fix(template-transform): use base64 encoding for Jinja2 templates to fix #26818 (#30223)

Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com>
Co-authored-by: crazywoola <100913391+crazywoola@users.noreply.github.com>

api/core/helper/code_executor/jinja2/jinja2_transformer.py

@@ -1,9 +1,14 @@
+from collections.abc import Mapping
 from textwrap import dedent
+from typing import Any
 
 from core.helper.code_executor.template_transformer import TemplateTransformer
 
 
 class Jinja2TemplateTransformer(TemplateTransformer):
+    # Use separate placeholder for base64-encoded template to avoid confusion
+    _template_b64_placeholder: str = "{{template_b64}}"
+
     @classmethod
     def transform_response(cls, response: str):
         """
@@ -13,18 +18,35 @@ class Jinja2TemplateTransformer(TemplateTransformer):
         """
         return {"result": cls.extract_result_str_from_response(response)}
 
+    @classmethod
+    def assemble_runner_script(cls, code: str, inputs: Mapping[str, Any]) -> str:
+        """
+        Override base class to use base64 encoding for template code.
+        This prevents issues with special characters (quotes, newlines) in templates
+        breaking the generated Python script. Fixes #26818.
+        """
+        script = cls.get_runner_script()
+        # Encode template as base64 to safely embed any content including quotes
+        code_b64 = cls.serialize_code(code)
+        script = script.replace(cls._template_b64_placeholder, code_b64)
+        inputs_str = cls.serialize_inputs(inputs)
+        script = script.replace(cls._inputs_placeholder, inputs_str)
+        return script
+
     @classmethod
     def get_runner_script(cls) -> str:
         runner_script = dedent(f"""
-            # declare main function
-            def main(**inputs):
-                import jinja2
-                template = jinja2.Template('''{cls._code_placeholder}''')
-                return template.render(**inputs)
-
+            import jinja2
             import json
             from base64 import b64decode
 
+            # declare main function
+            def main(**inputs):
+                # Decode base64-encoded template to handle special characters safely
+                template_code = b64decode('{cls._template_b64_placeholder}').decode('utf-8')
+                template = jinja2.Template(template_code)
+                return template.render(**inputs)
+
             # decode and prepare input dict
             inputs_obj = json.loads(b64decode('{cls._inputs_placeholder}').decode('utf-8'))
 

api/core/helper/code_executor/template_transformer.py

@@ -13,6 +13,15 @@ class TemplateTransformer(ABC):
     _inputs_placeholder: str = "{{inputs}}"
     _result_tag: str = "<<RESULT>>"
 
+    @classmethod
+    def serialize_code(cls, code: str) -> str:
+        """
+        Serialize template code to base64 to safely embed in generated script.
+        This prevents issues with special characters like quotes breaking the script.
+        """
+        code_bytes = code.encode("utf-8")
+        return b64encode(code_bytes).decode("utf-8")
+
     @classmethod
     def transform_caller(cls, code: str, inputs: Mapping[str, Any]) -> tuple[str, str]:
         """

api/tests/integration_tests/workflow/nodes/code_executor/test_code_jinja2.py

@@ -7,11 +7,14 @@ CODE_LANGUAGE = CodeLanguage.JINJA2
 
 
 def test_jinja2():
+    """Test basic Jinja2 template rendering."""
     template = "Hello {{template}}"
+    # Template must be base64 encoded to match the new safe embedding approach
+    template_b64 = base64.b64encode(template.encode("utf-8")).decode("utf-8")
     inputs = base64.b64encode(b'{"template": "World"}').decode("utf-8")
     code = (
         Jinja2TemplateTransformer.get_runner_script()
-        .replace(Jinja2TemplateTransformer._code_placeholder, template)
+        .replace(Jinja2TemplateTransformer._template_b64_placeholder, template_b64)
         .replace(Jinja2TemplateTransformer._inputs_placeholder, inputs)
     )
     result = CodeExecutor.execute_code(
@@ -21,6 +24,7 @@ def test_jinja2():
 
 
 def test_jinja2_with_code_template():
+    """Test template rendering via the high-level workflow API."""
     result = CodeExecutor.execute_workflow_code_template(
         language=CODE_LANGUAGE, code="Hello {{template}}", inputs={"template": "World"}
     )
@@ -28,7 +32,64 @@ def test_jinja2_with_code_template():
 
 
 def test_jinja2_get_runner_script():
+    """Test that runner script contains required placeholders."""
     runner_script = Jinja2TemplateTransformer.get_runner_script()
-    assert runner_script.count(Jinja2TemplateTransformer._code_placeholder) == 1
+    assert runner_script.count(Jinja2TemplateTransformer._template_b64_placeholder) == 1
     assert runner_script.count(Jinja2TemplateTransformer._inputs_placeholder) == 1
     assert runner_script.count(Jinja2TemplateTransformer._result_tag) == 2
+
+
+def test_jinja2_template_with_special_characters():
+    """
+    Test that templates with special characters (quotes, newlines) render correctly.
+    This is a regression test for issue #26818 where textarea pre-fill values
+    containing special characters would break template rendering.
+    """
+    # Template with triple quotes, single quotes, double quotes, and newlines
+    template = """<html>
+<body>
+    <input value="{{ task.get('Task ID', '') }}"/>
+    <textarea>{{ task.get('Issues', 'No issues reported') }}</textarea>
+    <p>Status: "{{ status }}"</p>
+    <pre>'''code block'''</pre>
+</body>
+</html>"""
+    inputs = {"task": {"Task ID": "TASK-123", "Issues": "Line 1\nLine 2\nLine 3"}, "status": "completed"}
+
+    result = CodeExecutor.execute_workflow_code_template(language=CODE_LANGUAGE, code=template, inputs=inputs)
+
+    # Verify the template rendered correctly with all special characters
+    output = result["result"]
+    assert 'value="TASK-123"' in output
+    assert "<textarea>Line 1\nLine 2\nLine 3</textarea>" in output
+    assert 'Status: "completed"' in output
+    assert "'''code block'''" in output
+
+
+def test_jinja2_template_with_html_textarea_prefill():
+    """
+    Specific test for HTML textarea with Jinja2 variable pre-fill.
+    Verifies fix for issue #26818.
+    """
+    template = "<textarea name='notes'>{{ notes }}</textarea>"
+    notes_content = "This is a multi-line note.\nWith special chars: 'single' and \"double\" quotes."
+    inputs = {"notes": notes_content}
+
+    result = CodeExecutor.execute_workflow_code_template(language=CODE_LANGUAGE, code=template, inputs=inputs)
+
+    expected_output = f"<textarea name='notes'>{notes_content}</textarea>"
+    assert result["result"] == expected_output
+
+
+def test_jinja2_assemble_runner_script_encodes_template():
+    """Test that assemble_runner_script properly base64 encodes the template."""
+    template = "Hello {{ name }}!"
+    inputs = {"name": "World"}
+
+    script = Jinja2TemplateTransformer.assemble_runner_script(template, inputs)
+
+    # The template should be base64 encoded in the script
+    template_b64 = base64.b64encode(template.encode("utf-8")).decode("utf-8")
+    assert template_b64 in script
+    # The raw template should NOT appear in the script (it's encoded)
+    assert "Hello {{ name }}!" not in script

api/tests/test_containers_integration_tests/workflow/nodes/code_executor/test_code_jinja2.py

@@ -12,10 +12,12 @@ class TestJinja2CodeExecutor(CodeExecutorTestMixin):
         _, Jinja2TemplateTransformer = self.jinja2_imports
 
         template = "Hello {{template}}"
+        # Template must be base64 encoded to match the new safe embedding approach
+        template_b64 = base64.b64encode(template.encode("utf-8")).decode("utf-8")
         inputs = base64.b64encode(b'{"template": "World"}').decode("utf-8")
         code = (
             Jinja2TemplateTransformer.get_runner_script()
-            .replace(Jinja2TemplateTransformer._code_placeholder, template)
+            .replace(Jinja2TemplateTransformer._template_b64_placeholder, template_b64)
             .replace(Jinja2TemplateTransformer._inputs_placeholder, inputs)
         )
         result = CodeExecutor.execute_code(
@@ -37,6 +39,34 @@ class TestJinja2CodeExecutor(CodeExecutorTestMixin):
         _, Jinja2TemplateTransformer = self.jinja2_imports
 
         runner_script = Jinja2TemplateTransformer.get_runner_script()
-        assert runner_script.count(Jinja2TemplateTransformer._code_placeholder) == 1
+        assert runner_script.count(Jinja2TemplateTransformer._template_b64_placeholder) == 1
         assert runner_script.count(Jinja2TemplateTransformer._inputs_placeholder) == 1
         assert runner_script.count(Jinja2TemplateTransformer._result_tag) == 2
+
+    def test_jinja2_template_with_special_characters(self, flask_app_with_containers):
+        """
+        Test that templates with special characters (quotes, newlines) render correctly.
+        This is a regression test for issue #26818 where textarea pre-fill values
+        containing special characters would break template rendering.
+        """
+        CodeExecutor, CodeLanguage = self.code_executor_imports
+
+        # Template with triple quotes, single quotes, double quotes, and newlines
+        template = """<html>
+<body>
+    <input value="{{ task.get('Task ID', '') }}"/>
+    <textarea>{{ task.get('Issues', 'No issues reported') }}</textarea>
+    <p>Status: "{{ status }}"</p>
+    <pre>'''code block'''</pre>
+</body>
+</html>"""
+        inputs = {"task": {"Task ID": "TASK-123", "Issues": "Line 1\nLine 2\nLine 3"}, "status": "completed"}
+
+        result = CodeExecutor.execute_workflow_code_template(language=CodeLanguage.JINJA2, code=template, inputs=inputs)
+
+        # Verify the template rendered correctly with all special characters
+        output = result["result"]
+        assert 'value="TASK-123"' in output
+        assert "<textarea>Line 1\nLine 2\nLine 3</textarea>" in output
+        assert 'Status: "completed"' in output
+        assert "'''code block'''" in output