Files
2026-08-03 13:00:21 +03:00

150 lines
4.2 KiB
Go

// Copyright (c) 2015-present Mattermost, Inc. All Rights Reserved.
// See LICENSE.txt for license information.
package markdown
import (
"sync/atomic"
)
// defaultMaxPostRunes is used only if no value has been registered via
// SetMaxPostRunes, which should not happen in practice. It assumes a post is at
// most 64KiB in DB. Assuming four bytes per rune, it gives us 16*1024 runes.
// If we allow scanning up to twice (arbitrary value) that value, we get:
const defaultMaxPostRunes = 2 * 16 * 1024
var maxPostRunes atomic.Int64
func init() {
maxPostRunes.Store(defaultMaxPostRunes)
}
// SetMaxPostRunes registers the real configured maximum post size, in runes, that MaxLen uses to
// compute the maximum markdown input length. This lets MaxLen reflect the real limit regardless
// of which code path first parses markdown, rather than depending on that path having already
// pushed the value in. Safe to call concurrently with MaxLen.
func SetMaxPostRunes(size int) {
maxPostRunes.Store(int64(size))
}
// MaxLen returns the current maximum markdown input length, in bytes, enforced by Parse and
// Inspect: four times the maximum post size, in runes, registered via SetMaxPostRunes (or
// defaultMaxPostRunes if it hasn't been called yet), assuming a worst case of four bytes per
// rune.
func MaxLen() int {
return 4 * int(maxPostRunes.Load())
}
// Inspect traverses the markdown tree in depth-first order. If f returns true, Inspect invokes f
// recursively for each child of the block or inline, followed by a call of f(nil).
func Inspect(markdown string, f func(any) bool) {
if len(markdown) > MaxLen() {
return
}
document, referenceDefinitions := Parse(markdown)
InspectBlock(document, func(block Block) bool {
if !f(block) {
return false
}
switch v := block.(type) {
case *Paragraph:
for _, inline := range MergeInlineText(v.ParseInlines(referenceDefinitions)) {
InspectInline(inline, func(inline Inline) bool {
return f(inline)
})
}
}
return true
})
}
// InspectBlock traverses the blocks in depth-first order, starting with block. If f returns true,
// InspectBlock invokes f recursively for each child of the block, followed by a call of f(nil).
func InspectBlock(block Block, f func(Block) bool) {
stack := []Block{block}
// Using seen for backtracking
seen := map[Block]bool{}
for len(stack) > 0 {
// "peek" the node from the stack
block := stack[len(stack)-1]
if seen[block] {
// "pop" the node only when backtracking(seen)
stack = stack[:len(stack)-1]
f(nil)
continue
}
seen[block] = true
// Process the node
if !f(block) {
continue
}
switch v := block.(type) {
case *Document:
for i := len(v.Children) - 1; i >= 0; i-- {
stack = append(stack, v.Children[i])
}
case *List:
for i := len(v.Children) - 1; i >= 0; i-- {
stack = append(stack, v.Children[i])
}
case *ListItem:
for i := len(v.Children) - 1; i >= 0; i-- {
stack = append(stack, v.Children[i])
}
case *BlockQuote:
for i := len(v.Children) - 1; i >= 0; i-- {
stack = append(stack, v.Children[i])
}
}
}
}
// InspectInline traverses the blocks in depth-first order, starting with block. If f returns true,
// InspectInline invokes f recursively for each child of the block, followed by a call of f(nil).
func InspectInline(inline Inline, f func(Inline) bool) {
stack := []Inline{inline}
// Using seen for backtracking
seen := map[Inline]bool{}
for len(stack) > 0 {
// "peek" the node from the stack
inline := stack[len(stack)-1]
if seen[inline] {
// "pop" the node only when backtracking(seen)
stack = stack[:len(stack)-1]
f(nil)
continue
}
seen[inline] = true
// Process the node
if !f(inline) {
continue
}
switch v := inline.(type) {
case *InlineImage:
for i := len(v.Children) - 1; i >= 0; i-- {
stack = append(stack, v.Children[i])
}
case *InlineLink:
for i := len(v.Children) - 1; i >= 0; i-- {
stack = append(stack, v.Children[i])
}
case *ReferenceImage:
for i := len(v.Children) - 1; i >= 0; i-- {
stack = append(stack, v.Children[i])
}
case *ReferenceLink:
for i := len(v.Children) - 1; i >= 0; i-- {
stack = append(stack, v.Children[i])
}
}
}
}