[ImageJ-devel] IJ1 unit tests

[Curtis Rueden]

Hi Wayne,

I was able to run all the failing tests.
>

Thanks for investigating. Barry is updating the tests now in response to
your comments, and will reply with any remaining issues. We should have all
the tests passing again very soon, and will get them automated soon as well.

Regards,
Curtis


On Sun, Oct 9, 2011 at 6:20 PM, Rasband, Wayne (NIH/NIMH) [E] <
rasbandw at mail.nih.gov> wrote:

> Hi Curtis & Barry,
>
> I was able to run all the failing tests. Thanks for pointing me to the
> right place on imagejdev website.
>
> >1. testDrawPixelsImageProcessor(ij.gui.ArrowTest): expected:<0> but
> was:<73>
>
> ImageJ 1.45 uses a different method for drawing arrows. This change was
> required to support outlined arrows. Arrows drawn with the new method do not
> have the exact same pixel layout. I would recommend removing this test.
>
> > 2. testFitSplineForStraightening(ij.gui.PolygonRoiTest): array lengths
> differed, expected.length=4 actual.length=8
>
> This test assumes that the number of points in a polygon ROI is the same as
> the size of the array returned by getXCoordinates(), which is not
> necessarily the case. It is better to use getPolygon() or getFloatPolygon().
>
> > 3. testGetUncalibratedLength(ij.gui.PolygonRoiTest): expected:<1.0> but
> was:<6.324555320336759>
>
> This is a bug in the test. The distance between (1,3) and (5,11) is not
> 1.0.
>
> > 4. testGetConvexHull(ij.gui.ShapeRoiTest)
>
> This test assumes that ShapeRoi.getConvexHull() always returns null, which
> is no longer the case.
>
> > 5. testGetFeretValues(ij.gui.ShapeRoiTest): arrays first differed at
> element [0]; expected:<10.44016> but was:<10.44030650891055>
>
> The ShapeRoi.getFeretValues() method in v1.45 calculates the Feret diameter
> of simple ROIs more accurately. Perhaps this test needs to use a larger
> tolerance value.
>
> > 6. testGetPixels(ij.VirtualStackTest): arrays first differed at element
> [2]; expected:<0> but was:<120>
> > 7.    testGetProcessor(ij.VirtualStackTest): arrays first differed at
> element [2]; expected:<0> but was:<120>
>
> The virtual stack tests assume the two images in the virtual stack are 2x3
> but they are actually 3x2.
>
> Best regards,
>
> -wayne
>
>
> On Oct 7, 2011, at 3:05 PM, Curtis Rueden wrote:
>
> > Hi Wayne et. al,
> >
> > For quite some time now, we have had unit tests (written by Barry) for
> testing the behavior of IJ1. Unfortunately, they were originally written to
> compile against a patched version of IJ1 and we neglected to update them to
> work with IJ1 "out of the box."
> >
> > I have had a ticket about this situation (
> http://dev.imagejdev.org/trac/imagej/ticket/598) for the past few months,
> but have only just now finally fixed the tests to work against various
> versions of IJ1 from 1.44 onwards.
> >
> > I ran them against all development versions of 1.45 (a through q) and
> saved the result as a blog post:
> >   http://imagejdev.org/2011/10/07/unit-tests-imagej-1x
> >
> > It looks like from 1.45a through 1.45q, a total of 7 code changes were
> introduced that result in test failures. These could be the result of
> bugfixes, or they could be new regressions. I wanted to let you know in case
> you were interested in pursuing these bugs, since I know how much you value
> backwards compatibility.
> >
> > Instructions on running the tests yourself can be found here:
> >   http://imagejdev.org/source-code#Running_ImageJ_1.x_unit_tests
> >
> > My plan is to hook up the tests to Hudson (
> http://dev.imagejdev.org:8080/) and have it send an email whenever a new
> version of IJ1 is released with failing tests.
> >
> > What do you think?
> >
> > Regards,
> > Curtis
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://imagej.net/pipermail/imagej-devel/attachments/20111010/45e82917/attachment.html>