Issues

Document what the fast code generation work learned about performance

hashgraph/pbj#1036 · opened

on Nov 2, 2026

Array parser: cut the per-nested-message cost of decoding

hashgraph/pbj#1031 · opened

on Oct 31, 2026

Views: cut the per-child cost of wrap and index

hashgraph/pbj#1030 · opened

on Oct 31, 2026

Writers: cut the per-nested-message cost of open and close

hashgraph/pbj#1029 · opened

on Oct 31, 2026

Synthetic HAPI block workload, benchmarks and reference codec matching the external block-stream workload

hashgraph/pbj#1027 · opened

on Oct 31, 2026

JSON codec writes and parses uint32/fixed32 >= 2^31 and uint64/fixed64 >= 2^63 as negative numbers

hashgraph/pbj#1025 · opened

on Oct 31, 2026

Byte-array writer writes corrupt wrapper messages for FloatValue/DoubleValue -0.0 and UInt32Value >= 2^31

hashgraph/pbj#1024 · opened

on Oct 31, 2026

A oneof message alternative set to null round-trips as the default instance

hashgraph/pbj#1021 · opened

on Oct 30, 2026

float and double fields set to -0.0 are not written, unlike protoc

hashgraph/pbj#1020 · opened

on Oct 30, 2026

1

Generated code does not compile for map fields with enum values

hashgraph/pbj#1019 · opened

on Oct 30, 2026

Zero-allocation flyweight writers (SBE-style encoders) for whole message trees

hashgraph/pbj#1018 · opened

on Oct 30, 2026

Unboxed oneof storage, part 1: storage, parsers and writers

hashgraph/pbj#1014 · opened

on Oct 30, 2026

Keep every generated method under HotSpot's 8,000-byte compile limit

hashgraph/pbj#1013 · opened

on Oct 30, 2026

JSON writer mis-indents a map field inside a nested message

hashgraph/pbj#1012 · opened

on Oct 30, 2026

1

Generated hashCode() repeats earlier fields' code before a map field

hashgraph/pbj#1011 · opened

on Oct 30, 2026

Unknown fields inside map entries always fail to parse with "Bad tag"

hashgraph/pbj#1009 · opened

on Oct 30, 2026

Parse string fields without per-call decoder and buffer allocations

hashgraph/pbj#1008 · opened

on Oct 30, 2026

Zero-copy views (part 2): oneofs, repeated fields and wrapper types

hashgraph/pbj#1006 · opened

on Oct 30, 2026

Zero-copy, zero-allocation views over encoded messages (part 1)

hashgraph/pbj#1004 · opened

on Oct 30, 2026

Reverse writer into a reusable buffer (fast_write)

hashgraph/pbj#1001 · opened

on Oct 30, 2026

1

Byte array parser with in-order fast paths (part 2): repeated fields, maps, oneofs

hashgraph/pbj#1000 · opened

on Oct 30, 2026

Byte array parser with in-order fast paths (part 1): ArrayCursor, routing, singular fields

hashgraph/pbj#998 · opened

on Oct 30, 2026

Parser clean-ups: schema lookup on unknown fields only, direct nested parse calls

hashgraph/pbj#993 · opened

on Oct 29, 2026

Generate straight-line size calculation and stream writers

hashgraph/pbj#990 · opened

on Oct 29, 2026

Generate straight-line byte array writers with direct nested calls

hashgraph/pbj#989 · opened

on Oct 29, 2026

Navigation Menu

Repository navigation

Issues

Document what the fast code generation work learned about performance

Array parser: cut the per-nested-message cost of decoding

Views: cut the per-child cost of wrap and index

Writers: cut the per-nested-message cost of open and close

Synthetic HAPI block workload, benchmarks and reference codec matching the external block-stream workload

JSON codec writes and parses uint32/fixed32 >= 2^31 and uint64/fixed64 >= 2^63 as negative numbers

Byte-array writer writes corrupt wrapper messages for FloatValue/DoubleValue -0.0 and UInt32Value >= 2^31

A oneof message alternative set to null round-trips as the default instance

float and double fields set to -0.0 are not written, unlike protoc

Generated code does not compile for map fields with enum values

Zero-allocation flyweight writers (SBE-style encoders) for whole message trees

Unboxed oneof storage, part 1: storage, parsers and writers

Keep every generated method under HotSpot's 8,000-byte compile limit

JSON writer mis-indents a map field inside a nested message

Generated hashCode() repeats earlier fields' code before a map field

Unknown fields inside map entries always fail to parse with "Bad tag"

Parse string fields without per-call decoder and buffer allocations

Zero-copy views (part 2): oneofs, repeated fields and wrapper types

Zero-copy, zero-allocation views over encoded messages (part 1)

Reverse writer into a reusable buffer (fast_write)

Byte array parser with in-order fast paths (part 2): repeated fields, maps, oneofs

Byte array parser with in-order fast paths (part 1): ArrayCursor, routing, singular fields

Parser clean-ups: schema lookup on unknown fields only, direct nested parse calls

Generate straight-line size calculation and stream writers

Generate straight-line byte array writers with direct nested calls