Rewrite parseName to be consistent with freetype2.
This commit is contained in:
parent
b96c7d6be0
commit
ae9e4ed6b9
1 changed files with 119 additions and 46 deletions
|
@ -20,11 +20,8 @@ package truetype // import "github.com/golang/freetype/truetype"
|
|||
import (
|
||||
"bytes"
|
||||
"fmt"
|
||||
"io/ioutil"
|
||||
|
||||
"golang.org/x/image/math/fixed"
|
||||
"golang.org/x/text/encoding/unicode"
|
||||
"golang.org/x/text/transform"
|
||||
)
|
||||
|
||||
// An Index is a Font's index of a rune.
|
||||
|
@ -93,9 +90,9 @@ type cm struct {
|
|||
start, end, delta, offset uint32
|
||||
}
|
||||
|
||||
// An nr holds a font's name, style
|
||||
type nr struct {
|
||||
fontInfo [2]string
|
||||
// nnameInfo holds a font's name and style.
|
||||
type nameInfo struct {
|
||||
name, style string
|
||||
}
|
||||
|
||||
// A Font represents a Truetype font.
|
||||
|
@ -114,8 +111,8 @@ type Font struct {
|
|||
bounds fixed.Rectangle26_6
|
||||
// Values from the maxp section.
|
||||
maxTwilightPoints, maxStorage, maxFunctionDefs, maxStackElements uint16
|
||||
// First value from head table
|
||||
nameRecord *nr
|
||||
// First value from head table.
|
||||
nameRecord nameInfo
|
||||
}
|
||||
|
||||
func (f *Font) parseCmap() error {
|
||||
|
@ -312,62 +309,138 @@ func (f *Font) parseMaxp() error {
|
|||
}
|
||||
|
||||
func (f *Font) parseName() error {
|
||||
if len(f.name) < 18 {
|
||||
return FormatError("name data too short")
|
||||
if len(f.name) < 18 { // The name table is three uint16s + at least one Name Record. Name Records are six uint16s.
|
||||
return FormatError("name table too short")
|
||||
}
|
||||
if u16(f.name, 0) != 0 {
|
||||
if u16(f.name, 0) != 0 { // The first uint16 in the name table should be set to 0, per the spec.
|
||||
return FormatError(fmt.Sprintf("invalid format specifier: %d", u16(f.name, 0)))
|
||||
}
|
||||
numNames := int(u16(f.name, 2))
|
||||
stringOffset := u16(f.name, 4)
|
||||
font := &nr{}
|
||||
|
||||
numNames := int(u16(f.name, 2)) // The number of Name Records is the second uint16 in the name table.
|
||||
stringOffset := u16(f.name, 4) // And the offset marking the beginning of the character data is the third int16.
|
||||
|
||||
if len(f.name) < int(stringOffset) {
|
||||
return FormatError("string offset is larger than the name table")
|
||||
}
|
||||
|
||||
const ( // The platform IDs in the name table.
|
||||
PLATFORM_UNICODE = 0
|
||||
PLATFORM_MACINTOSH = 1
|
||||
PLATFORM_MICROSOFT = 3
|
||||
)
|
||||
|
||||
const ( // Encoding IDs for the Microsoft platform.
|
||||
MICROSOFT_SYMBOL_CS = 1
|
||||
MICROSOFT_UNICODE_CS = 2
|
||||
MICROSOFT_UCS_4 = 10
|
||||
)
|
||||
|
||||
// Keep track of which records contain the name and style information.
|
||||
var foundUnicode, foundAppleEnglish, foundAppleRoman, foundMicrosoft, foundApple [2]int
|
||||
foundMicrosoftEnglish := false
|
||||
|
||||
for offset, i := 6, 0; i < numNames; i, offset = i+1, offset+12 {
|
||||
nameID := u16(f.name, offset+6)
|
||||
platformID, encodingID, languageID, nameID := u16(f.name, offset), u16(f.name, offset+2), u16(f.name, offset+4), u16(f.name, offset+6)
|
||||
|
||||
switch nameID {
|
||||
case 1, 2:
|
||||
/* Keeping these here in case they're requested later.
|
||||
|
||||
platformSpecificID, languageID,platformID := u16(f.name, offset+2), u16(f.name, offset+4),u16(f.name, offset)
|
||||
*/
|
||||
length, nameOffset := u16(f.name, offset+8), u16(f.name, offset+10)
|
||||
|
||||
// Assume a non-UTF-16 value
|
||||
value := f.name[stringOffset+nameOffset : stringOffset+nameOffset+length]
|
||||
|
||||
// but just in case... if the first byte is 0, as in luxi's 13th nameRecord,
|
||||
// assume UTF-16, try to parse it if its length is even
|
||||
if value[0] == 0 && length&1 == 0 {
|
||||
var err error
|
||||
|
||||
enc := unicode.UTF16(unicode.BigEndian, unicode.IgnoreBOM)
|
||||
value, err = ioutil.ReadAll(transform.NewReader(bytes.NewReader(value), enc.NewDecoder()))
|
||||
if err != nil {
|
||||
return err
|
||||
if u16(f.name, offset+8) == 0 { // Offset + 8 is the location of the string length. If zero, skip it.
|
||||
break
|
||||
}
|
||||
switch platformID {
|
||||
case PLATFORM_UNICODE:
|
||||
foundUnicode[nameID-1] = offset
|
||||
case PLATFORM_MACINTOSH:
|
||||
if languageID == 0 { // The language ID for English is zero.
|
||||
foundAppleEnglish[nameID-1] = offset
|
||||
} else if encodingID == 0 { // A zero encoding ID means Roman.
|
||||
foundAppleRoman[nameID-1] = offset
|
||||
}
|
||||
case PLATFORM_MICROSOFT:
|
||||
if foundMicrosoft[nameID-1] == 0 || (languageID&0x3FF) == 0x009 { // Prefer Microsoft fonts in English.
|
||||
if encodingID == MICROSOFT_SYMBOL_CS || encodingID == MICROSOFT_UNICODE_CS || encodingID == MICROSOFT_UCS_4 {
|
||||
foundMicrosoftEnglish = (languageID & 0x3FF) == 9
|
||||
foundMicrosoft[nameID-1] = offset
|
||||
}
|
||||
}
|
||||
}
|
||||
// TODO: find a fail-safe method for determine if a byte slice or string is
|
||||
// UTF-16 or not.
|
||||
font.fontInfo[nameID-1] = string(value)
|
||||
}
|
||||
|
||||
if font.fontInfo[0] != "" && font.fontInfo[1] != "" {
|
||||
f.nameRecord = font
|
||||
return nil
|
||||
}
|
||||
}
|
||||
|
||||
foundApple = foundAppleRoman
|
||||
if foundAppleEnglish[0] > 0 {
|
||||
foundApple = foundAppleEnglish
|
||||
}
|
||||
|
||||
nameRecord := nameInfo{}
|
||||
// Should we decode UTF-16 strings?
|
||||
switch {
|
||||
case foundMicrosoft[0] > 0 && !(foundApple[0] > 0 && !foundMicrosoftEnglish):
|
||||
nameOffset, styleOffset := u16(f.name, foundMicrosoft[0]+10)+stringOffset, u16(f.name, foundMicrosoft[1]+10)+stringOffset
|
||||
nameLength, styleLength := u16(f.name, foundMicrosoft[0]+8), u16(f.name, foundMicrosoft[1]+8)
|
||||
|
||||
nameRecord.name = f.nameEntryInASCII(f.name[nameOffset:nameOffset+nameLength], true)
|
||||
nameRecord.style = f.nameEntryInASCII(f.name[styleOffset:styleOffset+styleLength], true)
|
||||
case foundApple[0] > 0:
|
||||
nameOffset, styleOffset := u16(f.name, foundApple[0]+10)+stringOffset, u16(f.name, foundApple[1]+10)+stringOffset
|
||||
nameLength, styleLength := u16(f.name, foundApple[0]+8), u16(f.name, foundApple[1]+8)
|
||||
|
||||
nameRecord.name = f.nameEntryInASCII(f.name[nameOffset:nameOffset+nameLength], false)
|
||||
nameRecord.style = f.nameEntryInASCII(f.name[styleOffset:styleOffset+styleLength], false)
|
||||
case foundUnicode[0] > 0:
|
||||
nameOffset, styleOffset := u16(f.name, foundUnicode[0]+10)+stringOffset, u16(f.name, foundUnicode[1]+10)+stringOffset
|
||||
nameLength, styleLength := u16(f.name, foundUnicode[0]+8), u16(f.name, foundUnicode[1]+8)
|
||||
|
||||
nameRecord.name = f.nameEntryInASCII(f.name[nameOffset:nameOffset+nameLength], true)
|
||||
nameRecord.style = f.nameEntryInASCII(f.name[styleOffset:styleOffset+styleLength], true)
|
||||
}
|
||||
|
||||
f.nameRecord = nameRecord
|
||||
|
||||
return nil
|
||||
}
|
||||
|
||||
// Name returns the TrueType font's name
|
||||
func (f *Font) Name() string {
|
||||
return f.nameRecord.fontInfo[0]
|
||||
func (f *Font) nameEntryInASCII(b []byte, utf16 bool) string {
|
||||
buf := &bytes.Buffer{}
|
||||
if utf16 { // Equivalent to tt_name_ascii_from_utf16.
|
||||
j := len(b)
|
||||
if j&1 == 1 {
|
||||
return ""
|
||||
}
|
||||
|
||||
for i := 0; i < j; i += 2 {
|
||||
el := u16(b, i)
|
||||
if el == 0 {
|
||||
continue
|
||||
} else if el < 32 || el > 127 {
|
||||
buf.WriteByte('?')
|
||||
} else {
|
||||
buf.WriteByte(byte(el))
|
||||
}
|
||||
}
|
||||
} else { // Equivalent to tt_name_ascii_from_other.
|
||||
for _, el := range b {
|
||||
if el == 0 {
|
||||
continue
|
||||
} else if el < 32 || el > 127 {
|
||||
buf.WriteByte('?')
|
||||
} else {
|
||||
buf.WriteByte(el)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
return buf.String()
|
||||
}
|
||||
|
||||
// Style returns the TrueType font's style
|
||||
// Name returns the font's name.
|
||||
func (f *Font) Name() string {
|
||||
return f.nameRecord.name
|
||||
}
|
||||
|
||||
// Style returns the font's style.
|
||||
func (f *Font) Style() string {
|
||||
return f.nameRecord.fontInfo[1]
|
||||
return f.nameRecord.style
|
||||
}
|
||||
|
||||
// scale returns x divided by f.fUnitsPerEm, rounded to the nearest integer.
|
||||
|
|
Loading…
Reference in a new issue