/registerer.go:257):

type moduleArray []Module
func (u moduleArray) Len() int           { return len(u) }
func (u moduleArray) Swap(i, j int)      { u[i], u[j] = u[j], u[i] }
func (m moduleArray) Less(i, j int) bool { return bytes.Compare(...) < 0 }

sort.Sort(moduleArray(data))

After:

slices.SortFunc(data, func(a, b Module) int {
    return bytes.Compare(a.Address.Bytes(), b.Address.Bytes())
})

Deletes 13 lines of boilerplate. Also applies to dex/state_export.go which uses sort.Slice (can migrate to slices.SortFunc for consistency).

3c. rand.Read error checks (10 occurrences)

Since Go 1.24, crypto/rand.Read never returns an error [2]. Error checks on rand.Read are dead code.

File	Line	Current	After
x25519/deep_test.go	88	_, err := rand.Read(scalar)	rand.Read(scalar)
x25519/deep_test.go	112-113	_, _ = rand.Read(...)	rand.Read(...)
curve25519/deep_test.go	213	_, err := rand.Read(buf[:])	rand.Read(buf[:])
xwing/contract_test.go	130	_, err := rand.Read(garbage)	rand.Read(garbage)
Corona/contract_test.go	227	if _, err := rand.Read(prfKey); err != nil	rand.Read(prfKey)
ed25519/security_test.go	34,75	_, _ = rand.Read(message)	rand.Read(message)
ring/security_test.go	55,152	_, _ = rand.Read(...)	rand.Read(...)

3d. runtime/secret for cryptographic erasure (Go 1.26 experimental)

Go 1.26 adds runtime/secret (GOEXPERIMENT=runtimesecret) which zeroes registers and stack after a function returns [3]. Directly applicable to:

Package	Hot path	Why
frost/contract.go	verifySchnorrSignature	LiftX handles public key points
pqcrypto/contract.go	mlkemDecapsulate	Private key in calldata
pqcrypto/extended_contract.go	mldsaSign, slhdsaSign	Private key signing
threshold/accel.go	verifySingleSignature	Key material during verify
fhe/fhe_ops.go	initTFHE	Secret key generation

Implementation: Wrap with secret.Do(func() { ... }) once GOEXPERIMENT=runtimesecret is stable. Currently amd64+arm64 Linux only.

Risk: Experimental. Do not gate production builds on it. Use build tags:

//go:build goexperiment.runtimesecret

3e. crypto/hpke stdlib replacement (Go 1.26)

Go 1.26 added crypto/hpke implementing RFC 9180 [4]. Current code uses github.com/cloudflare/circl/hpke.

Compatibility check:

Feature	circl	stdlib crypto/hpke
P-256/P-384/P-521 KEMs	Yes	Yes
X25519 KEM	Yes	Yes
X25519+Kyber768 hybrid	Yes	Yes (post-quantum)
X-Wing hybrid	Yes	Needs verification
Deterministic Setup (custom rand)	Yes (sender.Setup accepts io.Reader)	Needs verification

Blocker: The hpke/contract.go:381 passes nil to sender.Setup() which circl interprets as "use default randomness". The deterministic reader is used via a different code path. Must verify stdlib crypto/hpke accepts a custom io.Reader for consensus-safe deterministic encryption before migrating.

Verdict: Investigate but do not migrate until the custom-reader API is confirmed. If confirmed, drop cloudflare/circl (saves ~15MB of transitive dependencies).

3f. go vet fixes (5 issues found)

blake3/deep_test.go:245     TestDeriveKey redeclared in this block
curve25519/deep_test.go:206 basepoint redeclared in this block
mlkem/deep_test.go:73       priv.Decapsulate undefined (API mismatch)
mldsa/deep_test.go:21       undefined: ContractAddress
pedersen/contract.go:273    unreachable code

These cause build failures in mldsa and mlkem packages and indicate stale deep_test.go files that were not updated after API changes.

3g. bytes.Equal in tests — use require.Equal

28 occurrences of require.True(t, bytes.Equal(a, b), ...) in test files. Replace with require.Equal(t, a, b) which provides better diff output on failure.

Affected packages: babyjubjub, x25519, pasta, pedersen, poseidon, curve25519, blake3.

4. Dependency Reductions

Can drop (with work)

Dependency	Used by	Replacement	Savings
cloudflare/circl	hpke/	crypto/hpke (stdlib)	~15MB transitive deps
golang.org/x/crypto/curve25519	x25519/	None (see below)	Minimal

Cannot drop

Dependency	Reason
luxfi/crypto/mlkem	Wraps all 3 modes (512/768/1024). Stdlib only has 768+1024. Mode-based API differs.
luxfi/crypto/mldsa	No stdlib equivalent. ML-DSA (FIPS 204) not in Go stdlib.
luxfi/crypto/slhdsa	No stdlib equivalent. SLH-DSA (FIPS 205) not in Go stdlib.
golang.org/x/crypto/curve25519	Stdlib crypto/ecdh wraps X25519 in high-level key API. Precompile needs raw X25519(scalar, point) which only x/crypto provides.
consensys/gnark-crypto	BLS12-381, BN254, Pasta curves, Poseidon hash. No stdlib equivalent.
supranational/blst	BLS12-381 via CGO. Performance-critical.
zeebo/blake3	No stdlib Blake3.
holiman/uint256	256-bit integer math for EVM. No stdlib equivalent.

Free performance from Go 1.26 (already active)

• Green Tea GC: 10-40% lower GC overhead for marking/scanning [5]. Benefits all precompiles that allocate during execution (ZK verifier, DEX math).
• 30% faster cgo: Benefits supranational/blst (BLS12-381) directly [6].
• Stack-allocated slices: Compiler allocates small slices on stack. Benefits the many make([]byte, 32) patterns in crypto precompiles.

5. Performance Wins

5a. PGO (Profile-Guided Optimization)

No PGO profile exists. Go 1.21+ supports PGO for 2-7% throughput improvement on hot paths, with some workloads seeing up to 15% [7].

Plan:

1. Add benchmark suite covering hot paths (FROST verify, BLS aggregate, ZK Groth16 verify, DEX swap math, Poseidon hash)
2. Run benchmarks with -cpuprofile=default.pgo
3. Commit default.pgo to repo root
4. Go toolchain automatically uses it on subsequent builds

Estimated gas savings: 2-7% reduction in wall-clock time for crypto precompiles translates to more accurate gas pricing (current gas costs are calibrated to unoptimized builds).

Hot paths for profiling (by test duration):

• fhe: 71.8s (FHE is inherently slow; PGO helps most here)
• pqcrypto: 4.7s
• Corona: 4.8s
• slhdsa: 3.7s
• zk: 3.8s

5b. Frost liftXCache improvement

frost/contract.go:147-182 uses a sync.RWMutex + map[[32]byte]curve.Point bounded to 1024 entries with no eviction policy. When full, new keys are silently dropped.

Improvement: Replace with sync.Map (better for read-heavy workloads, which this is — validators reuse the same keys) or implement LRU eviction. The current "stop caching when full" behavior means the 1025th unique public key gets zero cache benefit permanently.

5c. SIMD (not ready)

Go 1.26 adds experimental simd/archsimd (GOEXPERIMENT=simd) for AMD64 only [8]. ARM64 SVE support is planned but not yet available. Since production nodes run on both architectures, this is not actionable now. Revisit when ARM64 is supported.

Potential targets when ready:

• pasta/contract.go — field element arithmetic
• poseidon/contract.go — sponge state permutation
• bls12381/contract.go — point addition/multiplication
• dex/full_math.go — 256-bit multiplication

6. go fix Modernizer Batch

Go 1.26 ships 24 built-in modernizers [9]. Applicable ones:

Modernizer	Occurrences	Risk
any	30+	None (alias, identical semantics)
slicessort	1 (modules/) + 5 (dex/)	None
slicescontains	Check needed	None
forvar	0 (already correct)	N/A
rangeint	Check needed	None

Command: go fix -fix any,slicessort,slicescontains,rangeint ./...

Run go fix -diff ./... first to preview changes without applying them.

7. Existing Bugs Found During Analysis

Issue	File	Severity
Non-deterministic rand.Reader in encapsulate	pqcrypto/contract.go:339	CRITICAL
TestEdge_AllZeros panic	ed25519/deep_test.go:45	Medium
priv.Decapsulate undefined	mlkem/deep_test.go:73	Medium (build fail)
ContractAddress undefined	mldsa/deep_test.go:21	Medium (build fail)
TestDeriveKey redeclared	blake3/deep_test.go:245	Low (build fail)
basepoint redeclared	curve25519/deep_test.go:206	Low (build fail)
Unreachable code	pedersen/contract.go:273	Low

8. Summary of Recommendations

Priority order:

1. FIX pqcrypto/contract.go:339 consensus bug (deterministic encapsulation)
2. FIX 5 go vet failures (build-breaking in mldsa, mlkem, blake3, curve25519)
3. RUN go fix -fix any,slicessort ./... (mechanical, zero-risk)
4. REMOVE rand.Read error checks (10 occurrences, dead code since Go 1.24)
5. REPLACE sort.Sort(moduleArray(...)) with slices.SortFunc (delete boilerplate)
6. GENERATE PGO profile from benchmark suite (2-7% throughput improvement)
7. INVESTIGATE crypto/hpke stdlib migration (potential circl dep removal)
8. PLAN runtime/secret integration (when GOEXPERIMENT stabilizes)
9. BUMP liquidity/dex and liquidity/fhe from Go 1.26 to 1.26.1

Sources

• [1] https://go.dev/doc/devel/release — Go release history
• [2] https://go.dev/doc/go1.24 — crypto/rand.Read infallible since 1.24
• [3] https://pkg.go.dev/runtime/secret — runtime/secret experimental package
• [4] https://pkg.go.dev/crypto/hpke — stdlib HPKE (RFC 9180)
• [5] https://github.com/golang/go/issues/73581 — Green Tea GC
• [6] https://go.dev/doc/go1.26 — Go 1.26 release notes (cgo 30% faster)
• [7] https://go.dev/doc/pgo — Profile-Guided Optimization
• [8] https://pkg.go.dev/simd/archsimd — experimental SIMD
• [9] https://go.dev/blog/gofix — go fix modernizers (Go 1.26)
• [10] https://pkg.go.dev/crypto/mlkem — stdlib ML-KEM (FIPS 203)