27757714d0
* Move to goldmark Markdown rendering moved from blackfriday to the goldmark. Multiple subtle changes required to the goldmark extensions to keep current rendering and defaults. Can go further with goldmark linkify and have this work within markdown rendering making the link processor unnecessary. Need to think about how to go about allowing extensions - at present it seems that these would be hard to do without recompilation. * linter fixes Co-authored-by: Lauris BH <lauris@nix.lv>
242 lines
6.2 KiB
Go
242 lines
6.2 KiB
Go
package parser
|
|
|
|
import (
|
|
"fmt"
|
|
"strings"
|
|
"unicode"
|
|
|
|
"github.com/yuin/goldmark/ast"
|
|
"github.com/yuin/goldmark/text"
|
|
"github.com/yuin/goldmark/util"
|
|
)
|
|
|
|
// A DelimiterProcessor interface provides a set of functions about
|
|
// Deliiter nodes.
|
|
type DelimiterProcessor interface {
|
|
// IsDelimiter returns true if given character is a delimiter, otherwise false.
|
|
IsDelimiter(byte) bool
|
|
|
|
// CanOpenCloser returns true if given opener can close given closer, otherwise false.
|
|
CanOpenCloser(opener, closer *Delimiter) bool
|
|
|
|
// OnMatch will be called when new matched delimiter found.
|
|
// OnMatch should return a new Node correspond to the matched delimiter.
|
|
OnMatch(consumes int) ast.Node
|
|
}
|
|
|
|
// A Delimiter struct represents a delimiter like '*' of the Markdown text.
|
|
type Delimiter struct {
|
|
ast.BaseInline
|
|
|
|
Segment text.Segment
|
|
|
|
// CanOpen is set true if this delimiter can open a span for a new node.
|
|
// See https://spec.commonmark.org/0.29/#can-open-emphasis for details.
|
|
CanOpen bool
|
|
|
|
// CanClose is set true if this delimiter can close a span for a new node.
|
|
// See https://spec.commonmark.org/0.29/#can-open-emphasis for details.
|
|
CanClose bool
|
|
|
|
// Length is a remaining length of this delmiter.
|
|
Length int
|
|
|
|
// OriginalLength is a original length of this delimiter.
|
|
OriginalLength int
|
|
|
|
// Char is a character of this delimiter.
|
|
Char byte
|
|
|
|
// PreviousDelimiter is a previous sibling delimiter node of this delimiter.
|
|
PreviousDelimiter *Delimiter
|
|
|
|
// NextDelimiter is a next sibling delimiter node of this delimiter.
|
|
NextDelimiter *Delimiter
|
|
|
|
// Processor is a DelimiterProcessor associated with this delimiter.
|
|
Processor DelimiterProcessor
|
|
}
|
|
|
|
// Inline implements Inline.Inline.
|
|
func (d *Delimiter) Inline() {}
|
|
|
|
// Dump implements Node.Dump.
|
|
func (d *Delimiter) Dump(source []byte, level int) {
|
|
fmt.Printf("%sDelimiter: \"%s\"\n", strings.Repeat(" ", level), string(d.Text(source)))
|
|
}
|
|
|
|
var kindDelimiter = ast.NewNodeKind("Delimiter")
|
|
|
|
// Kind implements Node.Kind
|
|
func (d *Delimiter) Kind() ast.NodeKind {
|
|
return kindDelimiter
|
|
}
|
|
|
|
// Text implements Node.Text
|
|
func (d *Delimiter) Text(source []byte) []byte {
|
|
return d.Segment.Value(source)
|
|
}
|
|
|
|
// ConsumeCharacters consumes delimiters.
|
|
func (d *Delimiter) ConsumeCharacters(n int) {
|
|
d.Length -= n
|
|
d.Segment = d.Segment.WithStop(d.Segment.Start + d.Length)
|
|
}
|
|
|
|
// CalcComsumption calculates how many characters should be used for opening
|
|
// a new span correspond to given closer.
|
|
func (d *Delimiter) CalcComsumption(closer *Delimiter) int {
|
|
if (d.CanClose || closer.CanOpen) && (d.OriginalLength+closer.OriginalLength)%3 == 0 && closer.OriginalLength%3 != 0 {
|
|
return 0
|
|
}
|
|
if d.Length >= 2 && closer.Length >= 2 {
|
|
return 2
|
|
}
|
|
return 1
|
|
}
|
|
|
|
// NewDelimiter returns a new Delimiter node.
|
|
func NewDelimiter(canOpen, canClose bool, length int, char byte, processor DelimiterProcessor) *Delimiter {
|
|
c := &Delimiter{
|
|
BaseInline: ast.BaseInline{},
|
|
CanOpen: canOpen,
|
|
CanClose: canClose,
|
|
Length: length,
|
|
OriginalLength: length,
|
|
Char: char,
|
|
PreviousDelimiter: nil,
|
|
NextDelimiter: nil,
|
|
Processor: processor,
|
|
}
|
|
return c
|
|
}
|
|
|
|
// ScanDelimiter scans a delimiter by given DelimiterProcessor.
|
|
func ScanDelimiter(line []byte, before rune, min int, processor DelimiterProcessor) *Delimiter {
|
|
i := 0
|
|
c := line[i]
|
|
j := i
|
|
if !processor.IsDelimiter(c) {
|
|
return nil
|
|
}
|
|
for ; j < len(line) && c == line[j]; j++ {
|
|
}
|
|
if (j - i) >= min {
|
|
after := rune(' ')
|
|
if j != len(line) {
|
|
after = util.ToRune(line, j)
|
|
}
|
|
|
|
canOpen, canClose := false, false
|
|
beforeIsPunctuation := unicode.IsPunct(before)
|
|
beforeIsWhitespace := unicode.IsSpace(before)
|
|
afterIsPunctuation := unicode.IsPunct(after)
|
|
afterIsWhitespace := unicode.IsSpace(after)
|
|
|
|
isLeft := !afterIsWhitespace &&
|
|
(!afterIsPunctuation || beforeIsWhitespace || beforeIsPunctuation)
|
|
isRight := !beforeIsWhitespace &&
|
|
(!beforeIsPunctuation || afterIsWhitespace || afterIsPunctuation)
|
|
|
|
if line[i] == '_' {
|
|
canOpen = isLeft && (!isRight || beforeIsPunctuation)
|
|
canClose = isRight && (!isLeft || afterIsPunctuation)
|
|
} else {
|
|
canOpen = isLeft
|
|
canClose = isRight
|
|
}
|
|
return NewDelimiter(canOpen, canClose, j-i, c, processor)
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// ProcessDelimiters processes the delimiter list in the context.
|
|
// Processing will be stop when reaching the bottom.
|
|
//
|
|
// If you implement an inline parser that can have other inline nodes as
|
|
// children, you should call this function when nesting span has closed.
|
|
func ProcessDelimiters(bottom ast.Node, pc Context) {
|
|
lastDelimiter := pc.LastDelimiter()
|
|
if lastDelimiter == nil {
|
|
return
|
|
}
|
|
var closer *Delimiter
|
|
if bottom != nil {
|
|
if bottom != lastDelimiter {
|
|
for c := lastDelimiter.PreviousSibling(); c != nil; {
|
|
if d, ok := c.(*Delimiter); ok {
|
|
closer = d
|
|
}
|
|
prev := c.PreviousSibling()
|
|
if prev == bottom {
|
|
break
|
|
}
|
|
c = prev
|
|
}
|
|
}
|
|
} else {
|
|
closer = pc.FirstDelimiter()
|
|
}
|
|
if closer == nil {
|
|
pc.ClearDelimiters(bottom)
|
|
return
|
|
}
|
|
for closer != nil {
|
|
if !closer.CanClose {
|
|
closer = closer.NextDelimiter
|
|
continue
|
|
}
|
|
consume := 0
|
|
found := false
|
|
maybeOpener := false
|
|
var opener *Delimiter
|
|
for opener = closer.PreviousDelimiter; opener != nil; opener = opener.PreviousDelimiter {
|
|
if opener.CanOpen && opener.Processor.CanOpenCloser(opener, closer) {
|
|
maybeOpener = true
|
|
consume = opener.CalcComsumption(closer)
|
|
if consume > 0 {
|
|
found = true
|
|
break
|
|
}
|
|
}
|
|
}
|
|
if !found {
|
|
if !maybeOpener && !closer.CanOpen {
|
|
pc.RemoveDelimiter(closer)
|
|
}
|
|
closer = closer.NextDelimiter
|
|
continue
|
|
}
|
|
opener.ConsumeCharacters(consume)
|
|
closer.ConsumeCharacters(consume)
|
|
|
|
node := opener.Processor.OnMatch(consume)
|
|
|
|
parent := opener.Parent()
|
|
child := opener.NextSibling()
|
|
|
|
for child != nil && child != closer {
|
|
next := child.NextSibling()
|
|
node.AppendChild(node, child)
|
|
child = next
|
|
}
|
|
parent.InsertAfter(parent, opener, node)
|
|
|
|
for c := opener.NextDelimiter; c != nil && c != closer; {
|
|
next := c.NextDelimiter
|
|
pc.RemoveDelimiter(c)
|
|
c = next
|
|
}
|
|
|
|
if opener.Length == 0 {
|
|
pc.RemoveDelimiter(opener)
|
|
}
|
|
|
|
if closer.Length == 0 {
|
|
next := closer.NextDelimiter
|
|
pc.RemoveDelimiter(closer)
|
|
closer = next
|
|
}
|
|
}
|
|
pc.ClearDelimiters(bottom)
|
|
}
|