Source code for codegrade.models.extended_auto_test_run

"""The module that defines the ``ExtendedAutoTestRun`` model.

SPDX-License-Identifier: AGPL-3.0-only OR BSD-3-Clause-Clear
"""

import typing as t
from dataclasses import dataclass, field

import cg_request_args as rqa

from .. import parsers
from ..utils import to_dict
from .auto_test_result import AutoTestResult
from .auto_test_run import AutoTestRun


[docs]@dataclass class ExtendedAutoTestRun(AutoTestRun): """The run as extended JSON.""" #: The results in this run. This will only contain the result for the #: latest submissions. results: "t.Sequence[AutoTestResult]" #: The stdout output of the `run_setup_script`. Deprecated, please use the #: value on a result. setup_stdout: "str" #: The stderr output of the `run_setup_script`. Deprecated, please use the #: value on a result. setup_stderr: "str" raw_data: t.Optional[t.Dict[str, t.Any]] = field(init=False, repr=False) data_parser: t.ClassVar = rqa.Lazy( lambda: AutoTestRun.data_parser.parser.combine( rqa.FixedMapping( rqa.RequiredArgument( "results", rqa.List(parsers.ParserFor.make(AutoTestResult)), doc="The results in this run. This will only contain the result for the latest submissions.", ), rqa.RequiredArgument( "setup_stdout", rqa.SimpleValue.str, doc="The stdout output of the `run_setup_script`. Deprecated, please use the value on a result.", ), rqa.RequiredArgument( "setup_stderr", rqa.SimpleValue.str, doc="The stderr output of the `run_setup_script`. Deprecated, please use the value on a result.", ), ) ).use_readable_describe(True) ) def to_dict(self) -> t.Dict[str, t.Any]: res: t.Dict[str, t.Any] = { "results": to_dict(self.results), "setup_stdout": to_dict(self.setup_stdout), "setup_stderr": to_dict(self.setup_stderr), "id": to_dict(self.id), "created_at": to_dict(self.created_at), "state": to_dict(self.state), "is_continuous": to_dict(self.is_continuous), } return res @classmethod def from_dict( cls: t.Type["ExtendedAutoTestRun"], d: t.Dict[str, t.Any] ) -> "ExtendedAutoTestRun": parsed = cls.data_parser.try_parse(d) res = cls( results=parsed.results, setup_stdout=parsed.setup_stdout, setup_stderr=parsed.setup_stderr, id=parsed.id, created_at=parsed.created_at, state=parsed.state, is_continuous=parsed.is_continuous, ) res.raw_data = d return res
import os if os.getenv("CG_GENERATING_DOCS", "False").lower() in ("", "true"): import datetime