It appears there was a regression related to recent changes in mesa/readpix. It could be that the changes exposed something bad in our code, but it works everywhere else so ... Specifically it happened with Last week this test started failing on hythloth's nightly Mesa builds (along with a few other tests): * OpenGL2: https://open.cdash.org/viewTest.php?onlyfailed&buildid=4229706 https://open.cdash.org/testDetails.php?test=418931875&build=4229706 * OpenGL1: https://open.cdash.org/viewTest.php?onlyfailed&buildid=4230126 https://open.cdash.org/testDetails.php?test=418952240&build=4230126 For now I've checked out a mesa that still works and stopped updating. The problem bisects to: mesa/readpix: Clip ReadPixels() area to the ReadBuffer's https://cgit.freedesktop.org/mesa/mesa/commit/?id=605832736a6d9427ad894d403cceeb74a5b18dc1 and still occurs after some related follow-up changes to Mesa: mesa/readpix: Don't clip in _mesa_readpixels() https://cgit.freedesktop.org/mesa/mesa/commit/?id=b89a8a15c240418d1859947b5114993ecdf424fb mesa/readpix: Dedent former _mesa_readpixels() if block https://cgit.freedesktop.org/mesa/mesa/commit/?id=c624241ef47c3a6ea2d1177969328bd4547e45a2 Valgrind reports that vtkImageDifference encounters uninitialized memory after the change to Mesa but not before. That looks consistent with the messy corner of the test image seen in the above links. Mesa is no longer populating the full image with data. I'll post an apitrace file run with an old mesa that hopefully will show the problem for you should you not want to build VTK yourself.
Should already be fixed with Mesa commit 83b589301f4a150f4b1b13fd3ffd9f6d98ee6546 from an hour ago. Closing this as fixed. Please re-open if you still have a problem after updating Mesa.
Created attachment 121950 [details] apitrace apitrace file for the test that is exposing the problem. The test renders four times in a small window, reading the pixels after each render. Then pastes the images together into a larger image. See the links in the bug report for the failing versus valid images.
Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.