summaryrefslogtreecommitdiffstats
path: root/vendor/github.com/mattermost/mattermost-server/v5/shared/markdown/markdown.go
diff options
context:
space:
mode:
Diffstat (limited to 'vendor/github.com/mattermost/mattermost-server/v5/shared/markdown/markdown.go')
-rw-r--r--vendor/github.com/mattermost/mattermost-server/v5/shared/markdown/markdown.go147
1 files changed, 0 insertions, 147 deletions
diff --git a/vendor/github.com/mattermost/mattermost-server/v5/shared/markdown/markdown.go b/vendor/github.com/mattermost/mattermost-server/v5/shared/markdown/markdown.go
deleted file mode 100644
index 5ccdad8c..00000000
--- a/vendor/github.com/mattermost/mattermost-server/v5/shared/markdown/markdown.go
+++ /dev/null
@@ -1,147 +0,0 @@
-// Copyright (c) 2015-present Mattermost, Inc. All Rights Reserved.
-// See LICENSE.txt for license information.
-
-// This package implements a parser for the subset of the CommonMark spec necessary for us to do
-// server-side processing. It is not a full implementation and lacks many features. But it is
-// complete enough to efficiently and accurately allow us to do what we need to like rewrite image
-// URLs for proxying.
-package markdown
-
-import (
- "strings"
-)
-
-func isEscapable(c rune) bool {
- return c > ' ' && (c < '0' || (c > '9' && (c < 'A' || (c > 'Z' && (c < 'a' || (c > 'z' && c <= '~'))))))
-}
-
-func isEscapableByte(c byte) bool {
- return isEscapable(rune(c))
-}
-
-func isWhitespace(c rune) bool {
- switch c {
- case ' ', '\t', '\n', '\u000b', '\u000c', '\r':
- return true
- }
- return false
-}
-
-func isWhitespaceByte(c byte) bool {
- return isWhitespace(rune(c))
-}
-
-func isNumeric(c rune) bool {
- return c >= '0' && c <= '9'
-}
-
-func isNumericByte(c byte) bool {
- return isNumeric(rune(c))
-}
-
-func isHex(c rune) bool {
- return isNumeric(c) || (c >= 'a' && c <= 'f') || (c >= 'A' && c <= 'F')
-}
-
-func isHexByte(c byte) bool {
- return isHex(rune(c))
-}
-
-func isAlphanumeric(c rune) bool {
- return isNumeric(c) || (c >= 'a' && c <= 'z') || (c >= 'A' && c <= 'Z')
-}
-
-func isAlphanumericByte(c byte) bool {
- return isAlphanumeric(rune(c))
-}
-
-func nextNonWhitespace(markdown string, position int) int {
- for offset, c := range []byte(markdown[position:]) {
- if !isWhitespaceByte(c) {
- return position + offset
- }
- }
- return len(markdown)
-}
-
-func nextLine(markdown string, position int) (linePosition int, skippedNonWhitespace bool) {
- for i := position; i < len(markdown); i++ {
- c := markdown[i]
- if c == '\r' {
- if i+1 < len(markdown) && markdown[i+1] == '\n' {
- return i + 2, skippedNonWhitespace
- }
- return i + 1, skippedNonWhitespace
- } else if c == '\n' {
- return i + 1, skippedNonWhitespace
- } else if !isWhitespaceByte(c) {
- skippedNonWhitespace = true
- }
- }
- return len(markdown), skippedNonWhitespace
-}
-
-func countIndentation(markdown string, r Range) (spaces, bytes int) {
- for i := r.Position; i < r.End; i++ {
- if markdown[i] == ' ' {
- spaces++
- bytes++
- } else if markdown[i] == '\t' {
- spaces += 4
- bytes++
- } else {
- break
- }
- }
- return
-}
-
-func trimLeftSpace(markdown string, r Range) Range {
- s := markdown[r.Position:r.End]
- trimmed := strings.TrimLeftFunc(s, isWhitespace)
- return Range{r.Position, r.End - (len(s) - len(trimmed))}
-}
-
-func trimRightSpace(markdown string, r Range) Range {
- s := markdown[r.Position:r.End]
- trimmed := strings.TrimRightFunc(s, isWhitespace)
- return Range{r.Position, r.End - (len(s) - len(trimmed))}
-}
-
-func relativeToAbsolutePosition(ranges []Range, position int) int {
- rem := position
- for _, r := range ranges {
- l := r.End - r.Position
- if rem < l {
- return r.Position + rem
- }
- rem -= l
- }
- if len(ranges) == 0 {
- return 0
- }
- return ranges[len(ranges)-1].End
-}
-
-func trimBytesFromRanges(ranges []Range, bytes int) (result []Range) {
- rem := bytes
- for _, r := range ranges {
- if rem == 0 {
- result = append(result, r)
- continue
- }
- l := r.End - r.Position
- if rem < l {
- result = append(result, Range{r.Position + rem, r.End})
- rem = 0
- continue
- }
- rem -= l
- }
- return
-}
-
-func Parse(markdown string) (*Document, []*ReferenceDefinition) {
- lines := ParseLines(markdown)
- return ParseBlocks(markdown, lines)
-}