I recently attempted to use the -HC generated dmd/frontend.h to build LDC and discovered it doesn't work:

So far I have fixed:

• redefinition of size_t
• appending of _ to C keywords (to the disgruntlement of Iain), so that the tokens enum work with clang

Still to do:

• generate #include <stdarg.h> when va_list is used
• general issues of declaration order and forward declarations: some resolution is too eager resulting in subclasses being emitted before their base classes. other is not eager enough resulting in typedefs of templates not forward declaring their arguments ending up with typedef Array<BaseClass* > BaseClasses not referring to a valid (i.e. prior) definition of struct BaseClass`.

I'm currently in the weeds of resolving the order of declaration issues, but it is rather annoying and tedious, given DMD does not build on macOS. But before I lose the rest of my marbles trying to do that I thought I put out an invitation/request for help for anyone interested in learning a bit of DMD, given the success I had last time. Thanks already to Rikki for the header generation command!

I can provide an overview of what needs to be done along with a guided tour of the relevant bits of DMD and pull request reviews and so on.

Hit me up on the Discord if you are interested.

Thanks in advance,
Nic

I think one big problem is that a truly usable frontend.h would have to accomodate for all platforms. So all version(XY) stuff (Windows, Posix, …), 32/64-bit differences etc. would have to be accounted for.

On Thursday, 12 October 2023 at 14:34:01 UTC, kinke wrote:

Not really, you'd generate it "on demand" with the D compiler you are using to build LDC/GDC.

On Thursday, 12 October 2023 at 22:56:20 UTC, Nicholas Wilson wrote:

With all its bugs in generating C++ headers that don't exist in the latest version you're trying to compile.

Reducing what gets exposed to C++ interface would get you further.