For the segfaults, initially I was suspicious of qWaitForWindowExposed vs. qWaitForWindowActive but after changing those, while the segfaults appeared to happen less regularly; they still do occur.

Furthermore, they do not happen when running any single test by itself.

tests/graphicsItems/test_ArrowItem.py .                                                                                                                                                                  [ 74%]
tests/graphicsItems/test_AxisItem.py ......                                                                                                                                                              [ 78%]
tests/graphicsItems/test_ErrorBarItem.py .                                                                                                                                                               [ 79%]
tests/graphicsItems/test_GraphicsItem.py ..                                                                                                                                                              [ 80%]
tests/graphicsItems/test_ImageItem.py sszsh: segmentation fault  pytest -x -k "not examples"

~/Developer/pyqtgraph merge-test-data* 10s
pyqtgraph-pyside2_515-py38 ❯ pytest -k "test_ImageItem"
============================================================================================= test session starts ==============================================================================================
platform darwin -- Python 3.8.6, pytest-6.2.1, py-1.10.0, pluggy-0.13.1
rootdir: /Users/ogi/Developer/pyqtgraph, configfile: pytest.ini
plugins: cov-2.11.1, xdist-2.2.0, forked-1.3.0
collected 247 items / 239 deselected / 8 selected

tests/graphicsItems/test_ImageItem.py ss...                                                                                                                                                              [ 62%]
tests/graphicsItems/test_ImageItemFormat.py ...                                                                                                                                                          [100%]

================================================================================= 6 passed, 2 skipped, 239 deselected in 2.88s =================================================================================

~/Developer/pyqtgraph merge-test-data*
pyqtgraph-pyside2_515-py38 ❯

found a bunch more tests, migrated them over as well, ...segfaults occur in

• test_ImageItem.py
• test_ViewBox.py
• test_InfiniteLine.py
• test_NonUniformImage.py

this command runs without any errors! 😆

pytest tests -k "not test_ImageItem and not test_InfiniteLine and not test_NonUniformImage and not test_ViewBox"

now that the CI is looking at the tests directory, the segfault shows up there too, so at least we have that going for us...

I can no longer reproduce a segfault locally, but two segfaults in the CI is enough to make me suspicious.

Last two runs have shown no segfaults; will rerun a few more times to build some confidence in the results.

Tests are migrated over, static code quality checker is fine, "silenced" exceptions have been addressed. I think the next step is to regenerate images with Qt5 (or Qt6), and tighten the tolerance for assertImageApproved for Qt5 and Qt6.

Dear @j9ac9k , thanks for the massive overhaul.
With the revised tests, you might want to check if
#1025 and #1064 can be closed!

Good news regarding the pixel tolerance, it does not appear to be needed when the images are generated using Qt6 bindings. I've only done the PlotCurveItem images, I'm going to do the other images shortly (I'm setup to do one at a time, so it's going to take a bit!).

ok, I made CI angry; ...will investigate more later. With the new images, tox passed on macOS across the board with the new images

EDIT: I can replicate locally; must be something I just did....will investigate more later tonight

passes on macOS, segafaults on linux, crashes on windows, "works on my machine!"

EDIT: looks like it didn't like my changes to image_testing.py thank you sublime merge for letting me easily restore the change to that one file in the one commit (and leaving the rest of the commit untouched).

So this is the hold up on windows, there appears to be 1 blue and 1 yellow pixel that is the difference between what was shown, and what was expected. Will tinker with scaling settings....

If you want to refactor into the function getImageFromWidget(), then imageToArray() needs to be called with copy=True, or manually make a copy of the resultant ndarray.

If you want to refactor into the function getImageFromWidget(), then imageToArray() needs to be called with copy=True, or manually make a copy of the resultant ndarray.

I'll keep that in mind, but I reverted that change (due to the segfaults on ubuntu).... I may look to refactor it out, as I think it can provide useful functionality on its own; but right now I'm debating how to pass the assertImageAppoved tests, without having to provide a fudge-factor for Windows systems, which at a glance looks like what I will have to do.

Oh the test_getArrayRegion tests, only half of the tests within there fail, specifically the following two fail

        (pg.ROI([1, 1], [27, 28], pen='y'), 'baseroi'),
        (pg.RectROI([1, 1], [27, 28], pen='y'), 'rectroi'),

The other ones pass

        (pg.EllipseROI([1, 1], [27, 28], pen='y'), 'ellipseroi'),
        (pr, 'polylineroi'),

Just noticed the following comment in the code:

# on windows, one edge of one ROI handle is shifted slightly; letting this slide with pxCount=10

This was applied to one of the tests failing; but not to all the ones failing...

I'm doing some last minute cleanup on this PR, I'm going to let it sit for a day, and merge, so we can get back to #1796 and ensure the image tests pass that way.

and now I screwed up my git history here 🤦🏻

Ok, now I think this PR is ready.

@pijyoi the static code analyzer doesn't like the change to functions.py

    if depth in (8, 24, 32):
        dtype = np.uint8
        nchan = depth // 8
    elif depth in (16, 64):
        dtype = np.uint16
        nchan = depth // 16
    shape = qimg.height(), qimg.width()
    if nchan != 1:
        shape = shape + (nchan,)
    data = np.frombuffer(img_ptr, dtype=dtype).reshape(shape)
    return data

Specifically, it is complaining about nchan and dtype being potentially uninitialized. Should I provide an else with a runtime error or something along those lines?

Sure, please fix it with a ValueError, unsupported image type.

nonuniform_image/colormap-3x3.png looks very different. Probably due to some older PR breaking things w/o the test suite catching it.

master image from separate github repo

this PR

UPDATE: It seems that the commit 510626c that added the test was already generating the bottom image!
The offending line:
https://github.com/pyqtgraph/pyqtgraph/blob/master/pyqtgraph/graphicsItems/tests/test_NonUniformImage.py#L96

cmap = ColorMap(pos=[0.0, 1.0], color=[(0.0, 0.0, 0.0, 1.0), (1.0, 1.0, 1.0, 1.0)])

should perhaps be

cmap = ColorMap(pos=[0.0, 1.0], color=[(0, 0, 0), (255, 255, 255)])

I am not familiar with ColorMap usage, so maybe @NilsNemitz could help here.

@pijyoi you are spot on:

I'll see if I can reuse the image that's still on test-data, if not, can easily regenerate.

EDIT:

Sure enough, looks like I'll have to regenerate:

Updated the PR , unless anyone spots anything funky (including my screwing up of my git history, which I've only done a dozen or so times on this branch), I think this should be ready to merge (of course pending the result of the static code checker).