# Licensed under a 3-clause BSD style license - see LICENSE.rst

"""Test `astroquery.utils.timer`.

.. note::

The tests only compare rough estimates as

performance is machine-dependent.

"""

# STDLIB

import time

# THIRD-PARTY

import pytest

import numpy as np

from astropy.utils.exceptions import AstropyUserWarning

from astropy.modeling.fitting import ModelsError

# LOCAL

from ..timer import RunTimePredictor

def func_to_time(x):

"""This sleeps for y seconds for use with timing tests.

.. math::

y = 5 * x - 10

"""

y = 5.0 * np.asarray(x) - 10

time.sleep(y)

return y

@pytest.mark.flaky(reruns=1, reruns_delay=10)

def test_timer():

"""Test function timer."""

p = RunTimePredictor(func_to_time)

# --- These must run before data points are introduced. ---

with pytest.raises(ValueError):

p.do_fit()

with pytest.raises(RuntimeError):

p.predict_time(100)

# --- These must run next to set up data points. ---

with pytest.warns(AstropyUserWarning, match="ufunc 'multiply' did not "

"contain a loop with signature matching types"):

p.time_func([2.02, 2.04, 2.1, 'a', 2.3])

p.time_func(2.2) # Test OrderedDict

assert p._funcname == 'func_to_time'

assert p._cache_bad == ['a']

k = list(p.results.keys())

v = list(p.results.values())

np.testing.assert_array_equal(k, [2.02, 2.04, 2.1, 2.3, 2.2])

np.testing.assert_allclose(v, [0.1, 0.2, 0.5, 1.5, 1.0])

# --- These should only run once baseline is established. ---

with pytest.raises(ModelsError):

a = p.do_fit(model='foo')

with pytest.raises(ModelsError):

a = p.do_fit(fitter='foo')

a = p.do_fit()

assert p._power == 1

# Perfect slope is 5, with 10% uncertainty

assert 4.5 <= a[1] <= 5.5

# Perfect intercept is -10, with 1-sec uncertainty

assert -11 <= a[0] <= -9

# --- These should only run once fitting is completed. ---

# Perfect answer is 490, with 10% uncertainty

t = p.predict_time(100)

assert 441 <= t <= 539

# Repeated call to access cached run time

t2 = p.predict_time(100)

assert t == t2