Canonical serialization¶

ELANG defines a byte-deterministic serialization so that two semantically equal documents always produce identical bytes. This makes packages reproducible and content-addressable. The ELD canonical rules below are locked; the EPF and integrity-manifest canonical profiles have their own ratified serializers.

Ordering¶

Input declaration order is unrestricted — the canonical writer imposes a fixed order:

• Groups: consonants, then vowels, then custom groups sorted ASCII bytewise.
• Roots within a group: sorted by NFC IPA identifier, UTF-8 bytewise.
• Allophones within a root: sorted by NFC IPA identifier, UTF-8 bytewise.
• Root child order: roman, then when (if any), then allophones.
• Allophone child order: explicit roman (if any), then when (if any).

A fieldless allophone is emitted as a bare leaf with no braces; an allophone with fields gets a nonempty block.

Conditions¶

Inside when, group, and or, direct child expressions are sorted recursively by the Canonical Condition Sorting Key v1 (below). The writer preserves tree structure and duplicate predicates — there is no boolean simplification, no flattening, and references are not rewritten away from their owning roots.

Formatting¶

• 4 ASCII spaces per nesting level; no tabs.
• One declaration per line; zero blank lines.
• LF line endings only, with exactly one final LF.
• All comments are omitted from canonical output (source may use comments and any KDL 2.0 whitespace).
• Strings are always standard double-quoted, escaping embedded " and \. Permitted Unicode is preserved directly as UTF-8, and authored ASCII romanization case is preserved.

Because the characters rejected by the Unicode rules (controls, Cf, Co, Cn/noncharacters) can never appear in a valid value, no canonical control-escape form is needed. Validated NFC strings are preserved exactly — never silently re-normalized.

Canonical Condition Sorting Key v1¶

Each condition node is reduced to a binary key, and keys are compared as unsigned lexicographic byte strings.

Node ranks (u8):

Key formats:

atomic_key    = rank:u8
             || value_utf8_length:u16BE
             || exact_validated_value_utf8

composite_key = rank:u8
             || child_count:u16BE
             || repeated( sorted_child_key_length:u32BE || sorted_child_key )

Children are sorted recursively before their parent's key is built. The key discipline applies to the children of when (AND), group (AND), and or (two operands). There is no reordering across semantically different parent levels and no alias rewriting. when itself has no key rank because it is a root expression container.

Idempotence¶

Canonicalization is idempotent:

canon(canon(x)) == canon(x)

Equivalent inputs (for example, the same inventory with groups declared in a different order) produce identical canonical bytes, while genuinely different condition trees stay different.