fix(parser): coerce non-finite usage tokens and tidy review nits

Guard _safe_int against NaN/Infinity (json.loads accepts these literals;
int(nan)/int(inf) raise), broaden fuzz floats to cover them, and add an
explicit non-finite regression test. Revert entry_counts to truthiness so
empty-string types stay skipped, wrap service_tier in the message payload,
and drop the no-op [tool.hypothesis] block from pyproject.toml.

Author: clean6378-max-it

pyproject.toml

@@ -27,9 +27,6 @@ select = ["E", "F", "W", "I"]
 [tool.ruff.lint.isort]
 combine-as-imports = true
 
-[tool.hypothesis]
-max_examples = 200
-
 [tool.ruff.lint.per-file-ignores]
 # CLI bootstrap: sys.path must be set before local imports.
 "scripts/export.py" = ["E402"]

tests/test_parser_fuzz.py

@@ -53,7 +53,8 @@ def _write_jsonl(path: os.PathLike[str], lines: list[str]) -> str:
     st.none(),
     st.booleans(),
     st.integers(),
-    st.floats(allow_nan=False, allow_infinity=False),
+    # Allow NaN/Infinity: json.loads accepts these literals, so the parser must too.
+    st.floats(allow_nan=True, allow_infinity=True),
     st.text(max_size=200),
 )
 
@@ -263,3 +264,21 @@ def test_non_numeric_usage_tokens_do_not_crash(tmp_path: Path) -> None:
     assert session["metadata"]["total_input_tokens"] == 0
     assert session["metadata"]["total_output_tokens"] == 0
     assert session["metadata"]["total_ephemeral_5m_tokens"] == 0
+
+
+def test_non_finite_usage_tokens_do_not_crash(tmp_path: Path) -> None:
+    """json.loads accepts NaN/Infinity literals; int(nan)/int(inf) raise, so the
+    parser must coerce them to 0 rather than propagate ValueError/OverflowError."""
+    # Raw literals (not valid via json.dumps of finite floats) — written directly.
+    line = (
+        '{"type": "assistant", "message": {"usage": '
+        '{"input_tokens": NaN, "output_tokens": Infinity, '
+        '"cache_read_input_tokens": -Infinity, '
+        '"cache_creation": {"ephemeral_5m_input_tokens": NaN}}}}'
+    )
+    path = _write_jsonl(tmp_path / "nonfinite.jsonl", [line])
+    session = parse_session(path)
+    assert session["metadata"]["total_input_tokens"] == 0
+    assert session["metadata"]["total_output_tokens"] == 0
+    assert session["metadata"]["total_cache_read_tokens"] == 0
+    assert session["metadata"]["total_ephemeral_5m_tokens"] == 0

utils/jsonl_parser.py

@@ -2,6 +2,7 @@
 actually work with -- messages, tool calls, token counts, file activity, etc."""
 
 import json
+import math
 import os
 from datetime import datetime
 from typing import Any
@@ -41,12 +42,15 @@
 
 
 def _safe_int(val: Any) -> int:
-    """Coerce a value to int for token accounting; non-numeric input becomes 0
-    so fuzzed/malformed usage fields never raise during arithmetic."""
+    """Coerce a value to int for token accounting; non-numeric or non-finite
+    input becomes 0 so fuzzed/malformed usage fields never raise during
+    arithmetic. json.loads accepts NaN/Infinity literals, so guard against them."""
     if isinstance(val, bool):
         return 0
-    if isinstance(val, (int, float)):
-        return int(val)
+    if isinstance(val, int):
+        return val
+    if isinstance(val, float):
+        return int(val) if math.isfinite(val) else 0
     return 0
 
 
@@ -122,8 +126,9 @@ def parse_session(filepath: str) -> SessionDict:
                     metadata["first_timestamp"] = ts
                 metadata["last_timestamp"] = ts
 
-            # Count entry types (upstream may send non-str discriminants)
-            if entry_type is not None:
+            # Count entry types (upstream may send non-str/unhashable discriminants;
+            # coerce to str. Falsy types like "" are skipped, matching prior behavior).
+            if entry_type:
                 type_key = entry_type if isinstance(entry_type, str) else str(entry_type)
                 metadata["entry_counts"][type_key] = metadata["entry_counts"].get(type_key, 0) + 1
 
@@ -306,7 +311,7 @@ def _process_assistant(
                 "output_tokens": _safe_int(usage.get("output_tokens")),
                 "cache_read": _safe_int(usage.get("cache_read_input_tokens")),
                 "cache_creation": _safe_int(usage.get("cache_creation_input_tokens")),
-                "service_tier": usage.get("service_tier"),
+                "service_tier": tier if isinstance(tier, str) else None,
             },
         }
     )