Valery Mironov

hi, I read msvc 2013 log, but I can't understand what went wrong

I mean, it would be convenient if boost had the cross-platform function timegm (http://man7.org/linux/man-pages/man3/timegm.3.html, https://docs.microsoft.com/en-us/cpp/c-runtime-library/reference/mkgmtime-mkgmtime32-mkgmtime64?view=vs-2019). Especially if it is used in the boost https://github.com/boostorg/chrono/blob/develop/include/boost/chrono/io/time_point_io.hpp#L1166 One of my colleagues made a...

`boost::posix_time::to_time_t(boost::posix_time::ptime_from_tm(tm_time))` Sorry, for example, such code works correctly, but in general it may be worthwhile to achieve some kind of consistency

@ashvardanian > what do you think about fp8? I didn't see fp8, somewhere except faiss maybe. But ofc more supported types is good. > Are you looking to use SimSIMD...

> > 1. here should be `release` instead of `relaxed`: > > https://github.com/openssl/openssl/blob/6afaa3f41f5b65432b6700064b077032b9e0c625/include/internal/refcount.h#L51 > > > > Example of issue: [ClickHouse@10e427e](https://github.com/ClickHouse/openssl/commit/10e427ee32cf74bc2d1e4945c57e3d518c737e12) > > But tsan [doesn't work with such fences](https://github.com/google/sanitizers/issues/1415),...

In any case, I don't want to convince anyone that I'm right. Personally, I use the patch for these lines because it fixes tsan's errors. I just thought it was...

> What would be the semantics otherwise? 1) one thread: create object -- rc1 copy object -- rc2 start other thread 2) move copy of object to other thread 3)...

> The problem is these tsan errors might indicate a bug in how Poco uses OpenSSL. Ok whatever. Let's believe that everyone (rust std lib, boost, llvm libc++, etc) wrong...

@nhorman ### About memory_orders: > The fundamental issue is, when can there be a parallel increment and decrement when a refcount on an object is at 1. The answer is...

@kroeckx > T1: decrement counter atomic this should happens before this > T2: call free, writing to the location of the counter, non-atomic, and 8 bytes instead of 4. Otherwise...