What is the reason for the assertions in system/Platform.h ?

[Frank Drop]

In blaze/system/Platform.h I find the following three assertions:

BLAZE_STATIC_ASSERT( !BLAZE_WIN64_PLATFORM || ( !BLAZE_MINGW32_PLATFORM && !BLAZE_MINGW64_PLATFORM ) );
BLAZE_STATIC_ASSERT( !BLAZE_MINGW32_PLATFORM || ( !BLAZE_WIN64_PLATFORM && !BLAZE_MINGW64_PLATFORM ) );
BLAZE_STATIC_ASSERT( !BLAZE_MINGW64_PLATFORM || ( !BLAZE_WIN64_PLATFORM && !BLAZE_MINGW32_PLATFORM ) );

When compiling a project involving blaze with MSYS2 on MinGW64, these assertions fail, because all three defines are true. When out commenting these three assertions, the program compiles fine and is fully functional. I was wondering why these checks are implemented.

[Klaus Iglberger]

Hi Frank!

These static assertions document our expectations on the platforms. Our current expectations are that only one of these three settings can be true at any time. Apparently you have found a system that is both a 32-bit and a 64-bit system, which sounds strange to us (at least unexpected).

Please feel free to provide a pull request to enable the use of MYS2. The above settings are used for several system depending implementations: <math/Infinity.h> and <util/Limits.h>(both because ofsize_tandptrdiff_t) and<util/Memory.h>``` (for the platform-specific allocation of aligned memory). All of these need to be tested on all involved platforms (i.e. MinGW, Win64 and MSYS2) before a pull request can be accepted. Thanks,

Best regards,

Klaus!

[Klaus Iglberger]

Hi Frank!

I assume the question has been answered. If you still decide to provide a pull request, please refer back to this issue. Thanks,

Best regards,

Klaus!

[Thomas Castro]

Hi,

I was compiling a project and encountered the same assertions failing. I investigated a little about how the Pre-defined compiler macros work (here for compiler especific macros/ here for OS specific macros) and encountered that “__MINGW32__“ is defined whether I’m using Mingw32 of Mingw64 (32bit or 64bit). Also as far as I know MSYS2 uses Mingw64, In my case. I’m using niXman posix binaries. With that knowledge these assertions will always fail because all are set to 1 if I am using Mingw64.

I have checked the headers you mentioned that uses the *_PLATFORM macros (Infinity.h, Limits.h and Memory.h).

<util/Memory.h>:

I think the intention with the macros on the use of _aligned_malloc() and __mingw_aligned_malloc() was to identify if the user was using MSVC, MinGW-w64 or MinGW-w32. The only issue that I see is that it first checks if WIN32 set to true then uses _aligned_malloc(). However it does not ensure the user is not using MinGW-w32 as it could be possible both were defined.

Maybe in this case it is safer to have a macro indicating it is using MSVC compiler replacing the windows macros.

<math/Infinity.h> and <util/Limits.h>:

It just checks if I am at windows so I think this one is safe if the assertions changes. For clarity I checked that ptrdiff_t and size_t have the same sizes on both MSVC and MinGW-64 where the platform is 32 or 64.

I’m new on this platform so I apologize if I was not clear or any mistake on my side.

Best regards,

Tom.