fix: remove malignant alignas

[wdconinc]

Removing this alignas override fixes https://github.com/JeffersonLab/JANA2/issues/238, and may resolve a number of mysterious issues in EICrecon over the past year:

• from boost small_vector requiring initialization before JANA2,
• to Acts crashing when support for avx/avx2 is compiled in, which we narrowed down to https://github.com/paulgessinger/acts/blob/main/Core/include/Acts/Utilities/Any.hpp#L490-L503 and even discussed with Acts developers,
• to Eigen asserts failing with pointers to http://eigen.tuxfamily.org/dox-devel/group__TopicUnalignedArrayAssert.html,
• or just plain segfaults.

I can confirm that removing the alignas override fixes one reproducible Eigen assert failure in a local copy. Removing alignas is also confirmed to avoids our need to include boost/container/small_vector.hpp before including JANA2 headers. (We fixed the avx/avx2 issue with Acts already by increasing separation between JANA2 factories and tracking algorithms.)

[veprbl]

This should be good - alignas is part of the standard https://en.cppreference.com/w/cpp/language/alignas now

[faustus123]

This should be good - alignas is part of the standard https://en.cppreference.com/w/cpp/language/alignas now

I saw this too, but will admit to being unclear why the compiler is seeing it as undefined.

Regardless, This was all put in 6 years ago and I'm sure seemed like a good idea at the time. It needs to be removed. I would like to see the Get_CacheLineSize removed too unless someone can make an argument to keep it. Maybe @nathanwbrei can comment.

[wdconinc]

I saw this too, but will admit to being unclear why the compiler is seeing it as undefined.

Because it's not a macro. It's a keyword.

[wdconinc]

And even the fact that it's a keyword isn't important. At this point, only the preprocessor is active. It doesn't even know yet which language this is in. All it knows is that alignas is not a macro.

[faustus123]

OK. So was it a macro at one point? How could it ever have worked then?

[wdconinc]

Maybe it never worked. It just disables alignas everywhere. When you are looking at an issue on Mac (old clang I assume), then that issue was resolved by this. It just also broke alignas everywhere else. Just took until now to have someone realize it.

[nathanwbrei]

Regardless, This was all put in 6 years ago and I'm sure seemed like a good idea at the time. It needs to be removed. I would like to see the Get_CacheLineSize removed too unless someone can make an argument to keep it. Maybe @nathanwbrei can comment.

Let's get rid of it! The code is outright incorrect, and manages to break an impressive number of things. If we discover we still need it, we should reimplement it as something along the lines of:

#ifdef __APPLE__
#define JANA2_ALIGNAS 
#else
#define JANA2_ALIGNAS alignas
#endif

Also, we should make the cache line size be a parameter we can set externally in one place, e.g. JANA2_CACHE_LINE_BYTES. Currently, we have Get_CacheLineSize(), CACHE_LINE_BYTES, and one hardcoded 64.