blob: 3d07e2588e468370fc6a38af2e701dbba9962f50 [file] [log] [blame]
// Copyright 2016 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
// +build icu
package cases
import (
func TestICUConformance(t *testing.T) {
// Build test set.
input := []string{
"a.a a_a",
"a\u00ad\u00ada", // Format
"a''a", // SingleQuote
"a::a", // MidLetter
"a..a", // MidNumLet
"a;;a", // MidNum
"a__a", // ExtendNumlet
add := func(x interface{}) {
switch v := x.(type) {
case string:
input = append(input, v)
case []string:
for _, s := range v {
input = append(input, s)
for _, tc := range testCases {
for _, tc := range bufferTests {
for _, tc := range breakTest {
add(strings.Replace(tc, "|", "", -1))
for _, tc := range foldTestCases {
// Compare ICU to Go.
for _, c := range []string{"lower", "upper", "title", "fold"} {
for _, tag := range []string{
"und", "af", "az", "el", "lt", "nl", "tr",
} {
for _, s := range input {
if exclude(c, tag, s) {
testtext.Run(t, path.Join(c, tag, s), func(t *testing.T) {
want := doICU(tag, c, s)
got := doGo(tag, c, s)
if norm.NFC.String(got) != norm.NFC.String(want) {
t.Errorf("\n in %[3]q (%+[3]q)\n got %[1]q (%+[1]q)\n want %[2]q (%+[2]q)", got, want, s)
// exclude indicates if a string should be excluded from testing.
func exclude(cm, tag, s string) bool {
list := []struct{ cm, tags, pattern string }{
// TODO: Go does not handle certain esoteric breaks correctly. This will be
// fixed once we have a real word break iterator. Alternatively, it
// seems like we're not too far off from making it work, so we could
// fix these last steps. But first verify that using a separate word
// breaker does not hurt performance.
{"title", "af nl", "a''a"},
{"", "", "א'a"},
// All the exclusions below seem to be issues with the ICU
// implementation (at version 57) and thus are not marked as TODO.
// ICU does not handle leading apostrophe for Dutch and
// Afrikaans correctly. See
{"title", "af nl", "'n"},
{"title", "af nl", "'N"},
// Go terminates the final sigma check after a fixed number of
// ignorables have been found. This ensures that the algorithm can make
// progress in a streaming scenario.
{"lower title", "", "\u039f\u03a3...............................a"},
// This also applies to upper in Greek.
// NOTE: we could fix the following two cases by adding state to elUpper
// and aztrLower. However, considering a modifier to not belong to the
// preceding letter after the maximum modifiers count is reached is
// consistent with the behavior of unicode/norm.
{"upper", "el", "\u03bf" + strings.Repeat("\u0321", 29) + "\u0313"},
{"lower", "az tr lt", "I" + strings.Repeat("\u0321", 30) + "\u0307\u0300"},
{"upper", "lt", "i" + strings.Repeat("\u0321", 30) + "\u0307\u0300"},
{"lower", "lt", "I" + strings.Repeat("\u0321", 30) + "\u0300"},
// ICU title case seems to erroneously removes \u0307 from an upper case
// I unconditionally, instead of only when lowercasing. The ICU
// transform algorithm transforms these cases consistently with our
// implementation.
{"title", "az tr", "\u0307"},
// The spec says to remove \u0307 after Soft-Dotted characters. ICU
// transforms conform but ucasemap_utf8ToUpper does not.
{"upper title", "lt", "i\u0307"},
{"upper title", "lt", "i" + strings.Repeat("\u0321", 29) + "\u0307\u0300"},
// Both Unicode and CLDR prescribe an extra explicit dot above after a
// Soft_Dotted character if there are other modifiers.
// ucasemap_utf8ToUpper does not do this; ICU transforms do.
// The issue with ucasemap_utf8ToUpper seems to be that it does not
// consider the modifiers that are part of composition in the evaluation
// of More_Above. For instance, according to the More_Above rule for lt,
// a dotted capital I (U+0130) becomes i\u0307\u0307 (an small i with
// two additional dots). This seems odd, but is correct. ICU is
// definitely not correct as it produces different results for different
// normal forms. For instance, for an İ:
// \u0130 (NFC) -> i\u0307 (incorrect)
// I\u0307 (NFD) -> i\u0307\u0307 (correct)
// We could argue that we should not add a \u0307 if there already is
// one, but this may be hard to get correct and is not conform the
// standard.
{"lower title", "lt", "\u0130"},
{"lower title", "lt", "\u00cf"},
// We are conform ICU ucasemap_utf8ToUpper if we remove support for
// elUpper. However, this is clearly not conform the spec. Moreover, the
// ICU transforms _do_ implement this transform and produces results
// consistent with our implementation. Note that we still prefer to use
// ucasemap_utf8ToUpper instead of transforms as the latter have
// inconsistencies in the word breaking algorithm.
{"upper", "el", "\u0386"}, // GREEK CAPITAL LETTER ALPHA WITH TONOS
{"upper", "el", "\u0389"}, // GREEK CAPITAL LETTER ETA WITH TONOS
{"upper", "el", "\u038A"}, // GREEK CAPITAL LETTER IOTA WITH TONOS
{"upper", "el", "\u0391"}, // GREEK CAPITAL LETTER ALPHA
{"upper", "el", "\u0397"}, // GREEK CAPITAL LETTER ETA
{"upper", "el", "\u0399"}, // GREEK CAPITAL LETTER IOTA
{"upper", "el", "\u03AC"}, // GREEK SMALL LETTER ALPHA WITH TONOS
{"upper", "el", "\u03AE"}, // GREEK SMALL LETTER ALPHA WITH ETA
{"upper", "el", "\u03AF"}, // GREEK SMALL LETTER ALPHA WITH IOTA
{"upper", "el", "\u03B1"}, // GREEK SMALL LETTER ALPHA
{"upper", "el", "\u03B7"}, // GREEK SMALL LETTER ETA
{"upper", "el", "\u03B9"}, // GREEK SMALL LETTER IOTA
for _, x := range list {
if != "" && strings.Index(, cm) == -1 {
if x.tags != "" && strings.Index(x.tags, tag) == -1 {
if strings.Index(s, x.pattern) != -1 {
return true
return false
func doGo(tag, caser, input string) string {
var c Caser
t := language.MustParse(tag)
switch caser {
case "lower":
c = Lower(t)
case "upper":
c = Upper(t)
case "title":
c = Title(t)
case "fold":
c = Fold()
return c.String(input)