test: Add comprehensive edge case tests for TokenSplitter

textsplitter/token_splitter_edge_cases_test.go

@@ -0,0 +1,354 @@
+package textsplitter
+
+import (
+	"strings"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"github.com/tmc/langchaingo/schema"
+)
+
+func TestTokenSplitterEdgeCases(t *testing.T) {
+	t.Parallel()
+
+	t.Run("empty text", func(t *testing.T) {
+		t.Parallel()
+		splitter := NewTokenSplitter()
+		docs, err := CreateDocuments(splitter, []string{""}, nil)
+		require.NoError(t, err)
+		assert.Equal(t, []schema.Document{
+			{PageContent: "", Metadata: map[string]any{}},
+		}, docs)
+	})
+
+	t.Run("single character", func(t *testing.T) {
+		t.Parallel()
+		splitter := NewTokenSplitter(WithChunkSize(1))
+		docs, err := CreateDocuments(splitter, []string{"a"}, nil)
+		require.NoError(t, err)
+		assert.Equal(t, []schema.Document{
+			{PageContent: "a", Metadata: map[string]any{}},
+		}, docs)
+	})
+
+	t.Run("whitespace only", func(t *testing.T) {
+		t.Parallel()
+		splitter := NewTokenSplitter(WithChunkSize(10))
+		docs, err := CreateDocuments(splitter, []string{"   \n\t  "}, nil)
+		require.NoError(t, err)
+		assert.Len(t, docs, 1)
+		assert.Equal(t, "   \n\t  ", docs[0].PageContent)
+	})
+
+	t.Run("very large chunk size", func(t *testing.T) {
+		t.Parallel()
+		text := "This is a test text that should not be split because the chunk size is very large."
+		splitter := NewTokenSplitter(WithChunkSize(10000))
+		docs, err := CreateDocuments(splitter, []string{text}, nil)
+		require.NoError(t, err)
+		assert.Len(t, docs, 1)
+		assert.Equal(t, text, docs[0].PageContent)
+	})
+
+	t.Run("zero chunk overlap", func(t *testing.T) {
+		t.Parallel()
+		text := "This is a longer text that should be split into multiple chunks without any overlap between them."
+		splitter := NewTokenSplitter(
+			WithChunkSize(10),
+			WithChunkOverlap(0),
+		)
+		docs, err := CreateDocuments(splitter, []string{text}, nil)
+		require.NoError(t, err)
+		assert.Greater(t, len(docs), 1)
+
+		// Verify no overlap by checking that no two consecutive chunks share content
+		for i := 1; i < len(docs); i++ {
+			prev := strings.TrimSpace(docs[i-1].PageContent)
+			curr := strings.TrimSpace(docs[i].PageContent)
+			if prev != "" && curr != "" {
+				// Should not have overlapping words at boundaries
+				prevWords := strings.Fields(prev)
+				currWords := strings.Fields(curr)
+				if len(prevWords) > 0 && len(currWords) > 0 {
+					assert.NotEqual(t, prevWords[len(prevWords)-1], currWords[0],
+						"Chunks should not overlap when overlap is 0")
+				}
+			}
+		}
+	})
+
+	t.Run("chunk overlap equals chunk size", func(t *testing.T) {
+		t.Parallel()
+		text := "Word1 Word2 Word3 Word4 Word5 Word6 Word7 Word8"
+		splitter := NewTokenSplitter(
+			WithChunkSize(5),
+			WithChunkOverlap(5),
+		)
+		docs, err := CreateDocuments(splitter, []string{text}, nil)
+		require.NoError(t, err)
+		assert.Greater(t, len(docs), 1)
+
+		// With overlap equal to chunk size, chunks should have significant overlap
+		for i := 1; i < len(docs); i++ {
+			assert.NotEmpty(t, docs[i].PageContent)
+		}
+	})
+
+	t.Run("unicode and special characters", func(t *testing.T) {
+		t.Parallel()
+		text := "Hello 世界! 🌍 This contains émojis and spëcial characters: àáâãäå"
+		splitter := NewTokenSplitter(WithChunkSize(20))
+		docs, err := CreateDocuments(splitter, []string{text}, nil)
+		require.NoError(t, err)
+		assert.NotEmpty(t, docs)
+
+		// Verify all content is preserved
+		combined := ""
+		for _, doc := range docs {
+			combined += doc.PageContent
+		}
+		// Remove potential whitespace differences for comparison
+		assert.Contains(t, combined, "世界")
+		assert.Contains(t, combined, "🌍")
+		assert.Contains(t, combined, "émojis")
+		assert.Contains(t, combined, "àáâãäå")
+	})
+
+	t.Run("very long single word", func(t *testing.T) {
+		t.Parallel()
+		longWord := strings.Repeat("a", 1000)
+		splitter := NewTokenSplitter(WithChunkSize(10))
+		docs, err := CreateDocuments(splitter, []string{longWord}, nil)
+		require.NoError(t, err)
+		assert.Greater(t, len(docs), 1)
+
+		// Verify the long word is split appropriately
+		combined := ""
+		for _, doc := range docs {
+			combined += doc.PageContent
+		}
+		assert.Equal(t, longWord, combined)
+	})
+
+	t.Run("newlines and formatting preservation", func(t *testing.T) {
+		t.Parallel()
+		text := "Line 1\n\nLine 2\n\n\nLine 3\n\t\tIndented line\n"
+		splitter := NewTokenSplitter(WithChunkSize(15))
+		docs, err := CreateDocuments(splitter, []string{text}, nil)
+		require.NoError(t, err)
+		assert.NotEmpty(t, docs)
+
+		// Verify formatting is preserved in the split
+		combined := ""
+		for _, doc := range docs {
+			combined += doc.PageContent
+		}
+		assert.Contains(t, combined, "\n\n")
+		assert.Contains(t, combined, "\t\t")
+	})
+}
+
+func TestTokenSplitterDifferentModels(t *testing.T) {
+	t.Parallel()
+
+	testCases := []struct {
+		name      string
+		modelName string
+		text      string
+	}{
+		{
+			name:      "gpt-4 model",
+			modelName: "gpt-4",
+			text:      "This is a test for GPT-4 tokenization.",
+		},
+		{
+			name:      "gpt-3.5-turbo model",
+			modelName: "gpt-3.5-turbo",
+			text:      "This is a test for GPT-3.5-turbo tokenization.",
+		},
+		{
+			name:      "text-davinci-003 model",
+			modelName: "text-davinci-003",
+			text:      "This is a test for text-davinci-003 tokenization.",
+		},
+	}
+
+	for _, tc := range testCases {
+		tc := tc
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			splitter := NewTokenSplitter(
+				WithModelName(tc.modelName),
+				WithChunkSize(10),
+			)
+			docs, err := CreateDocuments(splitter, []string{tc.text}, nil)
+			require.NoError(t, err)
+			assert.NotEmpty(t, docs)
+
+			// Verify content is preserved
+			combined := ""
+			for _, doc := range docs {
+				combined += doc.PageContent
+			}
+			assert.Contains(t, combined, "test")
+			assert.Contains(t, combined, "tokenization")
+		})
+	}
+}
+
+func TestTokenSplitterDifferentEncodings(t *testing.T) {
+	t.Parallel()
+
+	testCases := []struct {
+		name         string
+		encodingName string
+		text         string
+	}{
+		{
+			name:         "cl100k_base encoding",
+			encodingName: "cl100k_base",
+			text:         "Testing cl100k_base encoding with various tokens.",
+		},
+		{
+			name:         "p50k_base encoding",
+			encodingName: "p50k_base",
+			text:         "Testing p50k_base encoding with various tokens.",
+		},
+		{
+			name:         "r50k_base encoding",
+			encodingName: "r50k_base",
+			text:         "Testing r50k_base encoding with various tokens.",
+		},
+	}
+
+	for _, tc := range testCases {
+		tc := tc
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			splitter := NewTokenSplitter(
+				WithEncodingName(tc.encodingName),
+				WithChunkSize(15),
+			)
+			docs, err := CreateDocuments(splitter, []string{tc.text}, nil)
+			require.NoError(t, err)
+			assert.NotEmpty(t, docs)
+
+			// Verify content is preserved
+			combined := ""
+			for _, doc := range docs {
+				combined += doc.PageContent
+			}
+			assert.Contains(t, combined, "Testing")
+			assert.Contains(t, combined, "encoding")
+		})
+	}
+}
+
+func TestTokenSplitterSpecialTokens(t *testing.T) {
+	t.Parallel()
+
+	t.Run("with allowed special tokens", func(t *testing.T) {
+		t.Parallel()
+		text := "This text contains <|endoftext|> special token."
+		splitter := NewTokenSplitter(
+			WithAllowedSpecial([]string{"<|endoftext|>"}),
+			WithChunkSize(20),
+		)
+		docs, err := CreateDocuments(splitter, []string{text}, nil)
+		require.NoError(t, err)
+		assert.NotEmpty(t, docs)
+
+		// Verify special token is preserved
+		combined := ""
+		for _, doc := range docs {
+			combined += doc.PageContent
+		}
+		assert.Contains(t, combined, "<|endoftext|>")
+	})
+
+	t.Run("with disallowed special tokens", func(t *testing.T) {
+		t.Parallel()
+		text := "This is normal text without special tokens."
+		splitter := NewTokenSplitter(
+			WithDisallowedSpecial([]string{"all"}),
+			WithChunkSize(20),
+		)
+		docs, err := CreateDocuments(splitter, []string{text}, nil)
+		require.NoError(t, err)
+		assert.NotEmpty(t, docs)
+
+		// Verify content is preserved
+		combined := ""
+		for _, doc := range docs {
+			combined += doc.PageContent
+		}
+		assert.Contains(t, combined, "normal text")
+	})
+}
+
+func TestTokenSplitterErrorHandling(t *testing.T) {
+	t.Parallel()
+
+	t.Run("invalid model name", func(t *testing.T) {
+		t.Parallel()
+		splitter := NewTokenSplitter(WithModelName("invalid-model-name"))
+		_, err := CreateDocuments(splitter, []string{"test"}, nil)
+		assert.Error(t, err)
+		assert.Contains(t, err.Error(), "tiktoken")
+	})
+
+	t.Run("invalid encoding name", func(t *testing.T) {
+		t.Parallel()
+		splitter := NewTokenSplitter(WithEncodingName("invalid-encoding"))
+		_, err := CreateDocuments(splitter, []string{"test"}, nil)
+		assert.Error(t, err)
+		assert.Contains(t, err.Error(), "tiktoken")
+	})
+}
+
+func TestTokenSplitterConsistency(t *testing.T) {
+	t.Parallel()
+
+	t.Run("consistent splitting", func(t *testing.T) {
+		t.Parallel()
+		text := "This is a consistent test text that should be split the same way every time."
+		splitter := NewTokenSplitter(
+			WithChunkSize(10),
+			WithChunkOverlap(2),
+		)
+
+		// Split the same text multiple times
+		docs1, err1 := CreateDocuments(splitter, []string{text}, nil)
+		require.NoError(t, err1)
+
+		docs2, err2 := CreateDocuments(splitter, []string{text}, nil)
+		require.NoError(t, err2)
+
+		// Results should be identical
+		assert.Equal(t, docs1, docs2)
+	})
+
+	t.Run("order preservation", func(t *testing.T) {
+		t.Parallel()
+		text := "First sentence. Second sentence. Third sentence. Fourth sentence."
+		splitter := NewTokenSplitter(WithChunkSize(8))
+		docs, err := CreateDocuments(splitter, []string{text}, nil)
+		require.NoError(t, err)
+
+		// Verify order is preserved
+		combined := ""
+		for _, doc := range docs {
+			combined += doc.PageContent
+		}
+
+		firstPos := strings.Index(combined, "First")
+		secondPos := strings.Index(combined, "Second")
+		thirdPos := strings.Index(combined, "Third")
+		fourthPos := strings.Index(combined, "Fourth")
+
+		assert.True(t, firstPos < secondPos)
+		assert.True(t, secondPos < thirdPos)
+		assert.True(t, thirdPos < fourthPos)
+	})
+}