🔒 Covered by ReqProof — L3 Assurance (123 requirements, 0 errors, 0 warnings, 0 open known issues)

New API: Append

// Append to an array without knowing its length
data, _ = jsonparser.Append(data, []byte(`"new_item"`), "items")

Append(data, value, keys...) ([]byte, error) — clean array-append API. Works on top-level and nested arrays. Auto-creates missing paths as single-element arrays. No need for [N] path syntax.

Bug fixes — all known issues resolved

Zero open known issues. All 4 KIs are now status: fixed.

Full changelog: CHANGELOG.md

🔒 Covered by ReqProof — L3 Assurance (123 requirements, 0 errors, 0 warnings)

Performance — 6.1x large-payload speedup

Two optimizations found via ADHD-driven profiling + 10 parallel worktree experiments:

1. Fix stringEnd unbounded backslash scan — stringEndConfig was scanning the ENTIRE remaining parent document for backslashes instead of just the string body. Bounded to data[:firstQuote]. 128µs → 22µs (5.8x).

2. SWAR string scan — replaced two separate bytes.IndexByte calls with a single inline 8-byte SWAR loop checking for both " and \ simultaneously. 22µs → 21µs (additional 8%).

Fresh benchmarks (all libraries at latest versions)

Methodology: Apple M4 Max (ARM64), Go 1.26.3, median of 5 runs. All comparison libraries updated. The encoding/json benchmark no longer uses ffjson-generated methods (fixed #126 in v1.3.1).

jsonparser is the fastest library overall on large payloads and the only zero-allocation parser.

Full changelog: CHANGELOG.md

🔒 Covered by ReqProof — L3 Assurance (123 requirements, 0 errors, 0 warnings)

Config struct — opt-in lenient parsing (#160, #115)

var Lenient = jsonparser.Config{AllowSingleQuotes: true, AllowUnknownEscapes: true}
Lenient.Get(data, "key")  // parses {'key':'value'} and unknown escapes

Single-quote support and lenient escape handling via an opt-in Config. Default stays strict (RFC 8259).

Streaming ReaderParser (#132, #257)

rp := jsonparser.NewReaderParser(file)  // any io.Reader
rp.Get("users", "[0]", "name")          // path-based access from a stream

Path-based JSON access from an io.Reader — parse 10GB+ files without loading into memory. Buffers in 64KB chunks; memory bounded by the largest value.

Name aliases (#66)

EachArray, EachObject, EachArrayErr, EachArrayWildcard — canonical EachXxx pattern. Old XxxEach names kept for backward compatibility.

Proof coverage

• 2 new SYS-REQs (115: Config/lenient, 116: streaming)
• 123 requirements, 0 errors, 0 warnings, 279/279 functions traced

Full changelog: CHANGELOG.md

New APIs (all backward-compatible)

Iteration with error/break control

• ArrayEachErr — like ArrayEach but the callback returns error. Return io.EOF for graceful stop, any other error to abort. Resolves #53, #129, #176, #230, #255, #262.
• EachKeyErr — same pattern for EachKey.

Safe string handling

• Escape(s string) []byte — RFC 8259 string escaping (inverse of Unescape). Produces a quoted JSON string literal.
• SetString(data, val, keys...) — Set with auto-quoted value. No more invalid JSON from forgetting quotes. Resolves #144, #158, #218, #270.

Container accessors

• GetArrayLen(data, keys...) (int, error) — count array elements without a callback. Resolves #175, #261.
• GetObjectLen(data, keys...) (int, error) — count object key-value pairs.
• GetUint64(data, keys...) (uint64, error) — uint64 variant of GetInt. Resolves #271.

Delete found signal

• DeleteFound(data, keys...) ([]byte, bool) — returns whether the key was found. Resolves #229.

Wildcard paths

• EachKeyWildcard, ArrayEachWildcard, SetWildcard — [*] path component to iterate/set all elements. Resolves #112.

  jsonparser.SetWildcard(data, []byte("true"), "users", "[*]", "active")

JSONPath compiled paths

• ParsePath("$.users[0].name") → []string{"users", "[0]", "name"}
• CompilePath + CompiledPath — pre-compile and reuse with Get/Set/Delete/etc. Resolves #234, #251.

  path, _ := jsonparser.CompilePath("$.person.name.fullName")
  name, _ := path.Get(data)  // reuse across calls

Fixes

• EachKey no longer panics with >64 key components (#56)
• Set pre-allocates output buffer, reducing allocations from 6 to 1 (#107)

Proof coverage

• 3 new SYS-REQs (112: container length, 113: wildcard paths, 114: compiled paths)
• 121 total requirements, 0 errors, 0 warnings, 384 MC/DC witness rows (0 uncovered)
• All new APIs traced via source-native annotations

Acknowledgments

Implemented by codex (gpt-5-codex) via codex exec. Proof coverage by ReqProof.

Full changelog: CHANGELOG.md

Bug fixes

• Fix Set/Delete input-buffer aliasing (#209, #141) — Set and Delete no longer corrupt the caller's input []byte when the slice has spare capacity. The append() call path was writing into the backing array beyond the returned slice. Now all mutation paths allocate a fresh buffer.

• Fix EachKey array-index inconsistency (#232) — EachKey now descends into terminal array-index paths (e.g. "key", "[0]") consistently with Get. Previously EachKey returned empty where Get succeeded on the same path.

• Fix benchmark measuring ffjson, not encoding/json (#126) — the benchmark payload types had ffjson-generated MarshalJSON/UnmarshalJSON methods, so the "10x faster than encoding/json" comparison was silently measuring ffjson. Now uses plain types with no generated methods.

Proof strengthening (how these bugs escaped, and the gates that prevent recurrence)

Contributed by codex (gpt-5-codex) via codex exec.

🔒 Formally verified by ReqProof

jsonparser v1.3.0 is the first Go library proven to L3 assurance by ReqProof — a git-native requirements-engineering and formal-verification platform. Every public API is traced to a formal requirement, every requirement is tested with 100% MC/DC coverage, and the entire parser is fuzzed by a custom structure-aware JSON fuzzer at 250k inputs/sec.

The proof review caught 7 real bugs that years of community use, OSS-Fuzz, and standard fuzzing had missed.

Read the root-cause analysis →

Security / bug fixes

• Fix Delete panic on malformed input with leading comma (OSS-Fuzz 4649128545288192)
• Fix empty-string key-component panics (8 sites) — reported by @c-tonneslan (#284)
• Fix Set data loss on scalar arrays (#267) — reported by @Solaris-star (#286)
• Fix Set malformed-JSON output on cross-type paths (auto-coerce)
• Fix Delete trailing-comma malformation
• Fix ArrayEach spurious callback on non-array root
• Fix lone-Unicode-surrogate mishandling in Unescape (now matches encoding/json)

Performance

• parseInt fast-path — 22–37% faster on short integers — @trevorprater (#285)
• stringEnd SIMD fast path — 12× faster on no-escape strings

Full changelog: CHANGELOG.md

Acknowledgments

Thank you to @c-tonneslan, @Solaris-star, @trevorprater, and OSS-Fuzz for the reports and contributions.

What's Changed

• Fix 2 bugs, remove 7 dead code blocks, add formal verification using ReqProof by @buger in #281
• Add support for tinygo by @buger in #269

Full Changelog: v1.1.2...v1.2.0

What's Changed

• Updated travis to build for 1.13 to 1.15 by @janreggie in #225
• • eliminate 2 allocations in EachKey() by @Villenny in #223
• fix issue #150 (in deleting case) by @daria-kay in #226
• fixing the oss-fuzz issue by @daria-kay in #227
• Fix parseInt overflow check false negative by @carsonip in #231
• Added bespoke error for null cases by @jonomacd in #228
• Fuzzing: Add CIFuzz by @AdamKorcz in #239
• Added latest versions of go to tests by @moredure in #244
• fix EachKey pIdxFlags allocation by @unxcepted in #241
• fix: prevent panic on negative slice index in Delete with malformed JSON (GO-2026-4514) by @dbarrosop in #276

New Contributors

• @janreggie made their first contribution in #225
• @Villenny made their first contribution in #223
• @daria-kay made their first contribution in #226
• @carsonip made their first contribution in #231
• @jonomacd made their first contribution in #228
• @moredure made their first contribution in #244
• @unxcepted made their first contribution in #241
• @dbarrosop made their first contribution in #276

Full Changelog: v1.1.1...v1.1.2

Fix CVE-2020-35381. PR #221

Christmas present 🎁

• Improve Set performance by 20% #196
• Memory allocation improvements during Array iteration #208
• Add support for ppc64-le architecture #213
• Fixed repeated paths from incorrectly incrementing data offset #215
• Fixed array iteration when you have more than 64 items #216
• Fix possible memory confusion in unsafe slice cast #204
• Fixed usage of Array iteration with Set #199
• Integration with go-fuzz #217

Thanks to everyone who participated in the project!
@AllenX2018 @saginadir @xsandr @nagesh4193 @rrgilchrist @floren @AdamKorcz @jlauinger