C++ Target about weave HOT 3 CLOSED

And another showstopper bug was laying in wait, preventing the obvious workaround of getting the impl field out of the union: nim-lang/Nim#13093

So if that isn't solvable upstream in a reasonable time we will have to:

• either only support C (and so no MSVC support)
• or disable dataflow parallelism/pledges in C++ mode. (code touched is in nim-lang/Nim#13093)

from weave.

Adding to the C++ woes syncScope does not work with the C++ backend if used in a for loop: nim-lang/Nim#14118

Unfortunately that's where it's the most useful, to sync all tasks before going to the next iteration (which often updates the same buffer in-place hence why synchronization is needed)

from weave.

The bad codegen is due to atomics being non-copyable and non-movable in C++

Unfortunately due to Nim's new case-object rules we have to copy/move them in the code here:

Following nim-lang/RFCs#209 and nim-lang/Nim#13995 this will be relaxed for --gc:arc however --gc:arc had several other issues related to threading at the moment and Weave base functionalities should work with most GC.

--> Avoid Nim case-objects and build our own.

from weave.