From 13a2573ca260f18935af39cc5bf81c08ee46d7d4 Mon Sep 17 00:00:00 2001
From: Vyr Cossont <VyrCossont@users.noreply.github.com>
Date: Mon, 27 Jan 2025 09:41:10 -0800
Subject: [PATCH] Allow marks after NFC normalization

Includes regression test for the Tamil example from #3618
---
 internal/text/goldmark_parsers.go |  2 +-
 internal/text/markdown_test.go    |  8 ++++++++
 internal/text/normalize.go        | 17 ++++++++---------
 internal/text/plain_test.go       |  7 +++++--
 internal/text/util.go             | 17 +++++++----------
 5 files changed, 29 insertions(+), 22 deletions(-)

diff --git a/internal/text/goldmark_parsers.go b/internal/text/goldmark_parsers.go
index b7cf4f9e9..e2c87e057 100644
--- a/internal/text/goldmark_parsers.go
+++ b/internal/text/goldmark_parsers.go
@@ -177,7 +177,7 @@ func (p *hashtagParser) Parse(
 			// Ignore initial '#'.
 			continue
 
-		case !isPlausiblyInHashtag(r) &&
+		case !isPermittedInHashtag(r) &&
 			!isHashtagBoundary(r):
 			// Weird non-boundary character
 			// in the hashtag. Don't trust it.
diff --git a/internal/text/markdown_test.go b/internal/text/markdown_test.go
index 98ed3a96b..153673415 100644
--- a/internal/text/markdown_test.go
+++ b/internal/text/markdown_test.go
@@ -50,6 +50,8 @@ const (
 	withInlineCode2Expected         = "<p><code>Nobody tells you about the &lt;/code>&lt;del>SECRET CODE&lt;/del>&lt;code>, do they?</code></p>"
 	withHashtag                     = "# Title\n\nhere's a simple status that uses hashtag #Hashtag!"
 	withHashtagExpected             = "<h1>Title</h1><p>here's a simple status that uses hashtag <a href=\"http://localhost:8080/tags/hashtag\" class=\"mention hashtag\" rel=\"tag nofollow noreferrer noopener\" target=\"_blank\">#<span>Hashtag</span></a>!</p>"
+	withTamilHashtag                = "here's a simple status that uses a hashtag in Tamil #தமிழ்"
+	withTamilHashtagExpected        = "<p>here's a simple status that uses a hashtag in Tamil <a href=\"http://localhost:8080/tags/%E0%AE%A4%E0%AE%AE%E0%AE%BF%E0%AE%B4%E0%AF%8D\" class=\"mention hashtag\" rel=\"tag nofollow noreferrer noopener\" target=\"_blank\">#<span>தமிழ்</span></a></p>"
 	mdWithHTML                      = "# Title\n\nHere's a simple text in markdown.\n\nHere's a <a href=\"https://example.org\">link</a>.\n\nHere's an image: <img src=\"https://gts.superseriousbusiness.org/assets/logo.png\" alt=\"The GoToSocial sloth logo.\" width=\"500\" height=\"600\">"
 	mdWithHTMLExpected              = "<h1>Title</h1><p>Here's a simple text in markdown.</p><p>Here's a <a href=\"https://example.org\" rel=\"nofollow noreferrer noopener\" target=\"_blank\">link</a>.</p><p>Here's an image:</p>"
 	mdWithCheekyHTML                = "# Title\n\nHere's a simple text in markdown.\n\nHere's a cheeky little script: <script>alert(ahhhh)</script>"
@@ -121,6 +123,12 @@ func (suite *MarkdownTestSuite) TestParseWithHashtag() {
 	suite.Equal(withHashtagExpected, formatted.HTML)
 }
 
+// Regressiom test for https://github.com/superseriousbusiness/gotosocial/issues/3618
+func (suite *MarkdownTestSuite) TestParseWithTamilHashtag() {
+	formatted := suite.FromMarkdown(withTamilHashtag)
+	suite.Equal(withTamilHashtagExpected, formatted.HTML)
+}
+
 func (suite *MarkdownTestSuite) TestParseWithHTML() {
 	formatted := suite.FromMarkdown(mdWithHTML)
 	suite.Equal(mdWithHTMLExpected, formatted.HTML)
diff --git a/internal/text/normalize.go b/internal/text/normalize.go
index d2e633d1e..ea266fb33 100644
--- a/internal/text/normalize.go
+++ b/internal/text/normalize.go
@@ -50,17 +50,16 @@ func NormalizeHashtag(text string) (string, bool) {
 
 	// Validate normalized result.
 	var (
-		notJustUnderscores = false
-		onlyPermittedChars = true
-		lengthOK           = true
+		atLeastOneRequiredChar = false
+		onlyPermittedChars     = true
+		lengthOK               = true
 	)
 
 	for i, r := range normalized {
-		if r != '_' {
-			// This isn't an underscore,
-			// so the whole hashtag isn't
-			// just underscores.
-			notJustUnderscores = true
+		if !isPermittedIfNotEntireHashtag(r) {
+			// This isn't an underscore, mark, etc,
+			// so the hashtag contains at least one
+			atLeastOneRequiredChar = true
 		}
 
 		if i >= maximumHashtagLength {
@@ -74,5 +73,5 @@ func NormalizeHashtag(text string) (string, bool) {
 		}
 	}
 
-	return normalized, (lengthOK && onlyPermittedChars && notJustUnderscores)
+	return normalized, lengthOK && onlyPermittedChars && atLeastOneRequiredChar
 }
diff --git a/internal/text/plain_test.go b/internal/text/plain_test.go
index fac54a38e..50b13cc4d 100644
--- a/internal/text/plain_test.go
+++ b/internal/text/plain_test.go
@@ -170,8 +170,11 @@ func (suite *PlainTestSuite) TestDeriveMultiple() {
 func (suite *PlainTestSuite) TestZalgoHashtag() {
 	statusText := `yo who else loves #praying to #z̸͉̅a̸͚͋l̵͈̊g̸̫͌ỏ̷̪?`
 	f := suite.FromPlain(statusText)
-	suite.Len(f.Tags, 1)
-	suite.Equal("praying", f.Tags[0].Name)
+	if suite.Len(f.Tags, 2) {
+		suite.Equal("praying", f.Tags[0].Name)
+		// NFC doesn't do much for Zalgo text, but it's difficult to strip marks without affecting non-Latin text.
+		suite.Equal("z̸͉̅a̸͚͋l̵͈̊g̸̫͌ỏ̷̪", f.Tags[1].Name)
+	}
 }
 
 func TestPlainTestSuite(t *testing.T) {
diff --git a/internal/text/util.go b/internal/text/util.go
index af45cfaf0..6178766a6 100644
--- a/internal/text/util.go
+++ b/internal/text/util.go
@@ -19,19 +19,16 @@ package text
 
 import "unicode"
 
-func isPlausiblyInHashtag(r rune) bool {
-	// Marks are allowed during parsing
-	// prior to normalization, but not after,
-	// since they may be combined into letters
-	// during normalization.
-	return unicode.IsMark(r) ||
-		isPermittedInHashtag(r)
-}
-
 func isPermittedInHashtag(r rune) bool {
 	return unicode.IsLetter(r) ||
 		unicode.IsNumber(r) ||
-		r == '_'
+		isPermittedIfNotEntireHashtag(r)
+}
+
+// isPermittedIfNotEntireHashtag is true for characters that may be in a hashtag
+// but are not allowed to be the only characters making up the hashtag.
+func isPermittedIfNotEntireHashtag(r rune) bool {
+	return unicode.IsMark(r) || r == '_'
 }
 
 // isHashtagBoundary returns true if rune r