On Monday, 22 December 2014 at 23:14:44 UTC, Elie Morisse wrote:
>  • Structs
> https://github.com/Syniurge/Calypso/blob/master/tests/calypso/showcase.d

> testClass cls = new testInherit;

This should be written

testClass* cls = new testInherit;

In C++ struct and class differ only in symbol mangling. They are both value types, support inheritance, copy constructors etc. By looking at the declaration you can't decide if a class should be used as value type or reference type, so better treat it as value type like in C++.

> I have the pleasure to announce to you all the existence of a modified LDC able to interface directly to C++ libraries, wiping out the need to write bindings:

This is nothing short of amazing. I do not know much about streams, up and down. But getting the fantastic toy language that is D to seamlessly integrate C++ puts the whole thing to a new level. There is nothing that can compare.

On Tuesday, 23 December 2014 at 15:20:18 UTC, Kagamin wrote:
> On Monday, 22 December 2014 at 23:14:44 UTC, Elie Morisse wrote:
>> • Structs
>> https://github.com/Syniurge/Calypso/blob/master/tests/calypso/showcase.d
>
>> testClass cls = new testInherit;
>
> This should be written
>
> testClass* cls = new testInherit;
>
> In C++ struct and class differ only in symbol mangling. They are both value types, support inheritance, copy constructors etc. By looking at the declaration you can't decide if a class should be used as value type or reference type, so better treat it as value type like in C++.

At the moment Calypso maps C++ POD records to D structs, and non-POD ones to D classes. But then it's true that this creates a conundrum because I can't translate class value types to D types, so currently functions that expect class value arguments are ignored.

Making C++ classes value types would have lots of implications, so the way I see it (for now, not sure if that's the best solution) I'll probably stick to D class semantics but I'll have to create C++-specific class value types to map C++ fields or functions that take class value arguments.

On Tuesday, 23 December 2014 at 15:20:18 UTC, Kagamin wrote:
> On Monday, 22 December 2014 at 23:14:44 UTC, Elie Morisse wrote:
>> • Structs
>> https://github.com/Syniurge/Calypso/blob/master/tests/calypso/showcase.d
>
>> testClass cls = new testInherit;
>
> This should be written
>
> testClass* cls = new testInherit;
>
> In C++ struct and class differ only in symbol mangling. They are both value types, support inheritance, copy constructors etc. By looking at the declaration you can't decide if a class should be used as value type or reference type, so better treat it as value type like in C++.

Ah and you're right that this might confuse the user since he has to know whether the class/struct is POD or not to know whether to use values or references. Hmm.. also simply adding a virtual function to a C++ class would break the D code which previously considered that class POD :)

So maybe you're right, it might be a better idea to make C++ classes values.

Or it can switch depending on preprocessor definitions:
---
class WXDLLIMPEXP_BASE wxString
#ifdef wxNEEDS_WXSTRING_PRINTF_MIXIN
       : public wxStringPrintfMixin
#endif
{
---

On 23 Dec 2014 13:30, "CraigDillabaugh via Digitalmars-d-announce" < digitalmars-d-announce@puremagic.com> wrote:
>
> On Tuesday, 23 December 2014 at 11:53:38 UTC, Dicebot wrote:
>>
>> On Tuesday, 23 December 2014 at 10:52:58 UTC, Joseph Rushton Wakeling
wrote:
>>>
>>> clip
>>
>>
>> Consider both things like embedded/MIPS and Windows64 - LLVM tooling is
not as strong on those right now, GCC does not provide such easy way to reuse C++ frontend and with DMD/MSVC it is simply beyond feasibility.
>>
>> clip
>
>
> Regarding GCC C++ frontend, will this be of help (also recently posted on
this list):
>
> http://forum.dlang.org/thread/weodkqwxrqetvolhbghb@forum.dlang.org
>

I would say no.  The interface exposed by the jit frontend only supports a small subset of C-like code at this point in time.  And doesn't allow you to access other frontends.

Iain.

Could you explain how to build it's on Windows. I read readme, but do not fully understand what I should to to to get it's work?

On 12/22/2014 3:14 PM, Elie Morisse wrote:
> Hi everyone,
>
> I have the pleasure to announce to you all the existence of a modified LDC able
> to interface directly to C++ libraries, wiping out the need to write bindings:
>
>   https://github.com/Syniurge/Calypso

I think this is an exciting development!

Considering the new D support for C++ namespaces, template mangling, etc., I think it would make Calypso's job easier. The big challenge is to get an interface to C++ STL, so one could do:

  import core.stdcpp.stl.vector;

for example.

I'd also like to encourage you to submit a Dconf 2015 presentation proposal on this.

On Wednesday, 21 January 2015 at 10:37:18 UTC, Suliman wrote:
> Could you explain how to build it's on Windows. I read readme, but do not fully understand what I should to to to get it's work?

You should follow the guides for LDC only you need to build Clang 3.5 as well beforehand and add the -DLLVM_SOURCE_PATH="/path/to/llvm/source/tree" argument to the cmake command while building Calypso.

Using Visual Studio:
http://wiki.dlang.org/Building_and_hacking_LDC_on_Windows_using_MSVC

Using MinGW :
http://wiki.dlang.org/Building_LDC_on_MinGW_x86

Currently making D classes derive from C++ classes with a virtual table only works with MinGW because only the Itanium ABI is supported. I'm going to add the Microsoft ABI ASAP (it's just a few lines of code) but it'll be untested.

On Wednesday, 21 January 2015 at 21:27:27 UTC, Walter Bright wrote:
> I think this is an exciting development!
>
> Considering the new D support for C++ namespaces, template mangling, etc., I think it would make Calypso's job easier. The big challenge is to get an interface to C++ STL, so one could do:
>
>   import core.stdcpp.stl.vector;
>
> for example.
>
> I'd also like to encourage you to submit a Dconf 2015 presentation proposal on this.

Thanks, Walter!


A small update since the announcement: instantiation of C++ class templates is now working, with the right partial specialization selected if any.

I'm still working as fast as I can to get Ogre3D working. Ogre3D makes wide usage of the standard C++ library so getting it running would be a milestone and at that point most C++ libraries will be usable or in close reach as well. And it'd also make a sweet demo :)

On 1/21/2015 3:21 PM, Elie Morisse wrote:
> I'm still working as fast as I can to get Ogre3D working. Ogre3D makes wide
> usage of the standard C++ library so getting it running would be a milestone and
> at that point most C++ libraries will be usable or in close reach as well. And
> it'd also make a sweet demo :)

Just making STL work would be an achievement!

Is the output of Calypso a file that can be imported?