X-Git-Url: http://git.giacomolli.uk/onechat/blobdiff_plain/09cde69fdffe2721434392036f98dc241177f5a1..cb514b662bcf9d90b2cf02789c989f777d774d3f:/telegram_markdown.go?ds=inline

diff --git a/telegram_markdown.go b/telegram_markdown.go
index 54bfbcb..8738bcd 100644
--- a/telegram_markdown.go
+++ b/telegram_markdown.go
@@ -11,6 +11,15 @@ import (
 // plain.
 const telegramMarkdownPunctuation = "!\"#$%&'()*+,-./:;<=>?@[\\]^_`{|}~"
 
+var (
+	// telegramTagEscaper writes the characters that rich Markdown reads as HTML
+	// as HTML entities.
+	telegramTagEscaper = strings.NewReplacer("&", "&amp;", "<", "&lt;")
+
+	// telegramTagUnescaper reverses telegramTagEscaper.
+	telegramTagUnescaper = strings.NewReplacer("&amp;", "&", "&lt;", "<")
+)
+
 // telegramMarkdownEntities converts the Markdown of a reply to plain text and
 // Telegram message entities. It knows bold (** and __), italic (* and _),
 // strikethrough (~~), inline code, fenced code blocks with an optional
@@ -68,6 +77,23 @@ func telegramMarkdownEntities(markdown string) (string, []TelegramEntity) {
 	return out.text.String(), out.entities
 }
 
+// telegramEscapeTags writes each & of markdown outside inline code and fenced
+// code blocks as &amp; and each < as &lt;. Rich Markdown reads raw HTML and
+// HTML entities in prose, and it does not read a backslash before < as an
+// escape. With the entities, it shows a tag name and an entity that the model
+// wrote in prose as text.
+func telegramEscapeTags(markdown string) string {
+	return telegramRewriteProse(markdown, telegramTagEscaper.Replace)
+}
+
+// telegramUnescapeTags gives back the Markdown that telegramEscapeTags escaped.
+// It is exact, because after the escape each & outside code starts &amp; or
+// &lt;. It also works on a part of telegramSplitRich, because a part has the
+// same code as the text that it comes from.
+func telegramUnescapeTags(markdown string) string {
+	return telegramRewriteProse(markdown, telegramTagUnescaper.Replace)
+}
+
 // telegramCodeFence returns the language of a line that opens a fenced code
 // block, and the fence. The fence is "" when line does not open a block.
 func telegramCodeFence(line string) (string, string) {
@@ -225,6 +251,69 @@ func telegramHeadingText(line string) string {
 	return rest
 }
 
+// telegramRewriteProse returns markdown with rewrite applied to each part of a
+// line outside inline code and fenced code blocks. A code block that does not
+// close runs to the end, as in Markdown.
+func telegramRewriteProse(markdown string, rewrite func(string) string) string {
+	var out strings.Builder
+	lines := strings.Split(markdown, "\n")
+	for index := 0; index < len(lines); index++ {
+		if index > 0 {
+			out.WriteString("\n")
+		}
+
+		line := lines[index]
+		_, fence := telegramCodeFence(line)
+		if fence == "" {
+			telegramRewriteLineProse(&out, line, rewrite)
+			continue
+		}
+
+		end := telegramClosingFence(lines, index, fence)
+		if end < 0 {
+			end = len(lines) - 1
+		}
+
+		out.WriteString(strings.Join(lines[index:end+1], "\n"))
+		index = end
+	}
+
+	return out.String()
+}
+
+// telegramRewriteLineProse writes line to out with rewrite applied to each part
+// outside inline code. A backtick after a backslash opens no code. A run of
+// backticks that opens no inline code stays as text whole, as in Markdown.
+func telegramRewriteLineProse(out *strings.Builder, line string, rewrite func(string) string) {
+	prose := 0
+	for index := 0; index < len(line); {
+		switch line[index] {
+		case '\\':
+			if index+1 < len(line) && strings.IndexByte(telegramMarkdownPunctuation, line[index+1]) >= 0 {
+				index += 2
+				continue
+			}
+		case '`':
+			end := index + telegramBackticks(line[index:])
+			span, found := telegramCodeSpan(line, index)
+			if found {
+				end = span.end
+			}
+
+			out.WriteString(rewrite(line[prose:index]))
+			out.WriteString(line[index:end])
+			index = end
+			prose = end
+
+			continue
+		}
+
+		index++
+	}
+
+	out.WriteString(rewrite(line[prose:]))
+}
+
 // telegramMarkSpan returns the inline mark that starts at index of line and
 // closes in line. It returns false when no mark starts there or the mark does
 // not close.