|
| 1 | +"""Tests for multi-part content handling in langchain_messages_to_copilotkit. |
| 2 | +
|
| 3 | +Covers the fix in PR #3844 / issue #1748: when AIMessage.content is a list |
| 4 | +of content blocks (e.g. Anthropic models), all text parts must be extracted |
| 5 | +and concatenated — not just the first element. |
| 6 | +""" |
| 7 | + |
| 8 | +import pytest |
| 9 | +from langchain_core.messages import AIMessage, HumanMessage, SystemMessage |
| 10 | + |
| 11 | +from copilotkit.langgraph import langchain_messages_to_copilotkit |
| 12 | + |
| 13 | + |
| 14 | +class TestMultiPartContentList: |
| 15 | + """AIMessage.content as a list should concatenate all text parts.""" |
| 16 | + |
| 17 | + def test_list_of_text_dicts(self): |
| 18 | + """Multiple {"type": "text", "text": "..."} dicts are all concatenated.""" |
| 19 | + msg = AIMessage( |
| 20 | + id="ai-1", |
| 21 | + content=[ |
| 22 | + {"type": "text", "text": "Hello "}, |
| 23 | + {"type": "text", "text": "world"}, |
| 24 | + ], |
| 25 | + ) |
| 26 | + result = langchain_messages_to_copilotkit([msg]) |
| 27 | + assert len(result) == 1 |
| 28 | + assert result[0]["content"] == "Hello world" |
| 29 | + assert result[0]["role"] == "assistant" |
| 30 | + |
| 31 | + def test_list_of_strings(self): |
| 32 | + """Content list of plain strings should be concatenated.""" |
| 33 | + msg = AIMessage( |
| 34 | + id="ai-2", |
| 35 | + content=["Part A", " Part B"], |
| 36 | + ) |
| 37 | + result = langchain_messages_to_copilotkit([msg]) |
| 38 | + assert len(result) == 1 |
| 39 | + assert result[0]["content"] == "Part A Part B" |
| 40 | + |
| 41 | + def test_mixed_strings_and_text_dicts(self): |
| 42 | + """Mix of plain strings and text dicts should all be concatenated.""" |
| 43 | + msg = AIMessage( |
| 44 | + id="ai-3", |
| 45 | + content=[ |
| 46 | + "Start ", |
| 47 | + {"type": "text", "text": "middle "}, |
| 48 | + {"text": "end"}, |
| 49 | + ], |
| 50 | + ) |
| 51 | + result = langchain_messages_to_copilotkit([msg]) |
| 52 | + assert len(result) == 1 |
| 53 | + assert result[0]["content"] == "Start middle end" |
| 54 | + |
| 55 | + def test_non_text_parts_are_skipped(self): |
| 56 | + """Non-text content blocks (e.g. images) should be ignored.""" |
| 57 | + msg = AIMessage( |
| 58 | + id="ai-4", |
| 59 | + content=[ |
| 60 | + {"type": "text", "text": "Sample png file"}, |
| 61 | + { |
| 62 | + "type": "image", |
| 63 | + "image_data": {"data": "base64data", "format": "image/png"}, |
| 64 | + }, |
| 65 | + ], |
| 66 | + ) |
| 67 | + result = langchain_messages_to_copilotkit([msg]) |
| 68 | + assert len(result) == 1 |
| 69 | + assert result[0]["content"] == "Sample png file" |
| 70 | + |
| 71 | + def test_empty_list_returns_empty_content(self): |
| 72 | + """Empty content list should produce empty string, not crash.""" |
| 73 | + msg = AIMessage( |
| 74 | + id="ai-5", |
| 75 | + content=[], |
| 76 | + ) |
| 77 | + result = langchain_messages_to_copilotkit([msg]) |
| 78 | + # Empty content means message is not included (content is falsy) |
| 79 | + assert len(result) == 0 |
| 80 | + |
| 81 | + def test_single_text_dict_in_list(self): |
| 82 | + """Single text dict in a list should still be extracted.""" |
| 83 | + msg = AIMessage( |
| 84 | + id="ai-6", |
| 85 | + content=[{"type": "text", "text": "Only one part"}], |
| 86 | + ) |
| 87 | + result = langchain_messages_to_copilotkit([msg]) |
| 88 | + assert len(result) == 1 |
| 89 | + assert result[0]["content"] == "Only one part" |
| 90 | + |
| 91 | + def test_dict_without_type_but_with_text_key(self): |
| 92 | + """A dict with "text" key but no "type" should still have text extracted.""" |
| 93 | + msg = AIMessage( |
| 94 | + id="ai-7", |
| 95 | + content=[{"text": "no type field"}], |
| 96 | + ) |
| 97 | + result = langchain_messages_to_copilotkit([msg]) |
| 98 | + assert len(result) == 1 |
| 99 | + assert result[0]["content"] == "no type field" |
| 100 | + |
| 101 | + |
| 102 | +class TestSingleDictContent: |
| 103 | + """AIMessage.content as a single dict (Anthropic style) should extract text. |
| 104 | +
|
| 105 | + Note: langchain_core.messages.AIMessage validates content as str | list, |
| 106 | + so a raw dict cannot be passed directly. We use a mock to exercise the |
| 107 | + dict-handling code path in langchain_messages_to_copilotkit, which exists |
| 108 | + to handle edge cases from deserialized or non-standard message objects. |
| 109 | + """ |
| 110 | + |
| 111 | + def test_dict_with_text_key(self): |
| 112 | + """A content dict with "text" key should have its text extracted.""" |
| 113 | + from unittest.mock import MagicMock |
| 114 | + |
| 115 | + msg = MagicMock(spec=AIMessage) |
| 116 | + msg.content = {"text": "dict content"} |
| 117 | + msg.id = "ai-8" |
| 118 | + msg.tool_calls = [] |
| 119 | + result = langchain_messages_to_copilotkit([msg]) |
| 120 | + assert len(result) == 1 |
| 121 | + assert result[0]["content"] == "dict content" |
| 122 | + |
| 123 | + |
| 124 | +class TestPlainStringContent: |
| 125 | + """Standard string content should still work as before.""" |
| 126 | + |
| 127 | + def test_plain_string_content(self): |
| 128 | + """Normal string content passes through unchanged.""" |
| 129 | + msg = AIMessage( |
| 130 | + id="ai-9", |
| 131 | + content="Just a string", |
| 132 | + ) |
| 133 | + result = langchain_messages_to_copilotkit([msg]) |
| 134 | + assert len(result) == 1 |
| 135 | + assert result[0]["content"] == "Just a string" |
| 136 | + |
| 137 | + def test_human_message_string(self): |
| 138 | + """HumanMessage with string content still works.""" |
| 139 | + msg = HumanMessage(id="human-1", content="Hello") |
| 140 | + result = langchain_messages_to_copilotkit([msg]) |
| 141 | + assert len(result) == 1 |
| 142 | + assert result[0]["role"] == "user" |
| 143 | + assert result[0]["content"] == "Hello" |
| 144 | + |
| 145 | + def test_system_message_string(self): |
| 146 | + """SystemMessage with string content still works.""" |
| 147 | + msg = SystemMessage(id="sys-1", content="System prompt") |
| 148 | + result = langchain_messages_to_copilotkit([msg]) |
| 149 | + assert len(result) == 1 |
| 150 | + assert result[0]["role"] == "system" |
| 151 | + assert result[0]["content"] == "System prompt" |
| 152 | + |
| 153 | + |
| 154 | +class TestIssue1748Reproduction: |
| 155 | + """Directly reproduces the scenario from issue #1748. |
| 156 | +
|
| 157 | + The original bug: when content is a list of dicts including an image block, |
| 158 | + only the first element was taken via `content[0]`, which was the dict itself, |
| 159 | + not a string. This caused the message to be silently dropped or mangled. |
| 160 | + """ |
| 161 | + |
| 162 | + def test_text_and_image_content_preserves_text(self): |
| 163 | + """The exact scenario from issue #1748: text + image content blocks.""" |
| 164 | + msg = AIMessage( |
| 165 | + id="ai-repro", |
| 166 | + content=[ |
| 167 | + {"type": "text", "text": "Sample png file"}, |
| 168 | + { |
| 169 | + "type": "image", |
| 170 | + "image_data": {"data": "aW1hZ2VfZGF0YQ==", "format": "image/png"}, |
| 171 | + }, |
| 172 | + ], |
| 173 | + ) |
| 174 | + result = langchain_messages_to_copilotkit([msg]) |
| 175 | + assert len(result) == 1 |
| 176 | + assert result[0]["content"] == "Sample png file" |
| 177 | + assert result[0]["role"] == "assistant" |
| 178 | + |
| 179 | + def test_multiple_text_parts_are_not_truncated(self): |
| 180 | + """The core bug: only the first element was kept. All text must survive.""" |
| 181 | + msg = AIMessage( |
| 182 | + id="ai-trunc", |
| 183 | + content=[ |
| 184 | + {"type": "text", "text": "First part. "}, |
| 185 | + {"type": "text", "text": "Second part. "}, |
| 186 | + {"type": "text", "text": "Third part."}, |
| 187 | + ], |
| 188 | + ) |
| 189 | + result = langchain_messages_to_copilotkit([msg]) |
| 190 | + assert len(result) == 1 |
| 191 | + assert result[0]["content"] == "First part. Second part. Third part." |
0 commit comments