v1.4.0

Highlights

serialized_size — exact streaming length without producing bytes (#22)

serialized_size / serialized_size_with_options compute the exact number of bytes to_writer_streaming will emit for a value, without producing them. They drive the same encoder as the streaming serializer through a counting sink, so the measurement is correct by construction and cannot drift from what is actually written.

This enables single-pass, zero-intermediate-buffer length-prefixed framing over a non-seekable transport (e.g. a socket): measure the body, write the length prefix, then stream the body once. The measure is O(1) for serialize_bytes and TypedSlice<T> bodies and O(N) for a bare numeric Vec<T>; unknown-length containers are rejected exactly as to_writer_streaming rejects them.

Bulk-slice fast path for typed numeric slices (#21)

TypedSlice<T>, to_writer_typed_slice, and the closed-form typed_slice_size give O(1) encoding and measuring of a contiguous numeric slice: a typed-array header, a SIZE prefix, and a single bulk write_all of the payload on little-endian targets (with a per-element fallback on big-endian).

Other changes

• The field module is now public (pub mod field), surfacing its JSON Pointer (RFC 6901) reference documentation and examples. Additive — from_field / from_field_slice / skip_value remain re-exported at the crate root.
• Added benches/serialized_size.rs (Criterion) validating the O(1)/O(N) cost model.
• Documentation and rustdoc cleanup; cargo doc -D warnings is clean.

All changes are additive — no removals or signature changes — so this is a minor release.