Add comprehensive test scenarios for code interpreter

Co-authored-by: owndev <69784886+owndev@users.noreply.github.com>

Author: Copilot

TEST_SCENARIOS.md

@@ -0,0 +1,182 @@
+# Test Scenarios for Code Interpreter Fix
+
+## Prerequisites
+1. Open WebUI is running with the updated Gemini pipeline
+2. Code interpreter feature is enabled in Open WebUI settings
+3. Google API key or Vertex AI credentials are configured
+4. A Gemini model is selected (e.g., gemini-2.5-pro, gemini-2.0-flash)
+
+## Test Scenario 1: Simple Code Execution
+
+**Prompt:**
+```
+Write and execute Python code to calculate pi to 10 decimal places using the Leibniz formula.
+```
+
+**Expected Behavior:**
+- Gemini should respond with code explanation
+- Code should execute automatically
+- Results should be displayed showing pi ≈ 3.1415926536
+
+**What to Check:**
+- No repeated text (the original bug symptom)
+- Code is displayed in a code block
+- Execution results are shown
+- No errors in console logs
+
+## Test Scenario 2: Data Visualization
+
+**Prompt:**
+```
+Create a simple bar chart showing the first 5 Fibonacci numbers using matplotlib.
+```
+
+**Expected Behavior:**
+- Code generates a bar chart
+- Chart is displayed in the response
+- No repeated text errors
+
+**What to Check:**
+- Code executes successfully
+- Image/chart is visible
+- Proper error handling if matplotlib isn't available
+
+## Test Scenario 3: Error Handling
+
+**Prompt:**
+```
+Execute this Python code: print(1/0)
+```
+
+**Expected Behavior:**
+- Code attempts to execute
+- Division by zero error is caught and displayed
+- Error message is clear and doesn't break the UI
+
+**What to Check:**
+- Error is handled gracefully
+- No system crash or hung requests
+- Error message is visible to user
+
+## Test Scenario 4: Multi-turn Conversation
+
+**Prompt 1:**
+```
+Create a Python function to calculate factorial of a number.
+```
+
+**Prompt 2:**
+```
+Now use that function to calculate factorial of 10.
+```
+
+**Expected Behavior:**
+- First response creates and shows the function
+- Second response uses the function and shows result (3628800)
+- Context is maintained between turns
+
+**What to Check:**
+- Multi-turn context works correctly
+- Variables/functions from previous turns are available
+- No repeated text issues
+
+## Test Scenario 5: Complex Calculation
+
+**Prompt:**
+```
+Calculate the first 20 prime numbers using Python.
+```
+
+**Expected Behavior:**
+- Code is generated and executed
+- List of primes is displayed: [2, 3, 5, 7, 11, 13, 17, 19, 23, 29, 31, 37, 41, 43, 47, 53, 59, 61, 67, 71]
+
+**What to Check:**
+- Complex logic executes correctly
+- Results are accurate
+- No timeout errors
+
+## Test Scenario 6: Streaming Mode
+
+Enable streaming responses and test:
+
+**Prompt:**
+```
+Generate a simple "Hello, World!" program in Python and execute it.
+```
+
+**Expected Behavior:**
+- Response streams in real-time
+- Code execution still works
+- No "chunk too big" errors
+
+**What to Check:**
+- Streaming works smoothly
+- Tool calls are detected in streaming mode
+- No response corruption
+
+## Test Scenario 7: Non-Streaming Mode
+
+Disable streaming responses and test:
+
+**Prompt:**
+```
+Calculate the sum of numbers from 1 to 100 using Python.
+```
+
+**Expected Behavior:**
+- Response arrives all at once
+- Code executes and shows result (5050)
+- No repeated text
+
+**What to Check:**
+- Non-streaming mode works
+- Tool calls are detected and emitted
+- Format is correct
+
+## Debugging Tips
+
+If tests fail, check:
+
+1. **Browser Console**: Look for JavaScript errors or failed API calls
+2. **Open WebUI Logs**: Check for Python exceptions or warnings
+3. **Network Tab**: Inspect the API request/response format
+4. **Event Emitter**: Verify events are being emitted correctly
+
+Key indicators of success:
+- ✅ No repeated text in responses
+- ✅ Code blocks are properly formatted
+- ✅ Execution results are displayed
+- ✅ Tool call events appear in logs
+- ✅ Format matches OpenAI tool call structure
+
+Key indicators of issues:
+- ❌ Text repeats multiple times
+- ❌ Code doesn't execute
+- ❌ "function_call" errors in logs
+- ❌ Missing tool_calls in API response
+- ❌ Malformed JSON in arguments
+
+## Log Monitoring
+
+Watch for these log messages (set log level to DEBUG):
+
+**Success indicators:**
+```
+Detected tool call: <function_name> with args: <args>
+Emitted tool call: <function_name> with args: <args>
+```
+
+**Error indicators:**
+```
+Error processing content part: ...
+Failed to access content parts: ...
+```
+
+## Comparison with Azure Pipeline
+
+To verify the fix matches Azure's behavior, test the same prompts with both:
+1. Azure AI pipeline (known working)
+2. Gemini pipeline (after fix)
+
+Both should execute code successfully and show results.