The commit assumes we update the used Boost libs for Windows to version 1.49 to switch to the latest release.
If we switch over to 1.49 on Windows, we need to add oleaut32 lib in bitcoin-qt.pro, to avoid the compilation error "undefined reference to _imp__VariantClear@4", because VariantClear() is used in the boost::interprocess code.
In the end we would also need to update the qtgui_deps_1.zip with the 1.49 boost libs for Windows and the 1.49 pre-compiled header files.
Were you able to reproduce the crash before this change? I guess not, because this wasn't fixed in Boost until 1.48!
I debugged that crash during RC2 and it's clearly related to boost interprocess, see: https://bitcointalk.org/index.php?topic=66887.msg779221#msg779221. As I have never been able to compile bitcoin with the supplied 1.43 libs with the current Qt SDK I could simply not try to reproduce the error in the "lab", sorry.
How about using the stock release builds to try reproducing it?
RC2 used the gitian build process, which uses Boost 1.47.
See contrib/gitian-descriptors/gitian-win32.yml
It WOULD be interesting to know if upgrading to the latest boost (1.49) fixes the problem.
@luke-jr: What you mean by stock release builds? @gavinandresen Is there a place to download the pre-compiled boost libs which were used in RC2? I would like to make a file-compare. Another thing that is weird for me ... why is 1_43 in the bitcoin-qt.pro and why does it compile on Gitian without that VariantClear error?
I'll try to "crash" my PC by simply shutting it of, while I have a running instance of my version ... :-/.
Which MinGW version is used in Gitian?
1.47:
1.49:
1.49 fixes one thing unrelated to Bitcoin on Windows, now C:\ProgramData\boost_interprocess\ is used instead of C:\ProgramData\boost_interprocess\Select LastBootUpTime from Win32_OperatingSystem\ ...
So the problem still persists and I will investigate further!
Boost 1.47 libs on Windows
Boost 1.48 has a very nice fix to remove that strange "Select LastBootUpTime from Win32_OperatingSystem" directory (https://svn.boost.org/trac/boost/ticket/5392) used by boost interprocess on Windows and 1.49 includes more bugfixes for interprocess (http://www.boost.org/users/history/version_1_49_0.html).
This does NOT solve #956, but as I said it has other fixes and at least the mentioned one that is beneficial for solving/debugging #956 further I think and yes I'm on the right track ^^.
I think this is missing the gitian changes to build/use the new version?
Sorry no, the commits are current and mergeable. What do you mean by gitian changes? A link or hint would be nice, still treat me like a newb :D, I'm learning everyday.
gitian is the deterministic build system we use, that sets up a virtual machine environment with known versions of EVERYTHING and then builds in such a way that bit-for-bit identical .exes come out of multiple developer's machines. It means anybody can verify that we are using the source code we're saying we are to build the executables that we distribute (except for Mac, everybody has to trust me because we can't gitian-build the Mac release yet).
Source code for the gitian-build process is in contrib/gitian-descriptors/ Windows builds use deps-win32.yml boost-win32.yml qt-win32.yml and gitian-win32.yml If you want builds to use a different version of something, you (or somebody) needs to change one or more of those files.
To summarize, the bitcoin-qt.pro file I edited is only of interest for people who compile it for themselves, like I currently do. So a change to that file is of no benefit for official releases, but at least helps others in compiling, right?
Closed, I'm using a modified version of this for my local version, seems unneeded here.