I am pleased to announce that Autowrap has now gained the ability to generate .NET interfaces for D libraries! This means that if you have a D library that you would like to call from .NET you can now Autowrap it and use the library in .NET as if it were any other .NET assembly.

Autowrap will generate interfaces for the following features of D:

Primitive Types that map between the two languages (no cent/ucent/real)
Module-level Functions
Structs and Classes
  - Constructors
  - Fields
  - Properties
  - Methods
1-Dimensional Ranges

Please note that this is initial work and there may be bugs and things that don't work. For example we know the following things will not work:

SysTime/DateTime/UUID types
Multi-Dimensional Ranges
Templates
Delegates/Callbacks

Amoung the items that do work, 1D ranges have not been thoroughly tested and there may be corner cases around class/reference lifetime. At the moment, we only ensure that a D object that is referenced in C# code is pinned in the D GC until the C# GC runs the objects destructors. Therefore, multiple copies of the same D reference may experience use-after-free as the memory will be unpinned after the first destructor is called. For most scenarios this will not be a problem, but it is something to be aware of. I am exploring alternatives to handle this problem but nothing has yet been decided and I am curious what your opinions and experience would suggest.

Autowrap/.NET requires .NET Core 2.1 or greater and C# 7.3 or later.

More information can be found here:
GitHub Repo: https://github.com/kaleidicassociates/autowrap
DUB Package: http://code.dlang.org/packages/autowrap

I would like to thank Symmetry Investments for sponsoring this effort.

-- 
Adam Wilson
IRC: EllipticBit
import quiet.dlang.dev;

On Thursday, 13 December 2018 at 11:24:05 UTC, Adam Wilson wrote:
> I am pleased to announce that Autowrap has now gained the ability to generate .NET interfaces for D libraries! This means that if you have a D library that you would like to call from .NET you can now Autowrap it and use the library in .NET as if it were any other .NET assembly.

I've never done much with .NET. Will this allow calling of D libraries from F#?

On 12/13/18 6:28 AM, bachmeier wrote:
> On Thursday, 13 December 2018 at 11:24:05 UTC, Adam Wilson wrote:
>> I am pleased to announce that Autowrap has now gained the ability to generate .NET interfaces for D libraries! This means that if you have a D library that you would like to call from .NET you can now Autowrap it and use the library in .NET as if it were any other .NET assembly.
> 
> I've never done much with .NET. Will this allow calling of D libraries from F#?

Yes. The generated C# can be built into a .NET Class Library and then added as a reference to the F# project.

-- 
Adam Wilson
IRC: EllipticBit
import quiet.dlang.dev;

Is there any overhead on the generated interface? Or overhead the compiler can't trivially optimise away.

Do you have any recocmendations about mixing coe like, don't use strings for now or try to minimize switching from D to C# and vice-versa.

Do you have plans to incorportae this as a VisualD project .csproj project since it's already intended to be microsoft oriented? (It would seem like a good fit to me.) That way you could even take away the mixin code and the "running the main method" code.

Does it work with LDC or only with DMD? How about GCC on linux?

On 12/14/18 2:33 PM, Sjoerd Nijboer wrote:
> Is there any overhead on the generated interface? Or overhead the compiler can't trivially optimise away.
> 

Yes, any overheads that would normally be associated with a P/Invoke call will be present here.

> Do you have any recocmendations about mixing coe like, don't use strings for now or try to minimize switching from D to C# and vice-versa.
> 

Strings are fine. But definitely try to minimize switching as there is some pretty significant overheads on the D side with module constructors. And "switching" can happen in some pretty well hidden places, for example, appending to a range is switch.

> Do you have plans to incorportae this as a VisualD project .csproj project since it's already intended to be microsoft oriented? (It would seem like a good fit to me.) That way you could even take away the mixin code and the "running the main method" code.
> 

I do not since I wouldn't know where to start, but it's possible.

> Does it work with LDC or only with DMD? How about GCC on linux?

It works with anything that can output a shared library and C interfaces. :)

-- 
Adam Wilson
IRC: EllipticBit
import quiet.dlang.dev;

>> Do you have plans to incorportae this as a VisualD project .csproj project since it's already intended to be microsoft oriented? (It would seem like a good fit to me.) That way you could even take away the mixin code and the "running the main method" code.
>> 
>
> I do not since I wouldn't know where to start, but it's possible.
>


This is an interesting ideal Visual D already provides a D injection in to C++ projects.

Using DLangInNet(I'm renamed your project for you ;) one should be able to add D code in to a C#, F#, C++, VB, or any .net compatible project.

You should talk to Rainer about this. It shouldn't be all that difficult to do since C#'s PInvoke does all the real work. I assume DLangInNet just generates a C# equivalent that forwards all the calls to the D code using Pinvoke when necessary?

So the idea is that one can add .d files to .Net projects, when built:

1. The D files are compiled.
2. DInNet runs on the D code and generates the C# output(could be modified for many other lannguages such as python, F#, Haskell, etc). Basically an autobinding generator from D to whatever, might be a good project to develop.
3. .Net handles the rest.

The point of having it this way in Visual Studio is that one can on a single project that has many different languages involved and can setup a rather seamless connection. This allows one to tailor the program design to the language of choice. E.g., D for performance, C++ for barebones, C# for gui, F# for structure, etc.

This could be used for Unity, say, for using D in algorithms. Such a feature would attract far more users and give D a boost in it's userbase and more will get done.

What we need is an idea that just works and one can use any appropriate source language at any time and they all bind without hicup's in most cases.

One can easily update all the compilers at a click and unwind if any compiler fails to build a project, etc. So much more could be done. But a start would be getting some of these different zones(e.g., languages, 3D engines, compilers, etc) to all work seamlessly in an integrated fashion.

E.g., something along the lines of:

1. Pick your language's
2. Pick your rendering engine
3. Pick your standard library
4. Pick your extensions


And so one can mix and match all things and everything just connects together. E.g., your python file will "intellisense" with your D file. Write a function in Haskell and it is callable in VB which can be called by your python file.

of course, this would suggest that there is just one underlying language and many "versions"(mappings) on top... which is basically IR with .NET.

I'm just thinking grander and not having to write a new compiler... just using bindings and wrapper generators to generate the code. Since most languages use object files and pretty much follow the same basic principles, it probably could be done for most things needed by most people.

One would just have to come up with the proper design so that people could build on it and connect languages in to the ecosystem easily.

This is the direction in which IDE's are headed but I don't think they realize it yet.

It seems it wouldn't be all that much work either. Essentially one just maintains a list of "exportable" symbols for each language source, project, module, etc and all other language sources can see them.

E.g., write a public global function or variable in D, say, and it is automatically exported which can then be seen in any other language source. Proper interaction will be chosen and everything will just work.

This would be like being able to speak 10 languages nearly simultaneously! With such an IDE it could even deal with Marshalling and seg faulting issues such as apply warnings when a variable should be free'd when it is interopted with and the IDE knows it should(e.g., D and C++ issues) or whatever. (People would be able to write such rules so that errors could be minimized)

On Sat, 15 Dec 2018 00:43:42 +0000, JJS@jjs.com wrote:
>>> Do you have plans to incorportae this as a VisualD project .csproj

Retaining the "On Sat, 15 Dec, Person A wrote:" lines is helpful for keeping track of the conversation.

> Using DLangInNet(I'm renamed your project for you ;)

Only the project maintainers have that authority. The project exposes D types and functions in Python, Excel, and .NET, so DLangInNet would be a terrible name for it.

> You should talk to Rainer about this. It shouldn't be all that difficult to do since C#'s PInvoke does all the real work. I assume DLangInNet just generates a C# equivalent that forwards all the calls to the D code using Pinvoke when necessary?

This is discussed in the Autowrap readme file.

> So the idea is that one can add .d files to .Net projects, when built:

Emitting the C# interface is a separate build target. You need to be able to specify Autowrap and the D compiler as dependencies. You need to hook this up into the build script.

> 2. DInNet runs on the D code and generates the C# output(could be modified for many other lannguages such as python, F#, Haskell, etc). Basically an autobinding generator from D to whatever, might be a good project to develop.

Perhaps a good name for that would be "Autowrap".

> The point of having it this way in Visual Studio is that one can on a single project that has many different languages involved and can setup a rather seamless connection.

Add a field to a wrapped D type and it won't show up on the C# side until you rebuild. This is not seamless. Making it seamless requires your IDE to know how the binding will be constructed and to have good support for every language you're using. Plus each language plugin needs to use a common set of data structures to represent your source code.

Technically, this isn't hard for a common set of languages, at least if you're only binding from a statically typed, strongly typed language with ideally little metaprogramming and in any case no ability to change types at runtime.

> What we need is an idea that just works and one can use any appropriate source language at any time and they all bind without hicup's in most cases.

Languages need a certain level of similarity for that to work. Haskell assumes immutability and would not work particularly well with a D function that mutates its input. Javascript objects can gain and lose fields and methods at runtime, something that D can't model well. A Python function can return a value of any type at all.