0
0
Fork 0
mirror of https://github.com/go-gitea/gitea synced 2024-11-13 21:41:16 +01:00
gitea/vendor/github.com/jaytaylor/html2text/html2text.go

301 lines
6.2 KiB
Go
Raw Normal View History

2016-11-03 23:16:01 +01:00
package html2text
import (
"bytes"
"io"
"regexp"
"strings"
"unicode"
"golang.org/x/net/html"
"golang.org/x/net/html/atom"
)
var (
spacingRe = regexp.MustCompile(`[ \r\n\t]+`)
newlineRe = regexp.MustCompile(`\n\n+`)
)
type textifyTraverseCtx struct {
Buf bytes.Buffer
prefix string
blockquoteLevel int
lineLength int
endsWithSpace bool
endsWithNewline bool
justClosedDiv bool
}
func (ctx *textifyTraverseCtx) traverse(node *html.Node) error {
switch node.Type {
default:
return ctx.traverseChildren(node)
case html.TextNode:
data := strings.Trim(spacingRe.ReplaceAllString(node.Data, " "), " ")
return ctx.emit(data)
case html.ElementNode:
ctx.justClosedDiv = false
switch node.DataAtom {
case atom.Br:
return ctx.emit("\n")
case atom.H1, atom.H2, atom.H3:
subCtx := textifyTraverseCtx{}
if err := subCtx.traverseChildren(node); err != nil {
return err
}
str := subCtx.Buf.String()
dividerLen := 0
for _, line := range strings.Split(str, "\n") {
if lineLen := len([]rune(line)); lineLen-1 > dividerLen {
dividerLen = lineLen - 1
}
}
divider := ""
if node.DataAtom == atom.H1 {
divider = strings.Repeat("*", dividerLen)
} else {
divider = strings.Repeat("-", dividerLen)
}
if node.DataAtom == atom.H3 {
return ctx.emit("\n\n" + str + "\n" + divider + "\n\n")
}
return ctx.emit("\n\n" + divider + "\n" + str + "\n" + divider + "\n\n")
case atom.Blockquote:
ctx.blockquoteLevel++
ctx.prefix = strings.Repeat(">", ctx.blockquoteLevel) + " "
if err := ctx.emit("\n"); err != nil {
return err
}
if ctx.blockquoteLevel == 1 {
if err := ctx.emit("\n"); err != nil {
return err
}
}
if err := ctx.traverseChildren(node); err != nil {
return err
}
ctx.blockquoteLevel--
ctx.prefix = strings.Repeat(">", ctx.blockquoteLevel)
if ctx.blockquoteLevel > 0 {
ctx.prefix += " "
}
return ctx.emit("\n\n")
case atom.Div:
if ctx.lineLength > 0 {
if err := ctx.emit("\n"); err != nil {
return err
}
}
if err := ctx.traverseChildren(node); err != nil {
return err
}
var err error
if ctx.justClosedDiv == false {
err = ctx.emit("\n")
}
ctx.justClosedDiv = true
return err
case atom.Li:
if err := ctx.emit("* "); err != nil {
return err
}
if err := ctx.traverseChildren(node); err != nil {
return err
}
return ctx.emit("\n")
case atom.B, atom.Strong:
subCtx := textifyTraverseCtx{}
subCtx.endsWithSpace = true
if err := subCtx.traverseChildren(node); err != nil {
return err
}
str := subCtx.Buf.String()
return ctx.emit("*" + str + "*")
case atom.A:
// If image is the only child, take its alt text as the link text
if img := node.FirstChild; img != nil && node.LastChild == img && img.DataAtom == atom.Img {
if altText := getAttrVal(img, "alt"); altText != "" {
ctx.emit(altText)
}
} else if err := ctx.traverseChildren(node); err != nil {
return err
}
hrefLink := ""
if attrVal := getAttrVal(node, "href"); attrVal != "" {
attrVal = ctx.normalizeHrefLink(attrVal)
if attrVal != "" {
hrefLink = "( " + attrVal + " )"
}
}
return ctx.emit(hrefLink)
case atom.P, atom.Ul, atom.Table:
if err := ctx.emit("\n\n"); err != nil {
return err
}
if err := ctx.traverseChildren(node); err != nil {
return err
}
return ctx.emit("\n\n")
case atom.Tr:
if err := ctx.traverseChildren(node); err != nil {
return err
}
return ctx.emit("\n")
case atom.Style, atom.Script, atom.Head:
// Ignore the subtree
return nil
default:
return ctx.traverseChildren(node)
}
}
}
func (ctx *textifyTraverseCtx) traverseChildren(node *html.Node) error {
for c := node.FirstChild; c != nil; c = c.NextSibling {
if err := ctx.traverse(c); err != nil {
return err
}
}
return nil
}
func (ctx *textifyTraverseCtx) emit(data string) error {
if len(data) == 0 {
return nil
}
lines := ctx.breakLongLines(data)
var err error
for _, line := range lines {
runes := []rune(line)
startsWithSpace := unicode.IsSpace(runes[0])
if !startsWithSpace && !ctx.endsWithSpace {
ctx.Buf.WriteByte(' ')
ctx.lineLength++
}
ctx.endsWithSpace = unicode.IsSpace(runes[len(runes)-1])
for _, c := range line {
_, err = ctx.Buf.WriteString(string(c))
if err != nil {
return err
}
ctx.lineLength++
if c == '\n' {
ctx.lineLength = 0
if ctx.prefix != "" {
_, err = ctx.Buf.WriteString(ctx.prefix)
if err != nil {
return err
}
}
}
}
}
return nil
}
func (ctx *textifyTraverseCtx) breakLongLines(data string) []string {
// only break lines when we are in blockquotes
if ctx.blockquoteLevel == 0 {
return []string{data}
}
var ret []string
runes := []rune(data)
l := len(runes)
existing := ctx.lineLength
if existing >= 74 {
ret = append(ret, "\n")
existing = 0
}
for l+existing > 74 {
i := 74 - existing
for i >= 0 && !unicode.IsSpace(runes[i]) {
i--
}
if i == -1 {
// no spaces, so go the other way
i = 74 - existing
for i < l && !unicode.IsSpace(runes[i]) {
i++
}
}
ret = append(ret, string(runes[:i])+"\n")
for i < l && unicode.IsSpace(runes[i]) {
i++
}
runes = runes[i:]
l = len(runes)
existing = 0
}
if len(runes) > 0 {
ret = append(ret, string(runes))
}
return ret
}
func (ctx *textifyTraverseCtx) normalizeHrefLink(link string) string {
link = strings.TrimSpace(link)
link = strings.TrimPrefix(link, "mailto:")
return link
}
func getAttrVal(node *html.Node, attrName string) string {
for _, attr := range node.Attr {
if attr.Key == attrName {
return attr.Val
}
}
return ""
}
func FromReader(reader io.Reader) (string, error) {
doc, err := html.Parse(reader)
if err != nil {
return "", err
}
ctx := textifyTraverseCtx{
Buf: bytes.Buffer{},
}
if err = ctx.traverse(doc); err != nil {
return "", err
}
text := strings.TrimSpace(newlineRe.ReplaceAllString(
strings.Replace(ctx.Buf.String(), "\n ", "\n", -1), "\n\n"))
return text, nil
}
func FromString(input string) (string, error) {
text, err := FromReader(strings.NewReader(input))
if err != nil {
return "", err
}
return text, nil
}