Implementing a fixture with a single setup but a dedicated instance per test

[standage]

I have a scenario in which a fixture setup task consumes 30-45 seconds to populate a working directory. I then have about a dozen tests that use the working directory created by this fixture. Each test needs to have its own dedicated instance of the working directory, but given how time-intensive the setup is the fixture should only be run once. Is this possible with tmp_path, tmp_path_factory, or other pytest features?

To ground this scenario in a concrete example, I've included a mock test script below. The workdir function is the fixture that gets passed to each test function, and it in turn calls the long_setup_task function which simulates a time-consuming task. The (very creatively named) one, two, and three functions simulate different operations to be performed on a dedicated copy of a working directory. And then the test_[one|two|three] functions test the expected result.

The script below satisfies my requirements, but it feels like I'm abusing pytest's features rather than using them as intended. Is there any cleaner, more idiomatic way to accomplish this? Something that doesn't require me running copytree in each test?

from pathlib import Path
import pytest
from shutil import copytree
from tempfile import TemporaryDirectory
from time import sleep


def long_setup_task(tmp):
    sleep(30)
    with open(tmp / "data.tsv", "w") as fh:
        print("A", "1", sep="\t", file=fh)
        print("B", "2", sep="\t", file=fh)
        print("C", "3", sep="\t", file=fh)
    with open(tmp / "log.txt", "w") as fh:
        print("do_long_task complete", file=fh)


def one(wd):
    with open(wd / "data.tsv", "a") as fh:
        print("D", "4", sep="\t", file=fh)
    with open(wd / "log.txt", "a") as fh:
        print("one complete", file=fh)


def two(wd):
    with open(wd / "data.tsv", "a") as fh:
        print("G", "7", sep="\t", file=fh)
    with open(wd / "log.txt", "a") as fh:
        print("two complete", file=fh)


def three(wd):
    with open(wd / "data.tsv", "a") as fh:
        print("†", "0", sep="\t", file=fh)
    with open(wd / "log.txt", "a") as fh:
        print("three complete", file=fh)


@pytest.fixture(scope="session")
def workdir():
    with TemporaryDirectory() as tmp:
        tmp = Path(tmp)
        long_setup_task(tmp)
        yield tmp


def test_one(workdir, tmp_path):
    copytree(workdir, tmp_path, dirs_exist_ok=True)
    one(tmp_path)
    with open(tmp_path / "data.tsv", "r") as fh:
        observed = fh.read()
    expected = "A\t1\nB\t2\nC\t3\nD\t4\n"
    assert observed == expected


def test_two(workdir, tmp_path):
    copytree(workdir, tmp_path, dirs_exist_ok=True)
    two(tmp_path)
    with open(tmp_path / "data.tsv", "r") as fh:
        observed = fh.read()
    expected = "A\t1\nB\t2\nC\t3\nG\t7\n"
    assert observed == expected


def test_three(workdir, tmp_path):
    copytree(workdir, tmp_path, dirs_exist_ok=True)
    three(tmp_path)
    with open(tmp_path / "data.tsv", "r") as fh:
        observed = fh.read()
    expected = "A\t1\nB\t2\nC\t3\n†\t0\n"
    assert observed == expected

[RonnyPfannschmidt]

your question indicates that the "costly setup" cannot be separated form a "instance"
so unless you find a less expensive way to replicate a setup, there is simply not technical way to create it to begin with

[RonnyPfannschmidt]

nm, i missread the initial question

its perfectly valid to have a wd fixture that does a copytree from a wd_template fixture

[standage]

Do you mean something like this? It seems to work for me, and certainly is less kludgy.

@pytest.fixture(scope="session")
def workdir_template():
    with TemporaryDirectory() as tmp:
        tmp = Path(tmp)
        long_setup_task(tmp)
        yield tmp


@pytest.fixture
def workdir(workdir_template, tmp_path):
    copytree(workdir_template, tmp_path, dirs_exist_ok=True)
    return tmp_path


def test_one(workdir):
    one(workdir)
    with open(workdir / "data.tsv", "r") as fh:
        observed = fh.read()
    expected = "A\t1\nB\t2\nC\t3\nD\t4\n"
    assert observed == expected


def test_two(workdir):
    two(workdir)
    with open(workdir / "data.tsv", "r") as fh:
        observed = fh.read()
    expected = "A\t1\nB\t2\nC\t3\nG\t7\n"
    assert observed == expected


def test_three(workdir):
    three(workdir)
    with open(workdir / "data.tsv", "r") as fh:
        observed = fh.read()
    expected = "A\t1\nB\t2\nC\t3\n†\t0\n"
    assert observed == expected

[RonnyPfannschmidt]

Yes, alltho i recommend usage of tmp_path_factory to ensure debugging