1
0
mirror of https://github.com/golang/go synced 2024-11-18 14:24:44 -07:00
go/present/style.go
Andrew Gerrand 9fc516408c go.tools/godoc/present: move present package from go.talks
Godoc depends on go.talks/pkg/present by way of go.tools/pkg/blog.
Better to keep all godoc dependencies in one place.

R=golang-dev, dsymonds, r
CC=golang-dev
https://golang.org/cl/13656047
2013-09-19 10:55:46 +10:00

167 lines
3.8 KiB
Go

// Copyright 2012 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 present
import (
"bytes"
"html"
"html/template"
"strings"
"unicode"
"unicode/utf8"
)
/*
Fonts are demarcated by an initial and final char bracketing a
space-delimited word, plus possibly some terminal punctuation.
The chars are
_ for italic
* for bold
` (back quote) for fixed width.
Inner appearances of the char become spaces. For instance,
_this_is_italic_!
becomes
<i>this is italic</i>!
*/
func init() {
funcs["style"] = Style
}
// Style returns s with HTML entities escaped and font indicators turned into
// HTML font tags.
func Style(s string) template.HTML {
return template.HTML(font(html.EscapeString(s)))
}
// font returns s with font indicators turned into HTML font tags.
func font(s string) string {
if strings.IndexAny(s, "[`_*") == -1 {
return s
}
words := split(s)
var b bytes.Buffer
Word:
for w, word := range words {
if len(word) < 2 {
continue Word
}
if link, _ := parseInlineLink(word); link != "" {
words[w] = link
continue Word
}
const punctuation = `.,;:()!?—–'"`
const marker = "_*`"
// Initial punctuation is OK but must be peeled off.
first := strings.IndexAny(word, marker)
if first == -1 {
continue Word
}
// Is the marker prefixed only by punctuation?
for _, r := range word[:first] {
if !strings.ContainsRune(punctuation, r) {
continue Word
}
}
open, word := word[:first], word[first:]
char := word[0] // ASCII is OK.
close := ""
switch char {
default:
continue Word
case '_':
open += "<i>"
close = "</i>"
case '*':
open += "<b>"
close = "</b>"
case '`':
open += "<code>"
close = "</code>"
}
// Terminal punctuation is OK but must be peeled off.
last := strings.LastIndex(word, word[:1])
if last == 0 {
continue Word
}
head, tail := word[:last+1], word[last+1:]
for _, r := range tail {
if !strings.ContainsRune(punctuation, r) {
continue Word
}
}
b.Reset()
b.WriteString(open)
var wid int
for i := 1; i < len(head)-1; i += wid {
var r rune
r, wid = utf8.DecodeRuneInString(head[i:])
if r != rune(char) {
// Ordinary character.
b.WriteRune(r)
continue
}
if head[i+1] != char {
// Inner char becomes space.
b.WriteRune(' ')
continue
}
// Doubled char becomes real char.
// Not worth worrying about "_x__".
b.WriteByte(char)
wid++ // Consumed two chars, both ASCII.
}
b.WriteString(close) // Write closing tag.
b.WriteString(tail) // Restore trailing punctuation.
words[w] = b.String()
}
return strings.Join(words, "")
}
// split is like strings.Fields but also returns the runs of spaces
// and treats inline links as distinct words.
func split(s string) []string {
var (
words = make([]string, 0, 10)
start = 0
)
// appendWord appends the string s[start:end] to the words slice.
// If the word contains the beginning of a link, the non-link portion
// of the word and the entire link are appended as separate words,
// and the start index is advanced to the end of the link.
appendWord := func(end int) {
if j := strings.Index(s[start:end], "[["); j > -1 {
if _, l := parseInlineLink(s[start+j:]); l > 0 {
// Append portion before link, if any.
if j > 0 {
words = append(words, s[start:start+j])
}
// Append link itself.
words = append(words, s[start+j:start+j+l])
// Advance start index to end of link.
start = start + j + l
return
}
}
// No link; just add the word.
words = append(words, s[start:end])
start = end
}
wasSpace := false
for i, r := range s {
isSpace := unicode.IsSpace(r)
if i > start && isSpace != wasSpace {
appendWord(i)
}
wasSpace = isSpace
}
for start < len(s) {
appendWord(len(s))
}
return words
}