Changelog

All notable changes to this project are documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

0.3.0 - 2026-08-12

from_list/2 silently dropped the empty word: DawgEx.from_list([""]) built a DAWG that answered member?(dawg, "") with false. The empty word is now a valid entry that costs no edges.
Internal DawgEx.Builder typespecs drifted from the runtime values: stack() described a frame's third element as a list of nested stacks where it holds the node's accumulated child edges (now [edge()]), and t() described a bare map that its required associations kept the %DawgEx.Builder{} struct — and the initial empty register/nodes maps — from inhabiting (now a struct type with optional(...) keys). Typespec-only, no behavior change.

Breaking: the header now mirrors an edge's shape and is incompatible with binaries built by 0.2.0. The root offset spends offset_width bits instead of offset_width + 2, and the two freed bits carry flags — <<offset_width::8, root_offset::size(offset_width), empty?::1, 0::1>> — where empty? records whether the empty word is a member and the final bit is always clear.

Breaking: the binary encoding is denser and incompatible with binaries built by 0.1.0. An edge's two flag bits are packed into the same bytes as its child offset instead of a padded trailing byte, and all nodes without outgoing edges share a single sentinel slot at offset 0 rather than each occupying a placeholder edge. Consequently offset_width must now satisfy rem(offset_width + 2, 8) == 0 (6, 14, 22, …) instead of being a multiple of 8, and the default is 14 — 3 bytes per edge with room for 16_384 edges, down from 16 — 4 bytes per edge with room for 65_536.

DawgEx.from_list/2 — builds a minimal DAWG from a list of binaries and encodes it as a flat binary. Takes an optional offset_width (a positive multiple of 8, defaulting to 16) that sets how many bits each edge spends addressing its child, trading encoded size against the number of edges the automaton can hold. Raises ArgumentError rather than emitting a binary whose offsets wrap when the width is too narrow for the word list.
DawgEx.member?/2 — set-membership query directly against the encoded binary, with no intermediate decoding step. Reads the offset width from the binary's header, so the build-time width does not have to be threaded back in.