diff options
Diffstat (limited to 'vendor/golang.org/x/text/language/parse.go')
-rw-r--r-- | vendor/golang.org/x/text/language/parse.go | 228 |
1 files changed, 0 insertions, 228 deletions
diff --git a/vendor/golang.org/x/text/language/parse.go b/vendor/golang.org/x/text/language/parse.go deleted file mode 100644 index d50c8aac..00000000 --- a/vendor/golang.org/x/text/language/parse.go +++ /dev/null @@ -1,228 +0,0 @@ -// Copyright 2013 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. - -package language - -import ( - "errors" - "strconv" - "strings" - - "golang.org/x/text/internal/language" -) - -// ValueError is returned by any of the parsing functions when the -// input is well-formed but the respective subtag is not recognized -// as a valid value. -type ValueError interface { - error - - // Subtag returns the subtag for which the error occurred. - Subtag() string -} - -// Parse parses the given BCP 47 string and returns a valid Tag. If parsing -// failed it returns an error and any part of the tag that could be parsed. -// If parsing succeeded but an unknown value was found, it returns -// ValueError. The Tag returned in this case is just stripped of the unknown -// value. All other values are preserved. It accepts tags in the BCP 47 format -// and extensions to this standard defined in -// http://www.unicode.org/reports/tr35/#Unicode_Language_and_Locale_Identifiers. -// The resulting tag is canonicalized using the default canonicalization type. -func Parse(s string) (t Tag, err error) { - return Default.Parse(s) -} - -// Parse parses the given BCP 47 string and returns a valid Tag. If parsing -// failed it returns an error and any part of the tag that could be parsed. -// If parsing succeeded but an unknown value was found, it returns -// ValueError. The Tag returned in this case is just stripped of the unknown -// value. All other values are preserved. It accepts tags in the BCP 47 format -// and extensions to this standard defined in -// http://www.unicode.org/reports/tr35/#Unicode_Language_and_Locale_Identifiers. -// The resulting tag is canonicalized using the the canonicalization type c. -func (c CanonType) Parse(s string) (t Tag, err error) { - tt, err := language.Parse(s) - if err != nil { - return makeTag(tt), err - } - tt, changed := canonicalize(c, tt) - if changed { - tt.RemakeString() - } - return makeTag(tt), err -} - -// Compose creates a Tag from individual parts, which may be of type Tag, Base, -// Script, Region, Variant, []Variant, Extension, []Extension or error. If a -// Base, Script or Region or slice of type Variant or Extension is passed more -// than once, the latter will overwrite the former. Variants and Extensions are -// accumulated, but if two extensions of the same type are passed, the latter -// will replace the former. For -u extensions, though, the key-type pairs are -// added, where later values overwrite older ones. A Tag overwrites all former -// values and typically only makes sense as the first argument. The resulting -// tag is returned after canonicalizing using the Default CanonType. If one or -// more errors are encountered, one of the errors is returned. -func Compose(part ...interface{}) (t Tag, err error) { - return Default.Compose(part...) -} - -// Compose creates a Tag from individual parts, which may be of type Tag, Base, -// Script, Region, Variant, []Variant, Extension, []Extension or error. If a -// Base, Script or Region or slice of type Variant or Extension is passed more -// than once, the latter will overwrite the former. Variants and Extensions are -// accumulated, but if two extensions of the same type are passed, the latter -// will replace the former. For -u extensions, though, the key-type pairs are -// added, where later values overwrite older ones. A Tag overwrites all former -// values and typically only makes sense as the first argument. The resulting -// tag is returned after canonicalizing using CanonType c. If one or more errors -// are encountered, one of the errors is returned. -func (c CanonType) Compose(part ...interface{}) (t Tag, err error) { - var b language.Builder - if err = update(&b, part...); err != nil { - return und, err - } - b.Tag, _ = canonicalize(c, b.Tag) - return makeTag(b.Make()), err -} - -var errInvalidArgument = errors.New("invalid Extension or Variant") - -func update(b *language.Builder, part ...interface{}) (err error) { - for _, x := range part { - switch v := x.(type) { - case Tag: - b.SetTag(v.tag()) - case Base: - b.Tag.LangID = v.langID - case Script: - b.Tag.ScriptID = v.scriptID - case Region: - b.Tag.RegionID = v.regionID - case Variant: - if v.variant == "" { - err = errInvalidArgument - break - } - b.AddVariant(v.variant) - case Extension: - if v.s == "" { - err = errInvalidArgument - break - } - b.SetExt(v.s) - case []Variant: - b.ClearVariants() - for _, v := range v { - b.AddVariant(v.variant) - } - case []Extension: - b.ClearExtensions() - for _, e := range v { - b.SetExt(e.s) - } - // TODO: support parsing of raw strings based on morphology or just extensions? - case error: - if v != nil { - err = v - } - } - } - return -} - -var errInvalidWeight = errors.New("ParseAcceptLanguage: invalid weight") - -// ParseAcceptLanguage parses the contents of an Accept-Language header as -// defined in http://www.ietf.org/rfc/rfc2616.txt and returns a list of Tags and -// a list of corresponding quality weights. It is more permissive than RFC 2616 -// and may return non-nil slices even if the input is not valid. -// The Tags will be sorted by highest weight first and then by first occurrence. -// Tags with a weight of zero will be dropped. An error will be returned if the -// input could not be parsed. -func ParseAcceptLanguage(s string) (tag []Tag, q []float32, err error) { - var entry string - for s != "" { - if entry, s = split(s, ','); entry == "" { - continue - } - - entry, weight := split(entry, ';') - - // Scan the language. - t, err := Parse(entry) - if err != nil { - id, ok := acceptFallback[entry] - if !ok { - return nil, nil, err - } - t = makeTag(language.Tag{LangID: id}) - } - - // Scan the optional weight. - w := 1.0 - if weight != "" { - weight = consume(weight, 'q') - weight = consume(weight, '=') - // consume returns the empty string when a token could not be - // consumed, resulting in an error for ParseFloat. - if w, err = strconv.ParseFloat(weight, 32); err != nil { - return nil, nil, errInvalidWeight - } - // Drop tags with a quality weight of 0. - if w <= 0 { - continue - } - } - - tag = append(tag, t) - q = append(q, float32(w)) - } - sortStable(&tagSort{tag, q}) - return tag, q, nil -} - -// consume removes a leading token c from s and returns the result or the empty -// string if there is no such token. -func consume(s string, c byte) string { - if s == "" || s[0] != c { - return "" - } - return strings.TrimSpace(s[1:]) -} - -func split(s string, c byte) (head, tail string) { - if i := strings.IndexByte(s, c); i >= 0 { - return strings.TrimSpace(s[:i]), strings.TrimSpace(s[i+1:]) - } - return strings.TrimSpace(s), "" -} - -// Add hack mapping to deal with a small number of cases that that occur -// in Accept-Language (with reasonable frequency). -var acceptFallback = map[string]language.Language{ - "english": _en, - "deutsch": _de, - "italian": _it, - "french": _fr, - "*": _mul, // defined in the spec to match all languages. -} - -type tagSort struct { - tag []Tag - q []float32 -} - -func (s *tagSort) Len() int { - return len(s.q) -} - -func (s *tagSort) Less(i, j int) bool { - return s.q[i] > s.q[j] -} - -func (s *tagSort) Swap(i, j int) { - s.tag[i], s.tag[j] = s.tag[j], s.tag[i] - s.q[i], s.q[j] = s.q[j], s.q[i] -} |