Spaces:
Running
Running
| """Tests for services.llm_text — the shared free-text cleanup for model prose.""" | |
| from __future__ import annotations | |
| from services.llm_text import clean_llm_text, sanitize_tree, to_plain_text | |
| class TestCleanLlmText: | |
| def test_strips_think_blocks_and_fences(self): | |
| raw = "<think>planning</think>\n```markdown\n**Hi** there\n```\n(Word count: 12)" | |
| assert clean_llm_text(raw) == "**Hi** there" | |
| def test_unclosed_think_prefix_dropped(self): | |
| raw = "<think>let me reason about this\n\nThe slope is 2." | |
| assert clean_llm_text(raw) == "The slope is 2." | |
| def test_literal_newlines_and_tex_delimiters(self): | |
| raw = "Step 1\\nStep 2 uses \\(x^2\\) and \\[a = b\\]" | |
| assert clean_llm_text(raw) == "Step 1\nStep 2 uses $x^2$ and $$a = b$$" | |
| def test_tex_commands_starting_with_n_t_r_survive(self): | |
| raw = r"$G = r \times d$, $\text{Profit} = S - C$, $a \neq b$, $\nabla f$, $\theta \to \rho$" | |
| assert clean_llm_text(raw) == raw | |
| def test_literal_escapes_before_prose_still_decode(self): | |
| assert clean_llm_text(r"Step 1\nThe next\n\nnext\t2") == "Step 1\nThe next\n\nnext 2" | |
| def test_double_escaped_tex_unescaped(self): | |
| assert clean_llm_text("Use $\\\\frac{1}{2}$") == "Use $\\frac{1}{2}$" | |
| def test_inner_code_block_preserved(self): | |
| raw = "Run:\n```python\nprint('\\n')\n```\nDone" | |
| assert clean_llm_text(raw) == raw | |
| def test_wrapping_quotes_and_label_removed(self): | |
| assert clean_llm_text('"Great job today!"') == "Great job today!" | |
| assert clean_llm_text("Message: Keep going") == "Keep going" | |
| def test_none_and_non_string(self): | |
| assert clean_llm_text(None) == "" | |
| assert clean_llm_text(42) == "42" | |
| class TestToPlainText: | |
| def test_markdown_removed_math_kept(self): | |
| raw = "## **Key**: the *slope* is $m = \\frac{2}{3}$ — see [link](http://x)" | |
| assert to_plain_text(raw) == "Key: the slope is $m = \\frac{2}{3}$ — see link" | |
| def test_option_label_stripped(self): | |
| assert to_plain_text("A) **500**", strip_option_label=True) == "500" | |
| assert to_plain_text("(b) x = 2", strip_option_label=True) == "x = 2" | |
| assert to_plain_text("Always", strip_option_label=True) == "Always" | |
| def test_subscripts_untouched(self): | |
| assert to_plain_text("x_1 and y_2 are *points*") == "x_1 and y_2 are points" | |
| def test_single_line(self): | |
| assert to_plain_text("- one\n- two", single_line=True) == "one two" | |
| class TestSanitizeTree: | |
| def test_plain_keys_vs_markdown_keys(self): | |
| payload = { | |
| "question": "What is **x**?", | |
| "options": ["**A**. 2", "B. *3*"], | |
| "explanation": "Because \\(x=2\\)\\n- step", | |
| "hints": ["use **PEMDAS**"], | |
| "id": "**q1**", | |
| "nested": {"title": "## T", "content": "<think>x</think>body"}, | |
| } | |
| out = sanitize_tree(payload) | |
| assert out["question"] == "What is **x**?" | |
| assert out["options"] == ["A. 2", "B. 3"] | |
| assert out["explanation"] == "Because $x=2$\n- step" | |
| assert out["hints"] == ["use **PEMDAS**"] | |
| assert out["id"] == "**q1**" | |
| assert out["nested"] == {"title": "T", "content": "body"} | |
| def test_non_string_leaves_untouched(self): | |
| assert sanitize_tree({"points": 3, "ok": True, "n": None}) == {"points": 3, "ok": True, "n": None} | |