Windows CI: Test multiple fixes against CERN/master. #177
Conversation
marsupial
force-pushed
the
PR/21
branch
2 times, most recently
from
50089d9
to
7600cba
Compare
|
I quickly tried it and more tests are failing on my local desktop (the new Windows tests are failing) |
|
I'm still pushing into it, but last run on Appveyor was down to 6 fails with CMake, and 5 with Ninja. The new Windows test will likely fail until moving to COFF or someone writes the code to support COMDATs on ELF. The point was to pull out the point of failure from multiple tests into one location so for example OutputRedirect.C -real- failures could be adressed. |
|
OK, I see. Thanks. And I suppose this will supersede the ROOT PR-745 (originally Cling PR-169), right? |
|
I just got tired of "This doesn't work" so wanted to gather all the Windows stuff into one place, as there can be multiple reasons for failures in a single test. And cherry-picking these out of my tree was often causing conflicts. I think the the TLS stuff should go in regardless, as it will currently crash hard on OS X & Linux, so shouldn't depend on whenever these may get in (but it is necessary for Windows to pass as well). |
cling is currently tearing down the text input before the Interpreter is destroyed. Leading to problems with pipes and file handles being close in the text input layer.
…tor. DeRegisterEHFrames can be called without a a prior call to RegisterEHFrames. This can easily be detected so just do nothing when that occurs.
This reverts commit 5947e13.
…ests and make originals pass.
…tually tested, and not whether std::cout is still broken.
…ing when they don’t exist.
…Interpreter::Initialize.
…, use the module to get them.
MSVC registers the facets in a linked list via _Facet_Register to be destroyed after main returns, but the memory for these objects will have been deallocated before that occurs.
This can lead to corruption of the owning process’ static variables.
…_address. Fixes function level static variables.
No description provided.