Update dependencies to K8s 1.8

vendor/golang.org/x/text/language/common.go

@@ -1,4 +1,4 @@
-// This file was generated by go generate; DO NOT EDIT
+// Code generated by running "go generate" in golang.org/x/text. DO NOT EDIT.
 
 package language
 

vendor/golang.org/x/text/language/gen.go

@@ -678,6 +678,8 @@ func (b *builder) parseIndices() {
 	b.locale.parse(meta.DefaultContent.Locales)
 }
 
+// TODO: region inclusion data will probably not be use used in future matchers.
+
 func (b *builder) computeRegionGroups() {
 	b.groups = make(map[int]index)
 
@@ -686,6 +688,11 @@ func (b *builder) computeRegionGroups() {
 		b.groups[i] = index(len(b.groups))
 	}
 	for _, g := range b.supp.TerritoryContainment.Group {
+		// Skip UN and EURO zone as they are flattening the containment
+		// relationship.
+		if g.Type == "EZ" || g.Type == "UN" {
+			continue
+		}
 		group := b.region.index(g.Type)
 		if _, ok := b.groups[group]; !ok {
 			b.groups[group] = index(len(b.groups))
@@ -782,6 +789,7 @@ func (b *builder) writeLanguage() {
 	lang.updateLater("tw", "twi")
 	lang.updateLater("nb", "nob")
 	lang.updateLater("ak", "aka")
+	lang.updateLater("bh", "bih")
 
 	// Ensure that each 2-letter code is matched with a 3-letter code.
 	for _, v := range lang.s[1:] {
@@ -798,10 +806,10 @@ func (b *builder) writeLanguage() {
 		}
 	}
 
-	// Complete canonialized language tags.
+	// Complete canonicalized language tags.
 	lang.freeze()
 	for i, v := range lang.s {
-		// We can avoid these manual entries by using the IANI registry directly.
+		// We can avoid these manual entries by using the IANA registry directly.
 		// Seems easier to update the list manually, as changes are rare.
 		// The panic in this loop will trigger if we miss an entry.
 		add := ""
@@ -908,7 +916,7 @@ func (b *builder) writeRegion() {
 			i := b.region.index(s)
 			for _, d := range e.description {
 				if strings.Contains(d, "Private use") {
-					regionTypes[i] = iso3166UserAssgined
+					regionTypes[i] = iso3166UserAssigned
 				}
 			}
 			regionTypes[i] |= bcp47Region
@@ -1065,7 +1073,7 @@ const (
 )
 
 const (
-	iso3166UserAssgined = 1 << iota
+	iso3166UserAssigned = 1 << iota
 	ccTLD
 	bcp47Region
 )
@@ -1355,42 +1363,23 @@ func (b *builder) writeLikelyData() {
 
 type mutualIntelligibility struct {
 	want, have uint16
-	conf       uint8
+	distance   uint8
 	oneway     bool
 }
 
 type scriptIntelligibility struct {
-	lang       uint16 // langID or 0 if *
-	want, have uint8
-	conf       uint8
+	wantLang, haveLang     uint16
+	wantScript, haveScript uint8
+	distance               uint8
+	// Always oneway
 }
 
-type sortByConf []mutualIntelligibility
-
-func (l sortByConf) Less(a, b int) bool {
-	return l[a].conf > l[b].conf
-}
-
-func (l sortByConf) Swap(a, b int) {
-	l[a], l[b] = l[b], l[a]
-}
-
-func (l sortByConf) Len() int {
-	return len(l)
-}
-
-// toConf converts a percentage value [0, 100] to a confidence class.
-func toConf(pct uint8) uint8 {
-	switch {
-	case pct == 100:
-		return 3 // Exact
-	case pct >= 90:
-		return 2 // High
-	case pct > 50:
-		return 1 // Low
-	default:
-		return 0 // No
-	}
+type regionIntelligibility struct {
+	lang     uint16 // compact language id
+	script   uint8  // 0 means any
+	group    uint8  // 0 means any; if bit 7 is set it means inverse
+	distance uint8
+	// Always twoway.
 }
 
 // writeMatchData writes tables with languages and scripts for which there is
@@ -1400,13 +1389,50 @@ func toConf(pct uint8) uint8 {
 // We also drop all region-related data as we use a different algorithm to
 // determine region equivalence.
 func (b *builder) writeMatchData() {
+	lm := b.supp.LanguageMatching.LanguageMatches
+	cldr.MakeSlice(&lm).SelectAnyOf("type", "written_new")
+
+	regionHierarchy := map[string][]string{}
+	for _, g := range b.supp.TerritoryContainment.Group {
+		regions := strings.Split(g.Contains, " ")
+		regionHierarchy[g.Type] = append(regionHierarchy[g.Type], regions...)
+	}
+	regionToGroups := make([]uint8, len(b.region.s))
+
+	idToIndex := map[string]uint8{}
+	for i, mv := range lm[0].MatchVariable {
+		if i > 6 {
+			log.Fatalf("Too many groups: %d", i)
+		}
+		idToIndex[mv.Id] = uint8(i + 1)
+		// TODO: also handle '-'
+		for _, r := range strings.Split(mv.Value, "+") {
+			todo := []string{r}
+			for k := 0; k < len(todo); k++ {
+				r := todo[k]
+				regionToGroups[b.region.index(r)] |= 1 << uint8(i)
+				todo = append(todo, regionHierarchy[r]...)
+			}
+		}
+	}
+	b.writeSlice("regionToGroups", regionToGroups)
+
 	b.writeType(mutualIntelligibility{})
 	b.writeType(scriptIntelligibility{})
-	lm := b.supp.LanguageMatching.LanguageMatches
-	cldr.MakeSlice(&lm).SelectAnyOf("type", "written")
+	b.writeType(regionIntelligibility{})
 
-	matchLang := []mutualIntelligibility{}
+	matchLang := []mutualIntelligibility{{
+		// TODO: remove once CLDR is fixed.
+		want:     uint16(b.langIndex("sr")),
+		have:     uint16(b.langIndex("hr")),
+		distance: uint8(5),
+	}, {
+		want:     uint16(b.langIndex("sr")),
+		have:     uint16(b.langIndex("bs")),
+		distance: uint8(5),
+	}}
 	matchScript := []scriptIntelligibility{}
+	matchRegion := []regionIntelligibility{}
 	// Convert the languageMatch entries in lists keyed by desired language.
 	for _, m := range lm[0].LanguageMatch {
 		// Different versions of CLDR use different separators.
@@ -1414,33 +1440,38 @@ func (b *builder) writeMatchData() {
 		supported := strings.Replace(m.Supported, "-", "_", -1)
 		d := strings.Split(desired, "_")
 		s := strings.Split(supported, "_")
-		if len(d) != len(s) || len(d) > 2 {
-			// Skip all entries with regions and work around CLDR bug.
+		if len(d) != len(s) {
+			log.Fatalf("not supported: desired=%q; supported=%q", desired, supported)
 			continue
 		}
-		pct, _ := strconv.ParseInt(m.Percent, 10, 8)
-		if len(d) == 2 && d[0] == s[0] && len(d[1]) == 4 {
-			// language-script pair.
-			lang := uint16(0)
-			if d[0] != "*" {
-				lang = uint16(b.langIndex(d[0]))
+		distance, _ := strconv.ParseInt(m.Distance, 10, 8)
+		switch len(d) {
+		case 2:
+			if desired == supported && desired == "*_*" {
+				continue
 			}
+			// language-script pair.
 			matchScript = append(matchScript, scriptIntelligibility{
-				lang: lang,
-				want: uint8(b.script.index(d[1])),
-				have: uint8(b.script.index(s[1])),
-				conf: toConf(uint8(pct)),
+				wantLang:   uint16(b.langIndex(d[0])),
+				haveLang:   uint16(b.langIndex(s[0])),
+				wantScript: uint8(b.script.index(d[1])),
+				haveScript: uint8(b.script.index(s[1])),
+				distance:   uint8(distance),
 			})
 			if m.Oneway != "true" {
 				matchScript = append(matchScript, scriptIntelligibility{
-					lang: lang,
-					want: uint8(b.script.index(s[1])),
-					have: uint8(b.script.index(d[1])),
-					conf: toConf(uint8(pct)),
+					wantLang:   uint16(b.langIndex(s[0])),
+					haveLang:   uint16(b.langIndex(d[0])),
+					wantScript: uint8(b.script.index(s[1])),
+					haveScript: uint8(b.script.index(d[1])),
+					distance:   uint8(distance),
 				})
 			}
-		} else if len(d) == 1 && d[0] != "*" {
-			if pct == 100 {
+		case 1:
+			if desired == supported && desired == "*" {
+				continue
+			}
+			if distance == 1 {
 				// nb == no is already handled by macro mapping. Check there
 				// really is only this case.
 				if d[0] != "no" || s[0] != "nb" {
@@ -1448,28 +1479,57 @@ func (b *builder) writeMatchData() {
 				}
 				continue
 			}
+			// TODO: consider dropping oneway field and just doubling the entry.
 			matchLang = append(matchLang, mutualIntelligibility{
-				want:   uint16(b.langIndex(d[0])),
-				have:   uint16(b.langIndex(s[0])),
-				conf:   uint8(pct),
-				oneway: m.Oneway == "true",
+				want:     uint16(b.langIndex(d[0])),
+				have:     uint16(b.langIndex(s[0])),
+				distance: uint8(distance),
+				oneway:   m.Oneway == "true",
 			})
-		} else {
-			// TODO: Handle other mappings.
-			a := []string{"*;*", "*_*;*_*", "es_MX;es_419"}
-			s := strings.Join([]string{desired, supported}, ";")
-			if i := sort.SearchStrings(a, s); i == len(a) || a[i] != s {
-				log.Printf("%q not handled", s)
+		case 3:
+			if desired == supported && desired == "*_*_*" {
+				continue
 			}
+			if desired != supported { // (Weird but correct.)
+				log.Fatalf("not supported: desired=%q; supported=%q", desired, supported)
+				continue
+			}
+			ri := regionIntelligibility{
+				lang:     b.langIndex(d[0]),
+				distance: uint8(distance),
+			}
+			if d[1] != "*" {
+				ri.script = uint8(b.script.index(d[1]))
+			}
+			switch {
+			case d[2] == "*":
+				ri.group = 0x80 // not contained in anything
+			case strings.HasPrefix(d[2], "$!"):
+				ri.group = 0x80
+				d[2] = "$" + d[2][len("$!"):]
+				fallthrough
+			case strings.HasPrefix(d[2], "$"):
+				ri.group |= idToIndex[d[2]]
+			}
+			matchRegion = append(matchRegion, ri)
+		default:
+			log.Fatalf("not supported: desired=%q; supported=%q", desired, supported)
 		}
 	}
-	sort.Stable(sortByConf(matchLang))
-	// collapse percentage into confidence classes
-	for i, m := range matchLang {
-		matchLang[i].conf = toConf(m.conf)
-	}
+	sort.SliceStable(matchLang, func(i, j int) bool {
+		return matchLang[i].distance < matchLang[j].distance
+	})
 	b.writeSlice("matchLang", matchLang)
+
+	sort.SliceStable(matchScript, func(i, j int) bool {
+		return matchScript[i].distance < matchScript[j].distance
+	})
 	b.writeSlice("matchScript", matchScript)
+
+	sort.SliceStable(matchRegion, func(i, j int) bool {
+		return matchRegion[i].distance < matchRegion[j].distance
+	})
+	b.writeSlice("matchRegion", matchRegion)
 }
 
 func (b *builder) writeRegionInclusionData() {
@@ -1482,6 +1542,11 @@ func (b *builder) writeRegionInclusionData() {
 		containment = make(map[index][]index)
 	)
 	for _, g := range b.supp.TerritoryContainment.Group {
+		// Skip UN and EURO zone as they are flattening the containment
+		// relationship.
+		if g.Type == "EZ" || g.Type == "UN" {
+			continue
+		}
 		group := b.region.index(g.Type)
 		groupIdx := b.groups[group]
 		for _, mem := range strings.Split(g.Contains, " ") {
@@ -1508,7 +1573,6 @@ func (b *builder) writeRegionInclusionData() {
 		for _, v := range l {
 			regionContainment[g] |= 1 << v
 		}
-		// log.Printf("%d: %X", g, regionContainment[g])
 	}
 	b.writeSlice("regionContainment", regionContainment)
 

vendor/golang.org/x/text/language/language.go

@@ -2,7 +2,7 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-//go:generate go run maketables.go gen_common.go -output tables.go
+//go:generate go run gen.go gen_common.go -output tables.go
 //go:generate go run gen_index.go
 
 // Package language implements BCP 47 language tags and related functionality.
@@ -129,8 +129,15 @@ const (
 // specific language or locale. All language tag values are guaranteed to be
 // well-formed.
 type Tag struct {
-	lang     langID
-	region   regionID
+	lang   langID
+	region regionID
+	// TODO: we will soon run out of positions for script. Idea: instead of
+	// storing lang, region, and script codes, store only the compact index and
+	// have a lookup table from this code to its expansion. This greatly speeds
+	// up table lookup, speed up common variant cases.
+	// This will also immediately free up 3 extra bytes. Also, the pVariant
+	// field can now be moved to the lookup table, as the compact index uniquely
+	// determines the offset of a possible variant.
 	script   scriptID
 	pVariant byte   // offset in str, includes preceding '-'
 	pExt     uint16 // offset of first extension, includes preceding '-'
@@ -593,7 +600,7 @@ func (t Tag) Extension(x byte) (ext Extension, ok bool) {
 			return Extension{ext}, true
 		}
 	}
-	return Extension{string(x)}, false
+	return Extension{}, false
 }
 
 // Extensions returns all extensions of t.

vendor/golang.org/x/text/language/match.go

@@ -6,6 +6,16 @@ package language
 
 import "errors"
 
+// A MatchOption configures a Matcher.
+type MatchOption func(*matcher)
+
+// PreferSameScript will, in the absence of a match, result in the first
+// preferred tag with the same script as a supported tag to match this supported
+// tag. The default is currently true, but this may change in the future.
+func PreferSameScript(preferSame bool) MatchOption {
+	return func(m *matcher) { m.preferSameScript = preferSame }
+}
+
 // Matcher is the interface that wraps the Match method.
 //
 // Match returns the best match for any of the given tags, along with
@@ -36,23 +46,44 @@ func Comprehends(speaker, alternative Tag) Confidence {
 // matched tag in t, but is augmented with the Unicode extension ('u')of the
 // corresponding preferred tag. This allows user locale options to be passed
 // transparently.
-func NewMatcher(t []Tag) Matcher {
-	return newMatcher(t)
+func NewMatcher(t []Tag, options ...MatchOption) Matcher {
+	return newMatcher(t, options)
 }
 
 func (m *matcher) Match(want ...Tag) (t Tag, index int, c Confidence) {
 	match, w, c := m.getBest(want...)
-	if match == nil {
-		t = m.default_.tag
-	} else {
+	if match != nil {
 		t, index = match.tag, match.index
+	} else {
+		// TODO: this should be an option
+		t = m.default_.tag
+		if m.preferSameScript {
+		outer:
+			for _, w := range want {
+				script, _ := w.Script()
+				if script.scriptID == 0 {
+					// Don't do anything if there is no script, such as with
+					// private subtags.
+					continue
+				}
+				for i, h := range m.supported {
+					if script.scriptID == h.maxScript {
+						t, index = h.tag, i
+						break outer
+					}
+				}
+			}
+		}
+		// TODO: select first language tag based on script.
+	}
+	if w.region != 0 && t.region != 0 && t.region.contains(w.region) {
+		t, _ = Raw.Compose(t, Region{w.region})
 	}
 	// Copy options from the user-provided tag into the result tag. This is hard
 	// to do after the fact, so we do it here.
-	// TODO: consider also adding in variants that are compatible with the
-	// matched language.
-	// TODO: Add back region if it is non-ambiguous? Or create another tag to
-	// preserve the region?
+	// TODO: add in alternative variants to -u-va-.
+	// TODO: add preferred region to -u-rg-.
+	// TODO: add other extensions. Merge with existing extensions.
 	if u, ok := w.Extension('u'); ok {
 		t, _ = Raw.Compose(t, u)
 	}
@@ -388,16 +419,18 @@ func minimizeTags(t Tag) (Tag, error) {
 
 // matcher keeps a set of supported language tags, indexed by language.
 type matcher struct {
-	default_     *haveTag
-	index        map[langID]*matchHeader
-	passSettings bool
+	default_         *haveTag
+	supported        []*haveTag
+	index            map[langID]*matchHeader
+	passSettings     bool
+	preferSameScript bool
 }
 
 // matchHeader has the lists of tags for exact matches and matches based on
 // maximized and canonicalized tags for a given language.
 type matchHeader struct {
-	exact []haveTag
-	max   []haveTag
+	exact []*haveTag
+	max   []*haveTag
 }
 
 // haveTag holds a supported Tag and its maximized script and region. The maximized
@@ -440,8 +473,10 @@ func makeHaveTag(tag Tag, index int) (haveTag, langID) {
 // script to map to another and we rely on this to keep the code simple.
 func altScript(l langID, s scriptID) scriptID {
 	for _, alt := range matchScript {
-		if (alt.lang == 0 || langID(alt.lang) == l) && scriptID(alt.have) == s {
-			return scriptID(alt.want)
+		// TODO: also match cases where language is not the same.
+		if (langID(alt.wantLang) == l || langID(alt.haveLang) == l) &&
+			scriptID(alt.haveScript) == s {
+			return scriptID(alt.wantScript)
 		}
 	}
 	return 0
@@ -457,7 +492,7 @@ func (h *matchHeader) addIfNew(n haveTag, exact bool) {
 		}
 	}
 	if exact {
-		h.exact = append(h.exact, n)
+		h.exact = append(h.exact, &n)
 	}
 	// Allow duplicate maximized tags, but create a linked list to allow quickly
 	// comparing the equivalents and bail out.
@@ -472,7 +507,7 @@ func (h *matchHeader) addIfNew(n haveTag, exact bool) {
 			break
 		}
 	}
-	h.max = append(h.max, n)
+	h.max = append(h.max, &n)
 }
 
 // header returns the matchHeader for the given language. It creates one if
@@ -486,12 +521,26 @@ func (m *matcher) header(l langID) *matchHeader {
 	return h
 }
 
+func toConf(d uint8) Confidence {
+	if d <= 10 {
+		return High
+	}
+	if d < 30 {
+		return Low
+	}
+	return No
+}
+
 // newMatcher builds an index for the given supported tags and returns it as
 // a matcher. It also expands the index by considering various equivalence classes
 // for a given tag.
-func newMatcher(supported []Tag) *matcher {
+func newMatcher(supported []Tag, options []MatchOption) *matcher {
 	m := &matcher{
-		index: make(map[langID]*matchHeader),
+		index:            make(map[langID]*matchHeader),
+		preferSameScript: true,
+	}
+	for _, o := range options {
+		o(m)
 	}
 	if len(supported) == 0 {
 		m.default_ = &haveTag{}
@@ -502,8 +551,9 @@ func newMatcher(supported []Tag) *matcher {
 	for i, tag := range supported {
 		pair, _ := makeHaveTag(tag, i)
 		m.header(tag.lang).addIfNew(pair, true)
+		m.supported = append(m.supported, &pair)
 	}
-	m.default_ = &m.header(supported[0].lang).exact[0]
+	m.default_ = m.header(supported[0].lang).exact[0]
 	for i, tag := range supported {
 		pair, max := makeHaveTag(tag, i)
 		if max != tag.lang {
@@ -511,6 +561,9 @@ func newMatcher(supported []Tag) *matcher {
 		}
 	}
 
+	// TODO: include alt script.
+	// - don't replace regions, but allow regions to be made more specific.
+
 	// update is used to add indexes in the map for equivalent languages.
 	// If force is true, the update will also apply to derived entries. To
 	// avoid applying a "transitive closure", use false.
@@ -520,7 +573,8 @@ func newMatcher(supported []Tag) *matcher {
 				return
 			}
 			hw := m.header(langID(want))
-			for _, v := range hh.max {
+			for _, ht := range hh.max {
+				v := *ht
 				if conf < v.conf {
 					v.conf = conf
 				}
@@ -536,9 +590,9 @@ func newMatcher(supported []Tag) *matcher {
 	// Add entries for languages with mutual intelligibility as defined by CLDR's
 	// languageMatch data.
 	for _, ml := range matchLang {
-		update(ml.want, ml.have, Confidence(ml.conf), false)
+		update(ml.want, ml.have, toConf(ml.distance), false)
 		if !ml.oneway {
-			update(ml.have, ml.want, Confidence(ml.conf), false)
+			update(ml.have, ml.want, toConf(ml.distance), false)
 		}
 	}
 
@@ -580,7 +634,7 @@ func (m *matcher) getBest(want ...Tag) (got *haveTag, orig Tag, c Confidence) {
 				continue
 			}
 			for i := range h.exact {
-				have := &h.exact[i]
+				have := h.exact[i]
 				if have.tag.equalsRest(w) {
 					return have, w, Exact
 				}
@@ -591,7 +645,7 @@ func (m *matcher) getBest(want ...Tag) (got *haveTag, orig Tag, c Confidence) {
 			// Base language is not defined.
 			if h != nil {
 				for i := range h.exact {
-					have := &h.exact[i]
+					have := h.exact[i]
 					if have.tag.equalsRest(w) {
 						return have, w, Exact
 					}
@@ -609,11 +663,11 @@ func (m *matcher) getBest(want ...Tag) (got *haveTag, orig Tag, c Confidence) {
 		}
 		// Check for match based on maximized tag.
 		for i := range h.max {
-			have := &h.max[i]
+			have := h.max[i]
 			best.update(have, w, max.script, max.region)
 			if best.conf == Exact {
 				for have.nextMax != 0 {
-					have = &h.max[have.nextMax]
+					have = h.max[have.nextMax]
 					best.update(have, w, max.script, max.region)
 				}
 				return best.have, best.want, High
@@ -635,11 +689,12 @@ type bestMatch struct {
 	want Tag
 	conf Confidence
 	// Cached results from applying tie-breaking rules.
-	origLang   bool
-	origReg    bool
-	regDist    uint8
-	origScript bool
-	parentDist uint8 // 255 if have is not an ancestor of want tag.
+	origLang     bool
+	origReg      bool
+	regGroupDist uint8
+	regDist      uint8
+	origScript   bool
+	parentDist   uint8 // 255 if have is not an ancestor of want tag.
 }
 
 // update updates the existing best match if the new pair is considered to be a
@@ -693,6 +748,14 @@ func (m *bestMatch) update(have *haveTag, tag Tag, maxScript scriptID, maxRegion
 		beaten = true
 	}
 
+	regGroupDist := regionGroupDist(have.maxRegion, maxRegion, maxScript, tag.lang)
+	if !beaten && m.regGroupDist != regGroupDist {
+		if regGroupDist > m.regGroupDist {
+			return
+		}
+		beaten = true
+	}
+
 	// We prefer if the pre-maximized region was specified and identical.
 	origReg := have.tag.region == tag.region && tag.region != 0
 	if !beaten && m.origReg != origReg {
@@ -702,8 +765,22 @@ func (m *bestMatch) update(have *haveTag, tag Tag, maxScript scriptID, maxRegion
 		beaten = true
 	}
 
-	// Next we prefer smaller distances between regions, as defined by regionDist.
-	regDist := regionDist(have.maxRegion, maxRegion, tag.lang)
+	// TODO: remove the region distance rule. Region distance has been replaced
+	// by the region grouping rule. For now we leave it as it still seems to
+	// have a net positive effect when applied after the grouping rule.
+	// Possible solutions:
+	// - apply the primary locale rule first to effectively disable region
+	//   region distance if groups are defined.
+	// - express the following errors in terms of grouping (if possible)
+	// - find another method of handling the following cases.
+	// maximization of legacy: find mo in
+	//      "sr-Cyrl, sr-Latn, ro, ro-MD": have ro; want ro-MD (High)
+	// region distance French: find fr-US in
+	//      "en, fr, fr-CA, fr-CH": have fr; want fr-CA (High)
+
+	// Next we prefer smaller distances between regions, as defined by
+	// regionDist.
+	regDist := uint8(regionDistance(have.maxRegion, maxRegion))
 	if !beaten && m.regDist != regDist {
 		if regDist > m.regDist {
 			return
@@ -721,6 +798,9 @@ func (m *bestMatch) update(have *haveTag, tag Tag, maxScript scriptID, maxRegion
 	}
 
 	// Finally we prefer tags which have a closer parent relationship.
+	// TODO: the parent relationship no longer seems necessary. It doesn't hurt
+	// to leave it in as the final tie-breaker, though, especially until the
+	// grouping data has further matured.
 	parentDist := parentDistance(have.tag.region, tag)
 	if !beaten && m.parentDist != parentDist {
 		if parentDist > m.parentDist {
@@ -737,6 +817,7 @@ func (m *bestMatch) update(have *haveTag, tag Tag, maxScript scriptID, maxRegion
 		m.origLang = origLang
 		m.origReg = origReg
 		m.origScript = origScript
+		m.regGroupDist = regGroupDist
 		m.regDist = regDist
 		m.parentDist = parentDist
 	}
@@ -759,15 +840,27 @@ func parentDistance(haveRegion regionID, tag Tag) uint8 {
 	return d
 }
 
-// regionDist wraps regionDistance with some exceptions to the algorithmic distance.
-func regionDist(a, b regionID, lang langID) uint8 {
-	if lang == _en {
-		// Two variants of non-US English are close to each other, regardless of distance.
-		if a != _US && b != _US {
-			return 2
+// regionGroupDist computes the distance between two regions based on their
+// CLDR grouping.
+func regionGroupDist(a, b regionID, script scriptID, lang langID) uint8 {
+	aGroup := uint(regionToGroups[a]) << 1
+	bGroup := uint(regionToGroups[b]) << 1
+	for _, ri := range matchRegion {
+		if langID(ri.lang) == lang && (ri.script == 0 || scriptID(ri.script) == script) {
+			group := uint(1 << (ri.group &^ 0x80))
+			if 0x80&ri.group == 0 {
+				if aGroup&bGroup&group != 0 { // Both regions are in the group.
+					return ri.distance
+				}
+			} else {
+				if (aGroup|bGroup)&group == 0 { // Both regions are not in the group.
+					return ri.distance
+				}
+			}
 		}
 	}
-	return uint8(regionDistance(a, b))
+	const defaultDistance = 4
+	return defaultDistance
 }
 
 // regionDistance computes the distance between two regions based on the