Decode common binary metadata

Choose a type from the format's storage definition. Equal byte widths do not imply equal meanings: a checksum, fixed-point number, and integer can each occupy four bytes. The inspector's schemas use these types for outer headers and metadata; compressed or encrypted content still requires a separate decoder.

Storage	CStruct spelling	Executable example
Three-byte integer	uint24, int24, with optional < or >	24-bit fields
Encoded text of a known byte length	utf8, latin1, cp437, utf16le, utf16be arrays	Bounded encodings
Signed/unsigned LEB128	sleb128_32, sleb128_64, uleb128_32, uleb128_64	Variable integers
Binary-scaled integer	fixed16_16, ufixed16_16, fixed2_30, ufixed8_8	Fixed point
16-byte identifier	uuid for network order; guid for Windows field order	Identifier order
Discriminator-dependent fields	if/else and switch	Conditional records

Preserve storage boundaries

Three-byte integers and identifiers have alignment 1. LEB128 also has alignment 1, but its byte width is determined by the encoded value. Use a stream-backed address query for a field following a varint. LEB128 writers emit canonical encodings; an in-place update rejects a replacement whose encoded width differs.

Bounded text array counts are bytes, including for UTF-16. For example, utf16le name[4] can hold one supplementary character represented by two UTF-16 code units. Existing wchar[4] reserves four code units, or eight bytes. Decoders reject incomplete sequences at the field boundary. Writers reject unmappable or oversized text and zero-pad shorter text. See text encodings.

Fixed-point values use Double and must lie exactly on their storage grid. For example, unsigned 8.8 represents 0.5 exactly as raw integer 128; 0.1 requires rounding and is rejected. These values cannot serve as integer counts, enum backing types, or bitfield storage. UUID/GUID values use System.Guid in managed code and canonical strings in browser JSON. Neither is an integer count.

Let each record select its own fields

Learn when each decision runs and which values it can use in Choose fields with if and switch, with three focused browser lessons.

The conditional-record example uses one schema for text and numeric records. Only the active fields consume bytes or appear in parsed results and debug ranges. Selecting an inactive field raises a path error. Names must remain unique even in mutually exclusive arms; named inline structs make larger alternatives easier to navigate.

Case labels are compiled constants. Predicates can use earlier fields and caller variables; unavailable active expressions fail. Each group preserves its entry environment, and each array element selects independently. See the grammar and scope rules.

Serialization chooses branches from the supplied values and rejects supplied inactive members. An update cannot change the active storage plan, even if two branches have equal byte widths. Serialize a new buffer for that change.

See the primitive reference for ranges, endianness, representation, and supported combinations.