Related: https://github.com/dlang/dmd/pull/6334

dmd has converted 100% of the glue code to D. It no longer has any use for the .h files in the front end. In other words, the .h files exist only to support GDC and LDC.

What are the prospects for reducing the dependence of GDC/LDC on those .h files? I figure it is impractical to convert gcc/llvm header files to D in order to convert all the glue code to D, but could the glue code be split up into D and C++ sources so that the C++ code will no longer need those headers?

On Monday, 19 December 2016 at 09:43:42 UTC, Walter Bright wrote:
> Related: https://github.com/dlang/dmd/pull/6334
>
> dmd has converted 100% of the glue code to D. It no longer has any use for the .h files in the front end. In other words, the .h files exist only to support GDC and LDC.
>
> What are the prospects for reducing the dependence of GDC/LDC on those .h files? I figure it is impractical to convert gcc/llvm header files to D in order to convert all the glue code to D,

For LDC: absolutely impractical if not utopian.

> but could the glue code be split up into D and C++ sources so that the C++ code will no longer need those headers?

For LDC again: not anytime soon, that'd be a *huge* effort.