X-Git-Url: https://git.giacomolli.uk/onechat/blobdiff_plain/f80b058ae0550a56b4e62f8e65c2a74349b970e4..b64d7548bd9f61921bdc650228642e77c73b0777:/telegram.go?ds=sidebyside

diff --git a/telegram.go b/telegram.go
index 9a24a4e..9dbe86e 100644
--- a/telegram.go
+++ b/telegram.go
@@ -14,6 +14,7 @@ import (
 	"strings"
 	"sync"
 	"time"
+	"unicode/utf8"
 )
 
 const (
@@ -33,6 +34,12 @@ const (
 	// messageUnits is the most UTF-16 code units of one Telegram message.
 	messageUnits = 4096
 
+	// richMessageBytes is the most bytes of the Markdown of one rich
+	// message. The Bot API allows 32768 characters of rendered text, and
+	// the Markdown of a text has at least as many bytes as the text has
+	// characters.
+	richMessageBytes = 32768
+
 	// typingInterval is how often the bot shows the typing action while a
 	// turn runs. Telegram shows the action for about five seconds.
 	typingInterval = 4 * time.Second
@@ -548,3 +555,83 @@ func prefixEnd(text string, limit int) int {
 
 	return len(text)
 }
+
+// splitRich cuts the Markdown of a reply into parts of at most limit bytes. It
+// cuts after a blank line outside a code block, so a table, a code block, or a
+// paragraph stays whole. When no such cut fits in a part, it cuts after the
+// last line break outside a code block, then after the last line break in a
+// code block, which it closes at the end of the part and opens again in the
+// next part with the same fence line. A line longer than limit is cut at the
+// limit. It never cuts a character.
+func splitRich(markdown string, limit int) []string {
+	var parts []string
+	for len(markdown) > limit {
+		cut, fence, opening := richCut(markdown, limit)
+		if fence == "" {
+			parts = append(parts, markdown[:cut])
+			markdown = markdown[cut:]
+
+			continue
+		}
+
+		parts = append(parts, markdown[:cut]+fence+"\n")
+		markdown = opening + "\n" + markdown[cut:]
+	}
+
+	return append(parts, markdown)
+}
+
+// richCut returns where splitRich cuts markdown, which is longer than limit.
+// For a cut in a code block, it also returns the fence that closes the block
+// and the line that opens it. Each part before a cut holds text other than
+// space, and a cut in a code block leaves room for the fence, so each cut makes
+// the rest shorter.
+func richCut(markdown string, limit int) (int, string, string) {
+	blank, line, code := 0, 0, 0
+	var fence, opening, inFence, inOpening string
+	for start := 0; start < limit; {
+		end := strings.IndexByte(markdown[start:], '\n')
+		if end < 0 || start+end+1 > limit {
+			break
+		}
+
+		end += start + 1
+		text := strings.TrimRight(markdown[start:end], "\r\n")
+		_, opens := codeFence(text)
+		switch {
+		case fence != "":
+			if closesFence(text, fence) {
+				fence = ""
+			} else if end > len(opening)+1 && end+len(fence)+1 <= limit {
+				code, inFence, inOpening = end, fence, opening
+			}
+		case opens != "":
+			fence, opening = opens, text
+		case strings.TrimSpace(markdown[:end]) == "":
+		case strings.TrimSpace(text) == "":
+			blank = end
+		}
+
+		if fence == "" && strings.TrimSpace(markdown[:end]) != "" {
+			line = end
+		}
+
+		start = end
+	}
+
+	switch {
+	case blank > 0:
+		return blank, "", ""
+	case line > 0:
+		return line, "", ""
+	case code > 0:
+		return code, inFence, inOpening
+	}
+
+	cut := limit
+	for !utf8.RuneStart(markdown[cut]) {
+		cut--
+	}
+
+	return cut, "", ""
+}