Roadmap

The live status of each milestone. The milestones and their exit criteria come from design.md ยง11; the GitHub milestones and issues are the source of truth for open work. Last updated 2026-09-29, for 0.2.0.

Milestone Status Open work
M0: skeleton and CI Done The packages are published to nuget.org from the maintainer's account, since 0.1.1.
M1: schemas Done
M2: binary encoding and the generic model Done
M2.5: code generation from .avsc files Done Union classes (#14), required/init (#15), cyclic cross-file references (#16), options (#18), JSON for generated types (#19), random-schema tests (#141)
M3: schema resolution Done
M4: attribute generator and AvroSerializer<T> Not started #31
M5: container files, codecs, single-object encoding Done
M6: CLI tool and protocols Started The avrosharp tool (gen, schema canonical, schema fingerprint) ships in 0.2.0; protocols are open (#33)
M7: hardening and 1.0 Started #35: API review (#73), docs site and samples (#74, #126), public API before 1.0 (#134), CI consuming the packages (#136), dev container (#139). Done: package metadata (#66), coverage (#75), release workflows (#76), nightly fuzzing (#34)
Integrations Started #77: Confluent (#79). Done: schema references (#80), registry wire framing (#81) and the field walker (#82). Remaining core work: the typed serializer lookup (#31)

Done, in more detail

  • Schemas: parsing (System.Text.Json), writing, Parsing Canonical Form, CRC-64-AVRO/MD5/SHA-256 fingerprints, all Avro 1.12 logical types. Checked against Apache's schema-tests.txt.
  • Encoding: AvroWriter/AvroReader over spans, IBufferWriter<byte> and ReadOnlySequence<byte>. The generic model (AvroValue, GenericRecord) is binary and JSON, with schema resolution.
  • Code generation: records, enums and fixed types with direct serializers, the Apache.Avro compatibility mode, logical types, and resolution for generated types, from the source generator or the avrosharp tool.
  • Files and messages:
    • schema-registry wire framing (Confluent, Apicurio, AWS Glue) and schema references;
    • container files, sync and async, with seeking and splitting;
    • every codec: null and deflate built in, and snappy, zstandard, bzip2 and xz in AvroSharp.Codecs;
    • single-object encoding;
    • streams of objects without a container (AvroSharp.Streams).
  • Verification:
    • property tests against Apache.Avro C#;
    • files written by Apache Avro Java for every codec;
    • fuzz targets, run as a smoke test on every build;
    • a Native AOT smoke test.
  • Hardening (0.2.0): bounded nesting and memory for hostile schemas and data (#129), resolution that follows Java where the specification leaves a choice (#130), code generation fixes from a review with random names and defaults (#131), and ordered assertions, coverage checks, and Java- and Glue-written reference data in the tests (#133).

See CHANGELOG.md for the details of each change.

Edit this page