[bugfix] Extend parser to handle more non-Latin hashtags (#3700)

* Allow marks after NFC normalization

Includes regression test for the Tamil example from #3618

* Disallow just numbers + marks + underscore as hashtag

5 files changed, 48 insertions, 37 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..ffa64ce44 100644
--- a/internal/text/plain_test.go
+++ b/internal/text/plain_test.go
@@ -118,20 +118,20 @@ func (suite *PlainTestSuite) TestDeriveHashtagsOK() {
 `
 
 	tags := suite.FromPlain(statusText).Tags
-	suite.Len(tags, 13)
-	suite.Equal("testing123", tags[0].Name)
-	suite.Equal("also", tags[1].Name)
-	suite.Equal("thisshouldwork", tags[2].Name)
-	suite.Equal("dupe", tags[3].Name)
-	suite.Equal("ThisShouldAlsoWork", tags[4].Name)
-	suite.Equal("this_should_not_be_split", tags[5].Name)
-	suite.Equal("111111", tags[6].Name)
-	suite.Equal("alimentación", tags[7].Name)
-	suite.Equal("saúde", tags[8].Name)
-	suite.Equal("lävistää", tags[9].Name)
-	suite.Equal("ö", tags[10].Name)
-	suite.Equal("네", tags[11].Name)
-	suite.Equal("ThisOneIsThirteyCharactersLong", tags[12].Name)
+	if suite.Len(tags, 12) {
+		suite.Equal("testing123", tags[0].Name)
+		suite.Equal("also", tags[1].Name)
+		suite.Equal("thisshouldwork", tags[2].Name)
+		suite.Equal("dupe", tags[3].Name)
+		suite.Equal("ThisShouldAlsoWork", tags[4].Name)
+		suite.Equal("this_should_not_be_split", tags[5].Name)
+		suite.Equal("alimentación", tags[6].Name)
+		suite.Equal("saúde", tags[7].Name)
+		suite.Equal("lävistää", tags[8].Name)
+		suite.Equal("ö", tags[9].Name)
+		suite.Equal("네", tags[10].Name)
+		suite.Equal("ThisOneIsThirteyCharactersLong", tags[11].Name)
+	}
 
 	statusText = `#올빼미 hej`
 	tags = suite.FromPlain(statusText).Tags
@@ -170,8 +170,17 @@ 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 (suite *PlainTestSuite) TestNumbersAreNotHashtags() {
+	statusText := `yo who else thinks #19_98 is #1?`
+	f := suite.FromPlain(statusText)
+	suite.Len(f.Tags, 0)
 }
 
 func TestPlainTestSuite(t *testing.T) {
diff --git a/internal/text/util.go b/internal/text/util.go
index af45cfaf0..47b2416dd 100644
--- a/internal/text/util.go
+++ b/internal/text/util.go
@@ -19,19 +19,14 @@ 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) || isPermittedIfNotEntireHashtag(r)
 }
 
-func isPermittedInHashtag(r rune) bool {
-	return unicode.IsLetter(r) ||
-		unicode.IsNumber(r) ||
-		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.IsNumber(r) || unicode.IsMark(r) || r == '_'
 }
 
 // isHashtagBoundary returns true if rune r