mirror of
1
Fork 0
gotosocial/vendor/github.com/yuin/goldmark/ast/inline.go

550 lines
12 KiB
Go
Raw Normal View History

package ast
import (
"fmt"
"strings"
textm "github.com/yuin/goldmark/text"
"github.com/yuin/goldmark/util"
)
// A BaseInline struct implements the Node interface partialliy.
type BaseInline struct {
BaseNode
}
// Type implements Node.Type.
func (b *BaseInline) Type() NodeType {
return TypeInline
}
// IsRaw implements Node.IsRaw.
func (b *BaseInline) IsRaw() bool {
return false
}
// HasBlankPreviousLines implements Node.HasBlankPreviousLines.
func (b *BaseInline) HasBlankPreviousLines() bool {
panic("can not call with inline nodes.")
}
// SetBlankPreviousLines implements Node.SetBlankPreviousLines.
func (b *BaseInline) SetBlankPreviousLines(v bool) {
panic("can not call with inline nodes.")
}
// Lines implements Node.Lines.
func (b *BaseInline) Lines() *textm.Segments {
panic("can not call with inline nodes.")
}
// SetLines implements Node.SetLines.
func (b *BaseInline) SetLines(v *textm.Segments) {
panic("can not call with inline nodes.")
}
// A Text struct represents a textual content of the Markdown text.
type Text struct {
BaseInline
// Segment is a position in a source text.
Segment textm.Segment
flags uint8
}
const (
textSoftLineBreak = 1 << iota
textHardLineBreak
textRaw
textCode
)
func textFlagsString(flags uint8) string {
buf := []string{}
if flags&textSoftLineBreak != 0 {
buf = append(buf, "SoftLineBreak")
}
if flags&textHardLineBreak != 0 {
buf = append(buf, "HardLineBreak")
}
if flags&textRaw != 0 {
buf = append(buf, "Raw")
}
if flags&textCode != 0 {
buf = append(buf, "Code")
}
return strings.Join(buf, ", ")
}
// Inline implements Inline.Inline.
func (n *Text) Inline() {
}
// SoftLineBreak returns true if this node ends with a new line,
// otherwise false.
func (n *Text) SoftLineBreak() bool {
return n.flags&textSoftLineBreak != 0
}
// SetSoftLineBreak sets whether this node ends with a new line.
func (n *Text) SetSoftLineBreak(v bool) {
if v {
n.flags |= textSoftLineBreak
} else {
n.flags = n.flags &^ textSoftLineBreak
}
}
// IsRaw returns true if this text should be rendered without unescaping
// back slash escapes and resolving references.
func (n *Text) IsRaw() bool {
return n.flags&textRaw != 0
}
// SetRaw sets whether this text should be rendered as raw contents.
func (n *Text) SetRaw(v bool) {
if v {
n.flags |= textRaw
} else {
n.flags = n.flags &^ textRaw
}
}
// HardLineBreak returns true if this node ends with a hard line break.
// See https://spec.commonmark.org/0.30/#hard-line-breaks for details.
func (n *Text) HardLineBreak() bool {
return n.flags&textHardLineBreak != 0
}
// SetHardLineBreak sets whether this node ends with a hard line break.
func (n *Text) SetHardLineBreak(v bool) {
if v {
n.flags |= textHardLineBreak
} else {
n.flags = n.flags &^ textHardLineBreak
}
}
// Merge merges a Node n into this node.
// Merge returns true if the given node has been merged, otherwise false.
func (n *Text) Merge(node Node, source []byte) bool {
t, ok := node.(*Text)
if !ok {
return false
}
if n.Segment.Stop != t.Segment.Start || t.Segment.Padding != 0 ||
source[n.Segment.Stop-1] == '\n' || t.IsRaw() != n.IsRaw() {
return false
}
n.Segment.Stop = t.Segment.Stop
n.SetSoftLineBreak(t.SoftLineBreak())
n.SetHardLineBreak(t.HardLineBreak())
return true
}
// Text implements Node.Text.
func (n *Text) Text(source []byte) []byte {
return n.Segment.Value(source)
}
// Dump implements Node.Dump.
func (n *Text) Dump(source []byte, level int) {
fs := textFlagsString(n.flags)
if len(fs) != 0 {
fs = "(" + fs + ")"
}
fmt.Printf("%sText%s: \"%s\"\n", strings.Repeat(" ", level), fs, strings.TrimRight(string(n.Text(source)), "\n"))
}
// KindText is a NodeKind of the Text node.
var KindText = NewNodeKind("Text")
// Kind implements Node.Kind.
func (n *Text) Kind() NodeKind {
return KindText
}
// NewText returns a new Text node.
func NewText() *Text {
return &Text{
BaseInline: BaseInline{},
}
}
// NewTextSegment returns a new Text node with the given source position.
func NewTextSegment(v textm.Segment) *Text {
return &Text{
BaseInline: BaseInline{},
Segment: v,
}
}
// NewRawTextSegment returns a new Text node with the given source position.
// The new node should be rendered as raw contents.
func NewRawTextSegment(v textm.Segment) *Text {
t := &Text{
BaseInline: BaseInline{},
Segment: v,
}
t.SetRaw(true)
return t
}
// MergeOrAppendTextSegment merges a given s into the last child of the parent if
// it can be merged, otherwise creates a new Text node and appends it to after current
// last child.
func MergeOrAppendTextSegment(parent Node, s textm.Segment) {
last := parent.LastChild()
t, ok := last.(*Text)
if ok && t.Segment.Stop == s.Start && !t.SoftLineBreak() {
t.Segment = t.Segment.WithStop(s.Stop)
} else {
parent.AppendChild(parent, NewTextSegment(s))
}
}
// MergeOrReplaceTextSegment merges a given s into a previous sibling of the node n
// if a previous sibling of the node n is *Text, otherwise replaces Node n with s.
func MergeOrReplaceTextSegment(parent Node, n Node, s textm.Segment) {
prev := n.PreviousSibling()
if t, ok := prev.(*Text); ok && t.Segment.Stop == s.Start && !t.SoftLineBreak() {
t.Segment = t.Segment.WithStop(s.Stop)
parent.RemoveChild(parent, n)
} else {
parent.ReplaceChild(parent, n, NewTextSegment(s))
}
}
// A String struct is a textual content that has a concrete value.
type String struct {
BaseInline
Value []byte
flags uint8
}
// Inline implements Inline.Inline.
func (n *String) Inline() {
}
// IsRaw returns true if this text should be rendered without unescaping
// back slash escapes and resolving references.
func (n *String) IsRaw() bool {
return n.flags&textRaw != 0
}
// SetRaw sets whether this text should be rendered as raw contents.
func (n *String) SetRaw(v bool) {
if v {
n.flags |= textRaw
} else {
n.flags = n.flags &^ textRaw
}
}
// IsCode returns true if this text should be rendered without any
// modifications.
func (n *String) IsCode() bool {
return n.flags&textCode != 0
}
// SetCode sets whether this text should be rendered without any modifications.
func (n *String) SetCode(v bool) {
if v {
n.flags |= textCode
} else {
n.flags = n.flags &^ textCode
}
}
// Text implements Node.Text.
func (n *String) Text(source []byte) []byte {
return n.Value
}
// Dump implements Node.Dump.
func (n *String) Dump(source []byte, level int) {
fs := textFlagsString(n.flags)
if len(fs) != 0 {
fs = "(" + fs + ")"
}
fmt.Printf("%sString%s: \"%s\"\n", strings.Repeat(" ", level), fs, strings.TrimRight(string(n.Value), "\n"))
}
// KindString is a NodeKind of the String node.
var KindString = NewNodeKind("String")
// Kind implements Node.Kind.
func (n *String) Kind() NodeKind {
return KindString
}
// NewString returns a new String node.
func NewString(v []byte) *String {
return &String{
Value: v,
}
}
// A CodeSpan struct represents a code span of Markdown text.
type CodeSpan struct {
BaseInline
}
// Inline implements Inline.Inline .
func (n *CodeSpan) Inline() {
}
// IsBlank returns true if this node consists of spaces, otherwise false.
func (n *CodeSpan) IsBlank(source []byte) bool {
for c := n.FirstChild(); c != nil; c = c.NextSibling() {
text := c.(*Text).Segment
if !util.IsBlank(text.Value(source)) {
return false
}
}
return true
}
// Dump implements Node.Dump.
func (n *CodeSpan) Dump(source []byte, level int) {
DumpHelper(n, source, level, nil, nil)
}
// KindCodeSpan is a NodeKind of the CodeSpan node.
var KindCodeSpan = NewNodeKind("CodeSpan")
// Kind implements Node.Kind.
func (n *CodeSpan) Kind() NodeKind {
return KindCodeSpan
}
// NewCodeSpan returns a new CodeSpan node.
func NewCodeSpan() *CodeSpan {
return &CodeSpan{
BaseInline: BaseInline{},
}
}
// An Emphasis struct represents an emphasis of Markdown text.
type Emphasis struct {
BaseInline
// Level is a level of the emphasis.
Level int
}
// Dump implements Node.Dump.
func (n *Emphasis) Dump(source []byte, level int) {
m := map[string]string{
"Level": fmt.Sprintf("%v", n.Level),
}
DumpHelper(n, source, level, m, nil)
}
// KindEmphasis is a NodeKind of the Emphasis node.
var KindEmphasis = NewNodeKind("Emphasis")
// Kind implements Node.Kind.
func (n *Emphasis) Kind() NodeKind {
return KindEmphasis
}
// NewEmphasis returns a new Emphasis node with the given level.
func NewEmphasis(level int) *Emphasis {
return &Emphasis{
BaseInline: BaseInline{},
Level: level,
}
}
type baseLink struct {
BaseInline
// Destination is a destination(URL) of this link.
Destination []byte
// Title is a title of this link.
Title []byte
}
// Inline implements Inline.Inline.
func (n *baseLink) Inline() {
}
// A Link struct represents a link of the Markdown text.
type Link struct {
baseLink
}
// Dump implements Node.Dump.
func (n *Link) Dump(source []byte, level int) {
m := map[string]string{}
m["Destination"] = string(n.Destination)
m["Title"] = string(n.Title)
DumpHelper(n, source, level, m, nil)
}
// KindLink is a NodeKind of the Link node.
var KindLink = NewNodeKind("Link")
// Kind implements Node.Kind.
func (n *Link) Kind() NodeKind {
return KindLink
}
// NewLink returns a new Link node.
func NewLink() *Link {
c := &Link{
baseLink: baseLink{
BaseInline: BaseInline{},
},
}
return c
}
// An Image struct represents an image of the Markdown text.
type Image struct {
baseLink
}
// Dump implements Node.Dump.
func (n *Image) Dump(source []byte, level int) {
m := map[string]string{}
m["Destination"] = string(n.Destination)
m["Title"] = string(n.Title)
DumpHelper(n, source, level, m, nil)
}
// KindImage is a NodeKind of the Image node.
var KindImage = NewNodeKind("Image")
// Kind implements Node.Kind.
func (n *Image) Kind() NodeKind {
return KindImage
}
// NewImage returns a new Image node.
func NewImage(link *Link) *Image {
c := &Image{
baseLink: baseLink{
BaseInline: BaseInline{},
},
}
c.Destination = link.Destination
c.Title = link.Title
for n := link.FirstChild(); n != nil; {
next := n.NextSibling()
link.RemoveChild(link, n)
c.AppendChild(c, n)
n = next
}
return c
}
// AutoLinkType defines kind of auto links.
type AutoLinkType int
const (
// AutoLinkEmail indicates that an autolink is an email address.
AutoLinkEmail AutoLinkType = iota + 1
// AutoLinkURL indicates that an autolink is a generic URL.
AutoLinkURL
)
// An AutoLink struct represents an autolink of the Markdown text.
type AutoLink struct {
BaseInline
// Type is a type of this autolink.
AutoLinkType AutoLinkType
// Protocol specified a protocol of the link.
Protocol []byte
value *Text
}
// Inline implements Inline.Inline.
func (n *AutoLink) Inline() {}
// Dump implements Node.Dump.
func (n *AutoLink) Dump(source []byte, level int) {
segment := n.value.Segment
m := map[string]string{
"Value": string(segment.Value(source)),
}
DumpHelper(n, source, level, m, nil)
}
// KindAutoLink is a NodeKind of the AutoLink node.
var KindAutoLink = NewNodeKind("AutoLink")
// Kind implements Node.Kind.
func (n *AutoLink) Kind() NodeKind {
return KindAutoLink
}
// URL returns an url of this node.
func (n *AutoLink) URL(source []byte) []byte {
if n.Protocol != nil {
s := n.value.Segment
ret := make([]byte, 0, len(n.Protocol)+s.Len()+3)
ret = append(ret, n.Protocol...)
ret = append(ret, ':', '/', '/')
ret = append(ret, n.value.Text(source)...)
return ret
}
return n.value.Text(source)
}
// Label returns a label of this node.
func (n *AutoLink) Label(source []byte) []byte {
return n.value.Text(source)
}
// NewAutoLink returns a new AutoLink node.
func NewAutoLink(typ AutoLinkType, value *Text) *AutoLink {
return &AutoLink{
BaseInline: BaseInline{},
value: value,
AutoLinkType: typ,
}
}
// A RawHTML struct represents an inline raw HTML of the Markdown text.
type RawHTML struct {
BaseInline
Segments *textm.Segments
}
// Inline implements Inline.Inline.
func (n *RawHTML) Inline() {}
// Dump implements Node.Dump.
func (n *RawHTML) Dump(source []byte, level int) {
m := map[string]string{}
t := []string{}
for i := 0; i < n.Segments.Len(); i++ {
segment := n.Segments.At(i)
t = append(t, string(segment.Value(source)))
}
m["RawText"] = strings.Join(t, "")
DumpHelper(n, source, level, m, nil)
}
// KindRawHTML is a NodeKind of the RawHTML node.
var KindRawHTML = NewNodeKind("RawHTML")
// Kind implements Node.Kind.
func (n *RawHTML) Kind() NodeKind {
return KindRawHTML
}
// NewRawHTML returns a new RawHTML node.
func NewRawHTML() *RawHTML {
return &RawHTML{
Segments: textm.NewSegments(),
}
}