[nciric]

Fixes #215

Summary

Provides an ICU Data Build Tool filter configuration template (icu_filters.json.in) paired with a generator script (tools/scripts/generate_icu_filters.py), along with UCONFIG_* compile-time flags to minimize both ICU4C's data and code footprint for Unicode Inflection.

Changes

1. icu_filters.json.in + tools/scripts/generate_icu_filters.py + .gitignore:
   • Populates localeFilter.includelist dynamically from resources/org/unicode/inflection/locale/supported-locales.properties so there is a single source of truth for supported locales.
   • Strips unused ICU data trees (coll_tree, coll_ucadata, conversion_mappings, translit, zone_tree, zone_supplemental, unit_tree, lang_tree, region_tree, curr_supplemental, confusables, stringprep, unames, and non-word break iterator rules/dictionaries/LSTM/adaboost models) and retains only the resources required by Inflection.
2. .github/workflows/cmake-multi-platform.yml:
   • Generates icu_filters.json via tools/scripts/generate_icu_filters.py and builds ICU 79.1 on both ubuntu-latest and macos-latest with ICU_DATA_FILTER_FILE and CPPFLAGS="-DUCONFIG_NO_COLLATION=1 -DUCONFIG_NO_IDNA=1 -DUCONFIG_NO_LEGACY_CONVERSION=1 -DUCONFIG_NO_TRANSLITERATION=1 -DUCONFIG_USE_ML_PHRASE_BREAKING=1 -DUCONFIG_NO_FILTERED_BREAK_ITERATION=1 -DUCONFIG_NO_SERVICE=1".
   • Uses NPROC=$(getconf _NPROCESSORS_ONLN) on both Linux and macOS.
3. test/src/inflection/util/LocaleUtilsTest.cpp (LocaleUtilsTest#testMinimalICUData):
   • Dedicated unit test verifying that every supported locale in LocaleUtils::getSupportedLocaleList() has all ICU data and APIs required by Inflection (ures_open, ulocdata_getExemplarSet, ListFormatter, RuleBasedNumberFormat(URBNF_SPELLOUT), NumberFormatter, SimpleDateFormat, PluralRules, BreakIterator::createWordInstance, Normalizer2, CaseMap, and ubidi_*).
4. Documentation (README.md, documents/how_to_add_new_language.md, documents/how_to_update_icu.md):
   • Documents how to generate icu_filters.json and build a minimal ICU4C with both ICU_DATA_FILTER_FILE and UCONFIG_* flags.

Size Comparison (ICU 79.1 on Linux x86_64)

Library	Default Unfiltered (Stripped)	Filtered + UCONFIG_* (Stripped)	Reduction
libicudata.so.79.1	~32.3 MB	1.62 MiB (~310 KB compressed)	-95.0%
libicui18n.so.79.1	3.52 MiB	2.86 MiB	-18.8%
libicuuc.so.79.1	2.19 MiB	1.98 MiB	-9.6%
Total (data + i18n + uc)	~38.0 MiB	6.46 MiB	-83.0%

[grhoten]

Dedicated unit test verifying that every supported locale in LocaleUtils::getSupportedLocaleList() has all ICU data and APIs required by Inflection (ures_open, ulocdata_getExemplarSet, ListFormatter, RuleBasedNumberFormat(URBNF_SPELLOUT), NumberFormatter, SimpleDateFormat, PluralRules, BreakIterator::createWordInstance, Normalizer2, CaseMap, and ubidi_*).

I think your AI agent is hallucinating. This code does not use ubidi_*, nor does the new test test it. The changes seem fine otherwise.

I am curious to see what data is left, but it's not a blocking request for merging.

[grhoten]

Oops! I made a bad suggestion. I recommend removing the -DUCONFIG_USE_ML_PHRASE_BREAKING=1 part.