mirror of
https://github.com/kuhyx/testsAndMisc.git
synced 2026-11-11 00:47:01 +01:00
Krzysztof kuhy Rudnicki
40a4eaf5b5
_challenge_support.py 402 -> 237. _pytest_clean and _patch_and_test go to _pytest_runner.py with the subprocess/os/sys imports they solely used; _stream_verdict and _parse_verdict go to _verdict.py with rich's Live. code_tutor patches by absolute string rather than a MOD constant, so the seam cost 15 per-name rewrites instead of one sed. Two intermediate failures caught by running it: the test modules import the moved names directly, and five patch strings still named the old module. Pass count is 216 before and after. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011w2Pj4Ek3ZNMQVYcFHgXDY
90 lines
2.6 KiB
Python
90 lines
2.6 KiB
Python
"""Streaming an LLM verdict behind an elapsed-time display, and parsing it.
|
|
|
|
Split out of :mod:`python_pkg.code_tutor._challenge_support` to keep it under
|
|
the 250-line cap. ``Live`` is imported here rather than there because this is
|
|
the only code that used it, and the tests patch it on whichever module the
|
|
call resolves through.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import json
|
|
import re
|
|
import time
|
|
from typing import TYPE_CHECKING
|
|
|
|
from rich.live import Live
|
|
from rich.text import Text
|
|
|
|
if TYPE_CHECKING:
|
|
from rich.console import Console
|
|
|
|
from python_pkg.code_tutor._llm import Backend
|
|
|
|
|
|
def _stream_verdict(
|
|
system: str,
|
|
user_msg: str,
|
|
backend: Backend,
|
|
console: Console,
|
|
label: str = "Rating",
|
|
) -> str:
|
|
"""Stream an LLM call silently while showing an elapsed timer.
|
|
|
|
Args:
|
|
system: System prompt for the judge.
|
|
user_msg: User message to judge.
|
|
backend: LLM backend with a ``stream(system, user, callback)`` method.
|
|
console: Rich console for the live timer.
|
|
label: Label prefix shown in the timer (e.g. ``"Rating"``).
|
|
|
|
Returns:
|
|
Accumulated response text.
|
|
"""
|
|
parts: list[str] = []
|
|
start = time.monotonic()
|
|
|
|
def _on_token(token: str) -> None:
|
|
parts.append(token)
|
|
elapsed = int(time.monotonic() - start)
|
|
live.update(Text(f"{label}... {elapsed}s", style="yellow"))
|
|
|
|
with Live(
|
|
Text(f"{label}... 0s", style="yellow"),
|
|
console=console,
|
|
refresh_per_second=4,
|
|
transient=True,
|
|
) as live:
|
|
backend.stream(system, user_msg, _on_token)
|
|
|
|
return "".join(parts)
|
|
|
|
|
|
def _parse_verdict(raw: str) -> tuple[str, str]:
|
|
"""Parse ``{"verdict": ..., "gap": ...}`` JSON from *raw*, tolerating fences.
|
|
|
|
Args:
|
|
raw: Raw LLM response text.
|
|
|
|
Returns:
|
|
``(verdict, gap)`` where verdict is ``"PASS"`` or ``"FAIL"``.
|
|
"""
|
|
clean = re.sub(r"```(?:json)?\s*", "", raw, flags=re.DOTALL).strip()
|
|
start = clean.find("{")
|
|
end = clean.rfind("}") + 1
|
|
if start == -1 or end == 0:
|
|
return "FAIL", "Could not parse judge response."
|
|
try:
|
|
data = json.loads(clean[start:end])
|
|
except json.JSONDecodeError:
|
|
return "FAIL", "Could not parse judge response."
|
|
verdict = str(data.get("verdict", "FAIL")).upper()
|
|
if verdict not in {"PASS", "FAIL"}:
|
|
verdict = "FAIL"
|
|
return verdict, str(data.get("gap", ""))
|
|
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# "Write tests first" flow -- helpers
|
|
# ---------------------------------------------------------------------------
|