is this going to work if passed a numpy float?

perhaps numpy.isreal and numpy.isfinite would be better to use here.

I think it still works, but I think we should leverage as much as possible numpy considering it is one of our dependencies.

think of it this way. We recently discovered some bugs that were because numpy ints and python ints aren't the same in py3k, and so isinstance(a, int) of a numpy int wouldn't work anymore. Better safe than sorry.

These won't work because the default axes object do not have these methods. You will need to explicitly create the 3d axes object and call the methods on it.

I am not an expert in pytest, but I am not quite sure this is correct. Wouldn't the err object get clobbered in each test, so you are really only comparing with the last one?

Do exceptions have this message attribute in py3k? The test suite is failing for py3k, saying that it doesn't (but it works for python27).

along those lines, I'm comfortable with simply checking the error type

this is code block is repeated several times throughout the PR.

i recommend factoring it out into its own method. something like:

...
class _AxesBase(martist.Artist):
    ...
    def _validate_axis_limit(limit):
        if limit is not None:
            if isinstance(limit, float) and
                    (not np.isreal(limit) or not np.isfinite(right)):
                raise ValueError(...)
            return self.convert_xunits(limit)

and then all of these blocks can just be, e.g.

right = self._validate_axis_limit(right)

I think you should try to parametrize the tests too:

a first pass would be something as simple as:

@pytest.mark.parametrize('value', [np.inf, np.nan])
def test_invalid_axes_limits(value)
    with pytest.raises(ValueError):
        plt.xlim(left=value)

    with pytest.raises(ValueError):
        plt.xlim(right=value)

    with pytest.raises(ValueError):
        plt.ylim(bottom=value)

    with pytest.raises(ValueError):
        plt.ylim(top=value)

Going two steps further, you could parametrize the limit functions and sides, but that might be going a bit too far:

@pytest.mark.parametrize('value', [np.inf, np.nan])
@pytest.mark.parametrize(('setter', 'side'), [
    (plt.xlim, 'left'),
    (plt.xlim, 'right'),
    (plt.ylim, 'bottom'),
    (plt.ylim, 'top'),
])
def test_invalid_axes_limits(setter, side, value)
    limit = {side: value}
    with pytest.raises(ValueError):
        setter(**limit)

I think you might have the 'side' and 'setter' swapped in the parametrize call.

@QuLogic yup I'll fix

Cleanup will work on the test, but I don't think it will work right on the parametrization. Consequently, I don't think it makes sense to create the figure in this list. It should be created in the test itself and only the setter name should be here. You can use getattr to dynamically access the method you want.

I still recommend moving the convert_xunits into your _validate... method

might want to do the conversion first in case that causes it to be invalid?

First line should be a full sentence; can be shortened to "Raise if specified axis limits are infinite."

Users aren't going to know what "argument values" means; should say "Axis limits cannot be ..."

Only this line should be in the context, just in case the first two raise other things accidentally.

Oh bother, I just noticed this should probably be checking the converted limits. Do you want to open a new PR for that, @nvedant07?

Sure , i have opened a new PR for this - #8474