regexlib ships as a single, dependency-free header. The Unicode tables it needs are generated into regexlib.h from the Unicode Character Database (UCD) by the scripts here — there is no runtime or build-time dependency on any other library.

The data (property tables + the version-dependent GraphemeBreak / Emoji / Incb enum value lists) is generated. The algorithms that consume it — the UTF-8 codec and the UAX #29 grapheme break rules (is_grapheme_boundary) — are hand-maintained C++ in regexlib.h, outside the generated block. Algorithms are not auto-derived from the spec; only the property data they read is.

The current target is recorded in ucd/VERSION (Unicode 17.0.0) and pinned by SHA-256 in ucd/SHA256SUMS. The raw *.txt inputs are large and reproducible, so they are not committed (ucd/.gitignore); fetch them on demand.

python3 tools/fetch_ucd.py                  # download + verify pinned inputs
python3 tools/generate_ucd.py verify        # parse sanity checks
python3 tools/generate_ucd.py update regexlib.h   # splice the generated block

The generated block lives between // [BEGIN GENERATED UCD BLOCK] and // [END GENERATED UCD BLOCK] in regexlib.h.

1. Edit ucd/VERSION.
2. python3 tools/fetch_ucd.py --refresh-sums (re-pins SHA256SUMS; review the diff).
3. python3 tools/generate_ucd.py update regexlib.h.
4. Review the UAX #29 rule skeleton (is_grapheme_boundary in regexlib.h) against the new UAX #29 revision history — new rules (rare) are hand-applied; new property values come through automatically via the regenerated enums/tables.
5. Run the tests, including the grapheme oracle below.

• tools/generate_ucd.py verify — parse-level sanity.
• Full-range self-check — every regeneration re-derives each property directly from the parsed UCD and asserts the emitted table matches across all code points (General_Category, Grapheme_Cluster_Break, Emoji, White_Space, and simple case folding). Code points that have only a full fold (ß, İ, …) have no simple fold and return the code point unchanged.
• Official UAX #29 test — ucd/auxiliary/GraphemeBreakTest.txt passes 766/766 through reg::unicode::is_grapheme_boundary.
• The regexlib test/fuzz/corpus suites pass with the generated tables.

• check_doc_identifiers.py — verifies that backtick identifiers in docs/*.md / README.md still exist in regexlib.h / test/*.cc, so a rename cannot silently rot the docs. Run as just lint-docs; CI runs it on every push.