fix(dspy): Guard use_tip on tip value in GroundedProposer to avoid rendering None in prompt

dspy/propose/grounded_proposer.py

@@ -387,7 +387,7 @@ def propose_instruction_for_predictor(
             program_aware=self.program_aware,
             use_task_demos=self.use_task_demos and demo_candidates,
             use_instruct_history=self.use_instruct_history and instruction_history,
-            use_tip=self.use_tip,
+            use_tip=self.use_tip and bool(tip),
             verbose=self.verbose
         )
 

tests/propose/test_grounded_proposer.py

@@ -6,6 +6,49 @@
 from dspy.utils.dummies import DummyLM
 
 
+class RecordingDummyLM(DummyLM):
+    """DummyLM that records the last user-message content sent to it (or any copy of it).
+
+    ``GroundedProposer.propose_instruction_for_predictor`` sends the generation prompt to a
+    *copy* of the prompt model (``self.prompt_model.copy(...)``). ``BaseLM.copy`` performs a
+    shallow copy, so ``captured_prompts`` is shared by reference between the original and the
+    rollout copy, allowing the test to inspect the exact prompt sent to the LM.
+    """
+
+    def __init__(self, answers, **kwargs):
+        super().__init__(answers, **kwargs)
+        self.captured_prompts = []
+
+    def forward(self, prompt=None, messages=None, **kwargs):
+        messages = messages or [{"role": "user", "content": prompt}]
+        self.captured_prompts.append(messages[-1]["content"])
+        return super().forward(prompt=prompt, messages=messages, **kwargs)
+
+
+def _make_proposer(*, use_tip=True, set_tip_randomly=False):
+    """Build a minimal ``GroundedProposer`` whose only variable input field is ``tip``.
+
+    Disables dataset summary, program awareness, task demos, and instruction history so that
+    the only conditionally-included input field is ``tip``, making prompt assertions precise.
+    """
+    program = Predict("question -> answer")
+    lm = RecordingDummyLM([{"proposed_instruction": "instruction"}] * 20)
+    proposer = GroundedProposer(
+        prompt_model=lm,
+        program=program,
+        trainset=[],
+        verbose=False,
+        use_dataset_summary=False,
+        program_aware=False,
+        use_task_demos=False,
+        use_instruct_history=False,
+        use_tip=use_tip,
+        set_tip_randomly=set_tip_randomly,
+        set_history_randomly=False,
+    )
+    return proposer, lm, program
+
+
 @pytest.mark.parametrize(
     "demo_candidates",
     [
@@ -63,3 +106,62 @@ def copy(self, **kwargs):
     )
     assert result == "instruction"
     assert prompt_model.last_copy_kwargs["temperature"] == 0.7
+
+
+def test_propose_instruction_for_predictor_omits_tip_field_when_tip_is_none():
+    """``use_tip=True`` with ``tip=None`` must not render ``[[ ## tip ## ]]\nNone`` in the prompt.
+
+    Regression test for the ``set_tip_randomly=False`` path (supported per PR #3919), where
+    ``selected_tip`` stays ``None`` while ``use_tip`` keeps its constructor default ``True``.
+    Before the fix the ``tip`` input field was still declared on the signature and ``None`` was
+    forwarded into the prompt, where the adapter rendered it as the literal string ``"None"``.
+    """
+    proposer, lm, program = _make_proposer(use_tip=True, set_tip_randomly=False)
+    result = proposer.propose_instruction_for_predictor(
+        program=program,
+        predictor=None,
+        pred_i=0,
+        demo_candidates=None,
+        demo_set_i=0,
+        trial_logs={},
+        tip=None,
+    )
+    assert result == "instruction"
+    assert "[[ ## tip ## ]]" not in lm.captured_prompts[-1]
+
+
+def test_propose_instruction_for_predictor_includes_tip_field_when_tip_supplied():
+    """A real tip value must still declare the ``tip`` field and render the tip text.
+
+    Guards against an over-correction that would always omit the ``tip`` field.
+    """
+    proposer, lm, program = _make_proposer(use_tip=True, set_tip_randomly=False)
+    tip = "Don't be afraid to be creative when creating the new instruction!"
+    result = proposer.propose_instruction_for_predictor(
+        program=program,
+        predictor=None,
+        pred_i=0,
+        demo_candidates=None,
+        demo_set_i=0,
+        trial_logs={},
+        tip=tip,
+    )
+    assert result == "instruction"
+    prompt = lm.captured_prompts[-1]
+    assert "[[ ## tip ## ]]" in prompt
+    assert tip in prompt
+
+
+def test_propose_instructions_for_program_omits_tip_when_set_tip_randomly_false():
+    """End-to-end: ``set_tip_randomly=False`` keeps ``selected_tip=None``; the prompt must not
+    contain the ``tip`` block (the user-facing configuration from PR #3919)."""
+    proposer, lm, program = _make_proposer(use_tip=True, set_tip_randomly=False)
+    result = proposer.propose_instructions_for_program(
+        trainset=[],
+        program=program,
+        demo_candidates=None,
+        trial_logs={},
+        N=1,
+    )
+    assert result == {0: ["instruction"]}
+    assert "[[ ## tip ## ]]" not in lm.captured_prompts[-1]