ucd-tools provides compiler-oriented tooling for dealing with Unicode
properties and normalization in multiple source languages.
The Unicode Character Database is an incredibly useful and somewhat frustrating resource. The official state of the UCD is spread between multiple files having inconsistent formats. The XML version of the UCD omits the state of some newer properties, and also experimental properties. There is no library (not even ICU) that provides comprehensive support for all Unicode properties, and no straightforward way to implement locally-defined custom properties. If the International Components for Unicode can solve the problem in whatever source language you are using, it's definitely the right solution to adopt. Unfortunately that leaves most source languages without native support.
Where ICU aims at broad support of Unicode, many applications in practice adopt simplifying policies. Examples include:
- External and internal encodings are all done in UTF-8. No conversion support is required.
- A particular normalization form (typically NFC) is chosen for input; the application may validate normalization, but does not perform normalization.
- Sorting is only performed internally, and is not script- or locale-aware.
- All bidiretional text is handled in logical order; no bidirectional support is required.
These programs might be viewed as providing "arm's length" Unicode support, but there are some important applications in this category - most notably compilers. Compilers are also unusual because
- A Unicode-aware programming language is generally specified with respect to some particular version of Unicode. It is not a goal (in fact, it is an anti-goal) to always track the latest version of the Unicode standard.
- A compiler implementation really wants to be able to access Unicode tables and algorithms in the native programming language for reasons of text stream processing efficiency.
Enter ucd-tools.
ucd-tools consists of four parts:
-
A collection of XML files. Primarily the XML-formatted UCD database, but also some additional XML files that define properties not included in the XML version of the UCD. The XML files also supply property aliases and property value aliases.
These files, along with any additional XML files that define local, custom properties, will be digested by the first-stage tool.
-
A property compiler that digests the XML files and produces a monolithic C++ source file that contains all of this information in a single, monolithic file.
The property compiler also performs various validity and cross-version stability checks.
-
A property generator that produces compact property tables and supporting type definitions for the various Unicode and locally-defined properties and their values. The property generator can be extended to support additional source language targets in a modular way.
These tables depend in turn on a small, language-specific support library that is separately implemented for each target language.
-
A support library for each of the supported source languages. The support library provides:
-
Language-specific type definition for Unicode code points.
This typically is not the same as the language's character datatype, because many source languages have adopted a 16-bit character type, and Unicode code points range over 21-bit unsigned values.
-
A data structure that implements closed code point ranges.
-
A specialized set implementation for code points based on the closed code point range type.
Unicode code point sets often consist of sparse, dense ranges, so it is useful to be able to iterate by range and to store the set using a dense representation.
-
Validation algorithms for NFC-encoded input.
-
[Possible, Future:] Implementations of case conversion and case-aware string comparison.
-
The general idea is to minimize the amount of code that has to be written
for a new source language, and use table-driven algorithms whose tables can
be easily updated as the Unicode version advances. Because the ucd-tools
client program requires support for a specific version of Unicode, it
is assumed that the generated tables and the support library are likely
to be statically linked into the client binary.