We are forking D

[Timon Gehr]

On 1/11/24 18:40, a11e99z wrote:
> what should be done with D?
> 
> 1) /for any future path of D/
> drop DMD entirely and develop only LDC.

DMD is much easier to build so as things are it is worth keeping even if 
only for the sake of compiler development.

> ...
> - minus: to slow code generation.

It's a big one.

> well, Zig decides to generate 
> LLVM-binary-IR w/o LLVM framework/tools. and need to optimize compiler 
> itself not generated code.
> ...

That seems potentially interesting, though an additional step to 
transform the generated LLVM IR into machine code is certainly more 
expensive than outputting machine code directly. I am sure the Zig 
developers must be aware of this and are planning to do their own native 
backends for common architectures as well?

Though I hear that LDC spends a lot of time on IR validation to detect 
bugs by default and that disabling it using --disable-verify can 
increase LDC build speeds significantly.