Feedback on Átila's Vision for D

[Gregor Mückl]

On Thursday, 17 October 2019 at 13:16:07 UTC, Petar Kirov 
[ZombineDev] wrote:
> The last line of work appears to be in the clang-9 branch: 
> https://github.com/Syniurge/Calypso/commits/clang-9
>
> W.r.t to C++ interop there are 5 areas of development:
> 1. During the past 3 years there have been a ton of fixes to 
> extern (C++) (and to a lesser extent extern (C), which was 
> already close to rock-solid) and nowadays it should be pretty 
> usable (namespaces, classes and templates work in many cases). 
> You can directly call much of D from C++ and vice versa. 
> Obviously there are many D and C++ specific language features 
> that are not yet/can't be supported, but we're pretty further 
> along than what we had a couple of years ago. Having all 3 
> major compilers (first DMD, then LDC and now GDC) be written 
> half in D and half in C++ is a testament to that. (Actually now 
> DMD is D-only now, but it was in C++ + D mode for enough time 
> to force many of the developers at the time to at least make 
> the subset of extern (C++) that it used fully working.)
>
> 2. During the past year some parts of STL had bindings written 
> to them and tested for Linux, Mac and Windows (which is 
> surprisingly difficult, due to difference in C++ compilers and 
> STL implementations): 
> https://github.com/dlang/druntime/tree/master/src/core/stdcpp
>
> 3. Having tools like dstep and dpp support C++. As far as I 
> know both tools support only C header files, but at least for 
> dpp, Atila has listed C++ as a major planned feature. Work in 
> area 1. makes 3. now possible. The work of one of GSOC students 
> made a huge impact for dstep, so it's likely that a similar 
> tactic for dpp would also help.
>
> 4. Automatic C/C++ header generation for D code. There has been 
> solid progress, but we're probably a couple of months before 
> having the work finally merged upstream.
>
> 5. Making import C++ automagically work has been the holy grail 
> for many. The author of Calypso has had the unfortunate task of 
> having to single-handedly work on not only implementation, but 
> also solving many language-level challenges like template 
> instantiation, overloading and many more. Adding insult to 
> injury, the most efficient way to interface with libClang is 
> with C++ and for they don't guarantee a stable API, and even 
> break it very often, and D's frontend is likely worse then that 
> (at minimum it switched from being C++ only, to D only during 
> the time Calypso was developed.).
> I think the best way forward is to get it merged in LDC as soon 
> as possible - that way at least the burden of supporting new D 
> frontend and LLVM/clang backend versions could be shared by 
> other contributors.

Thank you for the detailed progress report. However, I'm afraid 
that this doesn't answer the one central question that a 
potential user would have: which version do I use and where/how 
do I get it?

Calypso has been mentioned as a possible path to get easy C++ 
interop here several times. But the current appearance of the 
github project does not instill any confidence in people who 
don't feel especially adventurous.

I'm amazed at the determination that it takes to create something 
as complex as Calypso. I'd very much like it to succeed.

Semi-OT, I would also like to see the automatic C++ header 
generation integrated into DMD/LDC/GDC. This would complement 
C++-to-D interop nicely. There's a pull request for this feature 
on github that is slowly getting stale. I don't know if it's 
merely being ignored or whether there are remaining issues with 
it.