On Fri, Oct 16, 2015 at 10:15 AM, Robert Ramey wrote:

On 10/15/15 6:54 PM, Tom Kent wrote:

...

I think there is a bug in the serialization library that is breaking out of the test harness and causing window's error handling to kick in (when running on Windows, obviously). Usually when this happens the test runner application eventually (I believe) times out the test and marks it as failed. However, there are an incredible number of failures happening (since late last week) and they are bogging down the entire test run. Most of the runs I have attempted in the last week on windows have failed...the ones that have succeeded have taken ~12hrs up from the normal 3-4hrs.

I was able to capture a stack trace of one of these random failures: msvcp140d.dll!00007ff9cfffb968() Unknown msvcp140d.dll!00007ff9cff73005() Unknown

...

test_list_ptrs_text_warchive.exe!std::basic_filebuf

...

::_Endwrite() Line 649 C++

test_list_ptrs_text_warchive.exe!std::basic_filebuf

...

::close() Line 341 C++

test_list_ptrs_text_warchive.exe!std::basic_filebuf

...

::~basic_filebuf >() Line 155 C++

test_list_ptrs_text_warchive.exe!std::basic_ofstream

...

::~basic_ofstream >() Line 1068 C++

test_list_ptrs_text_warchive.exe!std::basic_ofstream

...

::`vbase destructor'() C++

test_list_ptrs_text_warchive.exe!test_list() Line 65 C++ test_list_ptrs_text_warchive.exe!test_main(int __formal, char * * __formal) Line 176 C++ test_list_ptrs_text_warchive.exe!main(int argc, char * * argv) Line 203 C++ [External Code]

I'd be happy to help get to the bottom of this, if you can give me any more diagnostic steps to take. We need to resolve this ASAP because it is preventing the windows test runners from producing timely results and we have a release coming up.

I have been making changes in he serialization library recently - fixing bugs. This turns out to be more difficult than first meets the eye. So I'm a little bogged down on this. I did push some changes to develop, things looked OK so I merged to master. I don't know if this is related to the current issue.

As soon as I get my current issues resolved, I'll re-pull from master and quadiple check that things work in my environments. Unfortunately(?) that doesn't include windows.

Would it be possible to revert the changes to develop until you have a better handle on them, so that the runners can continue for all the other projects? This would also rule-out the problem actually being a change in some other library, that is just showing up here. Tom