perf(core/felt): reduce unneeded heap allocations

[brbrr]

note: the original implementation was failing no quotes test - it was successfully unmarshalling non-quoted hex strings, such as 0xdeadbeef. I think it make sense to reject them as, JSON context, all of the felt values should be quoted strings

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 75.84%. Comparing base (4924bc8) to head (cbaf1ac).
⚠️ Report is 8 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3565      +/-   ##
==========================================
+ Coverage   75.77%   75.84%   +0.06%     
==========================================
  Files         384      380       -4     
  Lines       33750    33740      -10     
==========================================
+ Hits        25575    25591      +16     
+ Misses       6386     6361      -25     
+ Partials     1789     1788       -1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Copilot]

Pull request overview

Refactors felt.Felt JSON (un)marshalling to reduce heap allocations and tighten validation so only quoted 0x-prefixed hex strings are accepted in JSON contexts.

Changes:

• Reimplements Felt.UnmarshalJSON with fixed-size padding + hex.Decode and stricter “must be quoted string” validation.
• Reimplements Felt.MarshalJSON to emit a quoted 0x hex string without unnecessary leading zeros, with fewer allocations.
• Adds comprehensive JSON marshal/unmarshal and round-trip tests, including the previously-missed “no quotes” invalid case.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 4 comments.

File	Description
core/felt/felt.go	Updates JSON marshal/unmarshal implementations for lower allocation and stricter input rules.
core/felt/felt_test.go	Adds targeted test coverage for valid/invalid JSON inputs and round-trip behavior.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[rodrodros]

Felt JSON marshaling benchmarks

Before/after comparison of Felt.MarshalJSON and Felt.UnmarshalJSON.

• Host: Apple M2 Pro · darwin/arm64 · Go 1.26.1
• Command: go test -run=^$ -bench='BenchmarkMarshalJSON|BenchmarkUnmarshalJSON' -benchmem -count=10 ./core/felt/...
• old = main at merge-base (4924bc81)

Latency (sec/op)

Benchmark	old	new	Δ
MarshalJSON/zero	48.53 ns	28.59 ns	−41.09%
MarshalJSON/small	62.81 ns	32.22 ns	−48.71%
MarshalJSON/address	141.30 ns	46.29 ns	−67.24%
MarshalJSON/max_felt	141.80 ns	46.89 ns	−66.94%
MarshalJSON/padded_address	141.70 ns	46.56 ns	−67.15%
MarshalJSON/random	141.65 ns	47.08 ns	−66.76%
UnmarshalJSON/zero	85.66 ns	63.04 ns	−26.41%
UnmarshalJSON/small	131.90 ns	62.21 ns	−52.84%
UnmarshalJSON/address	438.65 ns	62.65 ns	−85.72%
UnmarshalJSON/max_felt	445.75 ns	62.93 ns	−85.88%
UnmarshalJSON/padded_address	438.10 ns	62.86 ns	−85.65%
UnmarshalJSON/random	442.90 ns	62.95 ns	−85.79%
geomean	168.5 ns	50.39 ns	−70.10%

All deltas are statistically significant (p = 0.000, n = 10).

Memory (B/op)

Benchmark	old	new	Δ
MarshalJSON/zero	8 B	8 B	—
MarshalJSON/small	24 B	16 B	−33.33%
MarshalJSON/address	288 B	80 B	−72.22%
MarshalJSON/max_felt	288 B	80 B	−72.22%
MarshalJSON/padded_address	288 B	80 B	−72.22%
MarshalJSON/random	288 B	80 B	−72.22%
UnmarshalJSON/zero	37 B	0 B	−100.00%
UnmarshalJSON/small	48 B	0 B	−100.00%
UnmarshalJSON/address	112 B	0 B	−100.00%
UnmarshalJSON/max_felt	112 B	0 B	−100.00%
UnmarshalJSON/padded_address	112 B	0 B	−100.00%
UnmarshalJSON/random	112 B	0 B	−100.00%

Allocations (allocs/op)

Benchmark	old	new	Δ
MarshalJSON/zero	1	1	—
MarshalJSON/small	2	1	−50%
MarshalJSON/address	4	1	−75%
MarshalJSON/max_felt	4	1	−75%
MarshalJSON/padded_address	4	1	−75%
MarshalJSON/random	4	1	−75%
UnmarshalJSON/zero	2	0	−100%
UnmarshalJSON/small	2	0	−100%
UnmarshalJSON/address	2	0	−100%
UnmarshalJSON/max_felt	2	0	−100%
UnmarshalJSON/padded_address	2	0	−100%
UnmarshalJSON/random	2	0	−100%

The remaining 1 alloc on MarshalJSON is the returned []byte — unavoidable since the method contract is to return a slice.

Takeaways

• UnmarshalJSON is now fully zero-alloc on every input shape. The old implementation allocated twice per call — once for strings.ToLower / strings.Trim, once for the pooled big.Int used by SetString.
• MarshalJSON drops from 4 → 1 alloc and ~66% faster on full-width inputs. The single remaining alloc is the return slice; it's also 3.6× smaller (80 B vs 288 B) because the new code sizes it precisely instead of going through a z.String() intermediate.
• Geomean latency across all 12 sub-benchmarks improved by −70.10%; the UnmarshalJSON group is where the biggest wins are (−86% on any realistic address-sized input).

[rodrodros]

Looks good to me ;)