Techniques

Frozen values are positional. The schema stays in the program. Dict keys are sorted by their UTF-8 bytes so two encodes of the same mapping match. Floats keep every bit, including the sign of zero and the payload of a NaN.

The tensor directory is one Frozen value. Its magic is KLNT and its version byte is 2. Descriptors are 128 bytes. Payloads start at a multiple of 64 from the start of the directory. A borrowed read returns a slice of the caller's buffer. Numeric payloads are copied with one memcpy on a little-endian host. Only the header and alignment gaps are zeroed; payload bytes are not cleared before that copy.

list<f32>, list<f64>, list<i32>, list<u32>, list<i64>, list<int>, list<u64>, and list<uint> accept a primitive slice (f32s, f64s, i32s, u32s, i64s, u64s). The element width is the schema width: a u32 slice is four bytes and is not written as eight-byte integers. On decode, an aligned little-endian body is borrowed from the input. Pass ListReuse in Options.reuse to copy into one buffer that is kept across calls instead of allocating on every decode. Writer.reset keeps the allocation and does not wipe the previous payload.

A count is checked against the unread bytes before Kiln allocates storage for the elements. The default cap is 16,777,216 counts and 16 MiB for one string or byte run. Pass cap on Options to raise it.

A frame is sixteen bytes: KLN0, four zero bytes, and the FNV-1a 64 hash of the canonical type text, then one Frozen value. A hash that does not match the schema the caller passed is SCHEMA.

Writer.reset clears the length and keeps the existing allocation. Call encodeTo or encodeTensorsTo in a loop when the same process encodes many records. Generated structs write fields directly into that buffer, with no per-field allocation.