refactor(tests): simplify YAML fixture schema and test runner

YAML: rimosse op/description/score_name/assertions block — ora detect/process come chiave diretta, assertions piatte sullo stesso livello del caso. Runner: eliminato _run_assertions engine, assertions inline in test_preprocess. Riduzione da ~170 a ~75 righe totali tra YAML + test. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-04-07 11:30:38 +02:00
parent 3cc32569d9
commit dcd14220ca
2 changed files with 106 additions and 234 deletions
--- a/tests/fixtures/preprocessors/cases.yaml
+++ b/tests/fixtures/preprocessors/cases.yaml
@@ -1,121 +1,71 @@
-# Preprocessor test cases — Step 1 (Local Agent V2)
+# Preprocessor test cases
 #
-# Schema per caso:
-#   id: "1.N"
-#   description: str
-#   score_name: str                     # nome score inviato a Langfuse
+# detect: <expected_type>   → chiama detect_content_type(filename, content)
+# process: <content_type>   → chiama preprocess(content_type, content)
 #
-# Sorgente contenuto (una delle due):
-#   file: <nome file in data/>          # letto come testo UTF-8
-#   generate: binary_noise              # contenuto generato dal runner (per test binari)
+# Sorgente: file: <nome in data/>  oppure  generate: binary_noise
+# filename: override del nome file passato a detect (default: valore di file:)
 #
-# Per op=detect:
-#   op: detect
-#   input_filename: str                 # filename passato a detect_content_type
-#   expected_content_type: str
-#
-# Per op=preprocess:
-#   op: preprocess
-#   input_content_type: str             # content_type passato a preprocess()
-#   assertions:
-#     no_html_tags: bool
-#     min_length: int
-#     compression_ratio_lt: float       # len(clean) / len(raw) < soglia
-#     metadata_keys: [str, ...]         # chiavi che devono essere in metadata
-#     contains: str | [str, ...]        # substring(s) presenti in clean_text
-#     not_contains: str | [str, ...]    # substring(s) assenti da clean_text
-#     content_type: str                 # valore atteso di result.content_type
+# Assertions piatte (solo per process):
+#   no_html: true           clean_text senza tag HTML
+#   min_chars: N            len(clean_text) >= N
+#   ratio_lt: F             len(clean) / len(raw) < F
+#   has_meta: [k, ...]      chiavi presenti in metadata
+#   contains: str | [str]   substring(s) presenti in clean_text
+#   excludes: str | [str]   substring(s) assenti da clean_text
+#   content_type: str       result.content_type == questo valore

-cases:
+- id: "1.1"
+  file: email_action.html
+  filename: email_export.html
+  detect: email_html

-  # ── Detection tests ────────────────────────────────────────────────
+- id: "1.2"
+  file: generic_page.html
+  filename: index.html
+  detect: generic_html

-  - id: "1.1"
-    description: "Detect email HTML"
-    score_name: preprocess.detect_email
-    file: email_action.html
-    op: detect
-    input_filename: email_export.html
-    expected_content_type: email_html
+- id: "1.3"
+  file: notes.txt
+  detect: plain_text

-  - id: "1.2"
-    description: "Detect generic HTML"
-    score_name: preprocess.detect_generic
-    file: generic_page.html
-    op: detect
-    input_filename: index.html
-    expected_content_type: generic_html
+- id: "1.4"
+  generate: binary_noise
+  filename: archive.xyz
+  detect: unknown

-  - id: "1.3"
-    description: "Detect plain text"
-    score_name: preprocess.detect_text
-    file: notes.txt
-    op: detect
-    input_filename: notes.txt
-    expected_content_type: plain_text
+- id: "1.5"
+  file: email_action.html
+  process: email_html
+  no_html: true
+  min_chars: 50
+  ratio_lt: 0.8

-  - id: "1.4"
-    description: "Detect unknown (binary-like content)"
-    score_name: preprocess.detect_unknown
-    generate: binary_noise
-    op: detect
-    input_filename: archive.xyz
-    expected_content_type: unknown
+- id: "1.6"
+  file: email_action.html
+  process: email_html
+  has_meta: [subject, from]

-  # ── Preprocess tests ───────────────────────────────────────────────
+- id: "1.7"
+  file: email_thread.html
+  process: email_html
+  contains: "Sure, I'll handle the deploy"
+  excludes: "Let's plan the deploy"

-  - id: "1.5"
-    description: "Email: strip HTML tags"
-    file: email_action.html
-    op: preprocess
-    input_content_type: email_html
-    assertions:
-      no_html_tags: true
-      min_length: 50
-      compression_ratio_lt: 0.8
+- id: "1.8"
+  file: email_single.html
+  process: email_html
+  contains: "deploy is done"

-  - id: "1.6"
-    description: "Email: extract metadata (Subject + From)"
-    file: email_action.html
-    op: preprocess
-    input_content_type: email_html
-    assertions:
-      metadata_keys: [subject, from]
+- id: "1.9"
+  file: email_heavy.html
+  process: email_html
+  no_html: true
+  min_chars: 30
+  excludes: [border-collapse, font-size]

-  - id: "1.7"
-    description: "Email: split thread — solo ultimo messaggio"
-    file: email_thread.html
-    op: preprocess
-    input_content_type: email_html
-    assertions:
-      contains: "Sure, I'll handle the deploy"
-      not_contains: "Let's plan the deploy"
-
-  - id: "1.8"
-    description: "Email: singolo messaggio senza thread"
-    file: email_single.html
-    op: preprocess
-    input_content_type: email_html
-    assertions:
-      contains: "deploy is done"
-
-  - id: "1.9"
-    description: "Email: HTML pesante con table layout"
-    file: email_heavy.html
-    op: preprocess
-    input_content_type: email_html
-    assertions:
-      no_html_tags: true
-      min_length: 30
-      not_contains:
-        - "border-collapse"
-        - "font-size"
-
-  - id: "1.10"
-    description: "Fallback: file sconosciuto → testo restituito"
-    file: fallback.txt
-    op: preprocess
-    input_content_type: unknown
-    assertions:
-      min_length: 1
-      content_type: unknown
+- id: "1.10"
+  file: fallback.txt
+  process: unknown
+  min_chars: 1
+  content_type: unknown
--- a/tests/test_preprocessors.py
+++ b/tests/test_preprocessors.py
@@ -1,26 +1,15 @@
 """Tests for the preprocessor system (Step 1 — Local Agent V2).

-Fixtures are driven by:
-  tests/fixtures/preprocessors/cases.yaml   — test case definitions
-  tests/fixtures/preprocessors/data/        — input files (HTML, txt, ...)
+Fixtures: tests/fixtures/preprocessors/cases.yaml + data/

 Run:
    pytest tests/test_preprocessors.py -v
-
-    # Only detection tests
-    pytest tests/test_preprocessors.py -v -k detect
-
-    # Only preprocess tests
-    pytest tests/test_preprocessors.py -v -k preprocess
-
-Langfuse scores are sent when LANGFUSE_SECRET_KEY / LANGFUSE_PUBLIC_KEY are set.
 """

 from __future__ import annotations

 import re
 from pathlib import Path
-from typing import Any

 import pytest
 import yaml
@@ -28,144 +17,77 @@ import yaml
 from app.core.langfuse_client import get_langfuse
 from app.core.preprocessors import detect_content_type, preprocess

-# ── Paths ──────────────────────────────────────────────────────────────
+_DATA_DIR = Path(__file__).parent / "fixtures" / "preprocessors" / "data"
+_CASES_FILE = Path(__file__).parent / "fixtures" / "preprocessors" / "cases.yaml"

-_FIXTURES_DIR = Path(__file__).parent / "fixtures" / "preprocessors"
-_DATA_DIR = _FIXTURES_DIR / "data"
-_CASES_FILE = _FIXTURES_DIR / "cases.yaml"
-
-# ── Content generators ─────────────────────────────────────────────────
-
-_GENERATORS: dict[str, str] = {
-    # High ratio of non-printable chars → triggers "unknown" heuristic
+_GENERATORS = {
    "binary_noise": "some\x00\x01\x02\x03\x04\x05content" * 20,
 }


-def _load_cases() -> list[dict]:
-    with _CASES_FILE.open(encoding="utf-8") as f:
-        return yaml.safe_load(f)["cases"]
+def _cases():
+    return yaml.safe_load(_CASES_FILE.read_text(encoding="utf-8"))


-def _read_content(case: dict) -> str:
+def _content(case: dict) -> str:
    if "generate" in case:
-        key = case["generate"]
-        if key not in _GENERATORS:
-            raise ValueError(f"Unknown generator '{key}' in case {case['id']}")
-        return _GENERATORS[key]
-    file_path = _DATA_DIR / case["file"]
-    return file_path.read_text(encoding="utf-8")
+        return _GENERATORS[case["generate"]]
+    return (_DATA_DIR / case["file"]).read_text(encoding="utf-8")


-# ── Langfuse helper ───────────────────────────────────────────────────
-
-def _lf_score(score_name: str, value: float, comment: str = "") -> None:
+def _lf_score(name: str, value: float, comment: str = "") -> None:
    lf = get_langfuse()
    if lf:
-        trace = lf.trace(name=f"eval-{score_name}")
-        lf.score(
-            trace_id=trace.id,
-            name=score_name,
-            value=value,
-            data_type="NUMERIC",
-            comment=comment,
-        )
+        trace = lf.trace(name=f"eval-{name}")
+        lf.score(trace_id=trace.id, name=name, value=value, data_type="NUMERIC", comment=comment)
        lf.flush()


-# ── Assertion engine ──────────────────────────────────────────────────
+# ── detect ────────────────────────────────────────────────────────────

-def _run_assertions(assertions: dict[str, Any], result: Any, raw: str) -> list[str]:
-    """Run all assertions declared in the YAML case. Returns failure messages."""
-    failures: list[str] = []
-
-    if assertions.get("no_html_tags"):
-        if re.search(r"<[^>]+>", result.clean_text):
-            failures.append("clean_text still contains HTML tags")
-
-    min_len = assertions.get("min_length")
-    if min_len is not None:
-        if len(result.clean_text) < min_len:
-            failures.append(
-                f"clean_text too short: {len(result.clean_text)} < {min_len}"
-            )
-
-    ratio_lt = assertions.get("compression_ratio_lt")
-    if ratio_lt is not None and len(raw) > 0:
-        ratio = len(result.clean_text) / len(raw)
-        if ratio >= ratio_lt:
-            failures.append(f"compression ratio {ratio:.2f} >= {ratio_lt}")
-
-    meta_keys = assertions.get("metadata_keys", [])
-    for key in meta_keys:
-        if not result.metadata.get(key):
-            failures.append(f"metadata missing key '{key}' (got {result.metadata})")
-
-    contains = assertions.get("contains")
-    if contains:
-        items = [contains] if isinstance(contains, str) else contains
-        for item in items:
-            if item not in result.clean_text:
-                failures.append(f"clean_text missing expected substring: {item!r}")
-
-    not_contains = assertions.get("not_contains")
-    if not_contains:
-        items = [not_contains] if isinstance(not_contains, str) else not_contains
-        for item in items:
-            if item in result.clean_text:
-                failures.append(f"clean_text contains forbidden substring: {item!r}")
-
-    expected_ct = assertions.get("content_type")
-    if expected_ct and result.content_type != expected_ct:
-        failures.append(
-            f"content_type mismatch: expected {expected_ct!r}, got {result.content_type!r}"
-        )
-
-    return failures
+_detect = [c for c in _cases() if "detect" in c]


-# ── Parametrized: detect ──────────────────────────────────────────────
-
-_detect_cases = [c for c in _load_cases() if c["op"] == "detect"]
-
-
-@pytest.mark.parametrize(
-    "case",
-    _detect_cases,
-    ids=[c["id"] for c in _detect_cases],
-)
+@pytest.mark.parametrize("case", _detect, ids=[c["id"] for c in _detect])
 def test_detect(case: dict) -> None:
-    raw = _read_content(case)
-    ct = detect_content_type(case["input_filename"], raw)
-
-    expected = case["expected_content_type"]
-    score = 1.0 if ct == expected else 0.0
-    _lf_score(case["score_name"], score, f"got={ct}, expected={expected}")
-
-    assert ct == expected, (
-        f"[{case['id']}] {case['description']}: "
-        f"expected content_type={expected!r}, got {ct!r}"
-    )
+    raw = _content(case)
+    filename = case.get("filename", case.get("file", ""))
+    ct = detect_content_type(filename, raw)
+    expected = case["detect"]
+    _lf_score(f"preprocess.detect.{case['id']}", 1.0 if ct == expected else 0.0)
+    assert ct == expected, f"[{case['id']}] expected {expected!r}, got {ct!r}"


-# ── Parametrized: preprocess ──────────────────────────────────────────
+# ── preprocess ────────────────────────────────────────────────────────

-_preprocess_cases = [c for c in _load_cases() if c["op"] == "preprocess"]
+_process = [c for c in _cases() if "process" in c]


-@pytest.mark.parametrize(
-    "case",
-    _preprocess_cases,
-    ids=[c["id"] for c in _preprocess_cases],
-)
+@pytest.mark.parametrize("case", _process, ids=[c["id"] for c in _process])
 def test_preprocess(case: dict) -> None:
-    raw = _read_content(case)
-    result = preprocess(case["input_content_type"], raw)
+    raw = _content(case)
+    result = preprocess(case["process"], raw)

-    assertions = case.get("assertions", {})
-    failures = _run_assertions(assertions, result, raw)
+    if case.get("no_html"):
+        assert not re.search(r"<[^>]+>", result.clean_text), "clean_text contains HTML tags"

-    assert not failures, (
-        f"[{case['id']}] {case['description']} — {len(failures)} assertion(s) failed:\n"
-        + "\n".join(f"  • {f}" for f in failures)
-    )
+    if "min_chars" in case:
+        assert len(result.clean_text) >= case["min_chars"], \
+            f"clean_text too short: {len(result.clean_text)} < {case['min_chars']}"
+
+    if "ratio_lt" in case:
+        ratio = len(result.clean_text) / len(raw)
+        assert ratio < case["ratio_lt"], f"compression ratio {ratio:.2f} >= {case['ratio_lt']}"
+
+    for key in case.get("has_meta", []):
+        assert result.metadata.get(key), f"metadata missing {key!r} (got {result.metadata})"
+
+    for item in ([case["contains"]] if isinstance(case.get("contains"), str) else case.get("contains", [])):
+        assert item in result.clean_text, f"clean_text missing {item!r}"
+
+    for item in ([case["excludes"]] if isinstance(case.get("excludes"), str) else case.get("excludes", [])):
+        assert item not in result.clean_text, f"clean_text contains forbidden {item!r}"
+
+    if "content_type" in case:
+        assert result.content_type == case["content_type"], \
+            f"expected content_type {case['content_type']!r}, got {result.content_type!r}"