176 行
6.1 KiB
Python
176 行
6.1 KiB
Python
from __future__ import annotations
|
|
|
|
import pytest
|
|
|
|
from backend.services.workflow_feedback_interpreter import (
|
|
WorkflowFeedbackInterpretationError,
|
|
WorkflowFeedbackInterpreter,
|
|
interpret_workflow_feedback,
|
|
)
|
|
|
|
|
|
CANDIDATES = {
|
|
"TX-15-01": "candidate-1",
|
|
"TX-15-02": "candidate-2",
|
|
"TX-15-03": "candidate-3",
|
|
}
|
|
|
|
|
|
def _ids(items: object) -> list[str]:
|
|
return [item.candidate_id for item in items] # type: ignore[union-attr]
|
|
|
|
|
|
def test_interprets_chinese_likes_and_strict_rejection_by_short_number() -> None:
|
|
result = interpret_workflow_feedback(
|
|
"1、2喜欢,3严格拒绝",
|
|
candidate_whitelist=CANDIDATES,
|
|
)
|
|
|
|
assert _ids(result.liked) == ["candidate-1", "candidate-2"]
|
|
assert result.rejected == ()
|
|
assert _ids(result.strict_rejected) == ["candidate-3"]
|
|
assert result.final is None
|
|
|
|
|
|
def test_interprets_full_display_code_as_final_candidate() -> None:
|
|
result = interpret_workflow_feedback(
|
|
"TX-15-03 采用",
|
|
candidate_whitelist=CANDIDATES,
|
|
)
|
|
|
|
assert result.final is not None
|
|
assert result.final.candidate_id == "candidate-3"
|
|
assert result.final.display_code == "TX-15-03"
|
|
|
|
|
|
def test_interprets_prefix_keep_and_postfix_reject_without_cross_contamination() -> None:
|
|
result = interpret_workflow_feedback(
|
|
"保留01和03,02不要",
|
|
candidate_whitelist=CANDIDATES,
|
|
)
|
|
|
|
assert _ids(result.liked) == ["candidate-1", "candidate-3"]
|
|
assert _ids(result.rejected) == ["candidate-2"]
|
|
assert result.strict_rejected == ()
|
|
|
|
|
|
def test_interprets_english_feedback_and_round_directives() -> None:
|
|
result = interpret_workflow_feedback(
|
|
"like 1 and 2; hard reject 3; next round: cooler lighting; material: black coat",
|
|
candidate_whitelist=CANDIDATES,
|
|
)
|
|
|
|
assert _ids(result.liked) == ["candidate-1", "candidate-2"]
|
|
assert _ids(result.strict_rejected) == ["candidate-3"]
|
|
assert result.next_round_prompt_patch == "cooler lighting"
|
|
assert result.material_intent == "black coat"
|
|
|
|
|
|
def test_interprets_chinese_ordinal_references_and_directives() -> None:
|
|
result = interpret_workflow_feedback(
|
|
"第一张喜欢,第三张采用,下一轮加强轮廓光,素材用黑色外套",
|
|
candidate_whitelist=CANDIDATES,
|
|
)
|
|
|
|
assert _ids(result.liked) == ["candidate-1"]
|
|
assert result.final is not None and result.final.candidate_id == "candidate-3"
|
|
assert result.next_round_prompt_patch == "加强轮廓光"
|
|
assert result.material_intent == "黑色外套"
|
|
|
|
|
|
def test_rejects_hallucinated_or_out_of_range_candidate() -> None:
|
|
with pytest.raises(WorkflowFeedbackInterpretationError) as exc_info:
|
|
interpret_workflow_feedback(
|
|
"04喜欢",
|
|
candidate_whitelist=CANDIDATES,
|
|
)
|
|
|
|
assert exc_info.value.reason == "unknown_candidate_reference"
|
|
assert exc_info.value.details["reference"] == "04"
|
|
assert exc_info.value.details["allowed_display_codes"] == list(CANDIDATES)
|
|
|
|
|
|
def test_short_number_must_be_unique_but_full_display_code_remains_safe() -> None:
|
|
ambiguous_candidates = {
|
|
"TX-15-01": "candidate-round-15",
|
|
"TX-16-01": "candidate-round-16",
|
|
}
|
|
|
|
exact = interpret_workflow_feedback(
|
|
"tx-16-01 adopt",
|
|
candidate_whitelist=ambiguous_candidates,
|
|
)
|
|
assert exact.final is not None and exact.final.candidate_id == "candidate-round-16"
|
|
|
|
with pytest.raises(WorkflowFeedbackInterpretationError) as exc_info:
|
|
interpret_workflow_feedback(
|
|
"01 adopt",
|
|
candidate_whitelist=ambiguous_candidates,
|
|
)
|
|
assert exc_info.value.reason == "ambiguous_candidate_reference"
|
|
|
|
|
|
def test_llm_result_is_validated_through_display_code_whitelist_without_network() -> None:
|
|
interpreter = WorkflowFeedbackInterpreter()
|
|
result = interpreter.validate_llm_result(
|
|
{
|
|
"liked": ["01", {"display_code": "TX-15-02"}],
|
|
"final": {"display_code": "TX-15-03"},
|
|
"material_intent": "keep the black coat reference",
|
|
"next_round_prompt_patch": "use a calmer expression",
|
|
},
|
|
candidate_whitelist=CANDIDATES,
|
|
source_text="LLM result fixture",
|
|
)
|
|
|
|
assert _ids(result.liked) == ["candidate-1", "candidate-2"]
|
|
assert result.final is not None and result.final.candidate_id == "candidate-3"
|
|
assert result.source == "llm_validated"
|
|
assert result.to_dict()["final"] == {
|
|
"candidate_id": "candidate-3",
|
|
"display_code": "TX-15-03",
|
|
}
|
|
|
|
|
|
def test_llm_result_rejects_raw_candidate_id_and_unknown_display_code() -> None:
|
|
interpreter = WorkflowFeedbackInterpreter()
|
|
|
|
with pytest.raises(WorkflowFeedbackInterpretationError) as raw_id_error:
|
|
interpreter.validate_llm_result(
|
|
{"liked": [{"candidate_id": "candidate-1"}]},
|
|
candidate_whitelist=CANDIDATES,
|
|
)
|
|
assert raw_id_error.value.reason == "invalid_llm_result"
|
|
|
|
with pytest.raises(WorkflowFeedbackInterpretationError) as unknown_code_error:
|
|
interpreter.validate_llm_result(
|
|
{"final": "TX-99-01"},
|
|
candidate_whitelist=CANDIDATES,
|
|
)
|
|
assert unknown_code_error.value.reason == "unknown_candidate_reference"
|
|
|
|
|
|
def test_rejects_conflicting_feedback_and_multiple_final_candidates() -> None:
|
|
with pytest.raises(WorkflowFeedbackInterpretationError) as conflict_error:
|
|
interpret_workflow_feedback(
|
|
"01喜欢,01不要",
|
|
candidate_whitelist=CANDIDATES,
|
|
)
|
|
assert conflict_error.value.reason == "conflicting_candidate_feedback"
|
|
|
|
with pytest.raises(WorkflowFeedbackInterpretationError) as final_error:
|
|
WorkflowFeedbackInterpreter().validate_llm_result(
|
|
{"final": ["01", "02"]},
|
|
candidate_whitelist=CANDIDATES,
|
|
)
|
|
assert final_error.value.reason == "multiple_final_candidates"
|
|
|
|
|
|
def test_prompt_numbers_are_not_mistaken_for_candidate_feedback() -> None:
|
|
result = interpret_workflow_feedback(
|
|
"下一轮使用2K分辨率,不要红色背景",
|
|
candidate_whitelist=CANDIDATES,
|
|
)
|
|
|
|
assert result.has_candidate_feedback is False
|
|
assert result.next_round_prompt_patch == "使用2K分辨率"
|