Add a logarithmic scale slider class

[GenevieveBuckley]

Juan originally suggested a log scale slider for napari (napari/napari#938) but this strikes me as something that would also be very useful to have available in magicgui.

Adds numpy as a base requirement.

The main limitation of the log slider is that the available range is restricted from 0 -> 14.5. If you go any higher than 14.5 you get an overflow error from Qt. Qt expects internal values in the range of -2147483647 to +2147483647, so the exponential can cause problems quickly. Still, I think there is some value in having a log slider available, even if there are some restrictions here.

[codecov]

Codecov Report

Merging #38 into master will increase coverage by 0.15%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master      #38      +/-   ##
==========================================
+ Coverage   95.82%   95.98%   +0.15%     
==========================================
  Files          14       16       +2     
  Lines         934      971      +37     
==========================================
+ Hits          895      932      +37     
  Misses         39       39              

Impacted Files	Coverage Δ
magicgui/_qt/widgets/__init__.py	100.00% <100.00%> (ø)
magicgui/_qt/widgets/double_slider.py	100.00% <100.00%> (ø)
magicgui/_qt/widgets/log_slider.py	100.00% <100.00%> (ø)
magicgui/_tests/test_qt.py	99.01% <100.00%> (ø)
magicgui/_tests/test_widgets.py	100.00% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update db4c816...981efac. Read the comment docs.

[GenevieveBuckley]

Ok, something weird is going on with the tests. I can't see how the changes here could possibly have caused this, but here's what I've found.

• Tests pass on the current master branch (I can reproduce this locally).
• On this log-slider branch, all the CI test runs report a test failure for magicgui/_tests/test_qt.py::test_event (I can reproduce this locally too).
• If you run that specific test by itself, it passes fine: 'pytest magicgui/_tests/test_qt.py::test_event`
• But if you run any other test using a qtbot fixture (or the magic_widget fixture, since that fixture also uses qtbot) before it, test_event will fail because of the line: assert not QtW.QApplication.instance(). I've checked this theory by running an example test before test_event that does nothing but assert True but does take in the qtbot fixture:

def test_example(qtbot):
    assert True

So, what to do?

If we can remove the line assert not QtW.QApplication.instance() from test_event then we'd have no more test failures - but I assume the reason it's there in the first place is because it's checking something important? I don't know why it's important that we see/expect a null pointer here (which is what the Qt docs say is supposed to happen if no instance has been allocated), so I'd really like to hear suggestions.

[tlambert03]

that was a pretty fragile test to begin with and was very dependent on the order of operations. Once a qtbot makes an app, it sticks around for the rest of the tests. I was asserting that there was no instance so that we could make sure that the event_loop context manager (same as napari.gui_qt basically) was actually creating the app.

I just pushed a fix here, which deletes any pre-existing application instance using sip.delete or shiboken.2 delete... a clever little trick that Phillip recently found on stack overflow.

[GenevieveBuckley]

Thank you for the fix! Very helpful :)

[tlambert03]

welp... looks like pyside2 was ok with that, but not pyqt5. I just skipped the test for pyqt5.

[tlambert03]

to make this slightly more flexible, perhaps we should add an attribute self.base = math.e ... then convert math.exp(value) to math.pow(self.base, value) and math.log(value) to math.log(value, base=self.base)

[GenevieveBuckley]

Nice touch, I like this suggestion a lot. Done!

[GenevieveBuckley]

Ok I think I've addressed all the reviewer suggestions, and we have a passing test suite now (thanks to Talley). I think things are looking good here.

[tlambert03]

thanks!