On 22/11/2018 11:22, Stephan Bergmann wrote:
What does make a difference for these Windows build failures is passing SAL_USE_VCLPLUGIN=svp into JunitTest. <https://gerrit.libreoffice.org/plugins/gitiles/core/+/6cdfe5ebb4f6c06bfa8b0e67e778dd68131c14e3%5E!/> "Drop some headless mode variants" did two seemingly independent, but actually interacting changes:For one, it changed OpenGLHelper::isVCLOpenGLEnabled (vcl/source/opengl/OpenGLHelper.cxx) to disable OpenGL when SAL_USE_VCLPLUGIN=svp is set (which, in the past, was only set for certain tests on Linux).For another, it added SAL_USE_VCLPLUGIN=svp to solenv/gbuild/CppunitTest.mk in such a way that it would effectively be enabled during all tests, on all platforms (and which <https://gerrit.libreoffice.org/plugins/gitiles/core/+/d5ed903618f200456feed9b410b7bd1ed8daeb62%5E!/> "Set CppunitTest-related env vars only during CppunitTest" then "fixed", by restricting it to just CppunitTests).I asked Jan-Marek to come up with a fix (presumably setting SAL_USE_VCLPLUGIN=svp in more places again) that makes OpenGLHelper::isVCLOpenGLEnabled behave as expected (whatever that means) during tests. I must confess I'm lost here... :)
...and it may also be that OpenGLHelper::isVCLOpenGLEnabled already behaves as expected, and SAL_USE_VCLPLUGIN=svp is set in exactly the right set of tests, and these Windows build failures are symptoms of other issues that need to be fixed (and had just been hidden by accident)