On Wednesday, January 25, 2012 20:38:01 Don wrote:
> Please note that the Win32 API project on dsource was created with the explicit intention of including it in Phobos. See bug 317.

Good to know. I would point out though that for the most part, std.c has been being phased out with all of its parts being moved to Phobos, so I'm not sure that putting the Win32 API bindings in std.c.windows makes sense.

Regardless, my point is that we need to figure out what we want to do at this point. As it stands, we end up with pull requests such as

https://github.com/D-Programming-Language/druntime/pull/139

which just add more bindings which a particular programmer happens to want. And while that might be okay in the short term, that doesn't work in the long term. We need to decide where we want the Win32 API bindings, how we want them organized more or less, and start putting them there. The Win32 API project is a great start, but it's been around for years without being moved into druntime or Phobos, and we should start looking at moving those bindings into either druntime or Phobos. At this point, given that they're OS bindings, and the OS bindings have been going in druntime, I would expect them to go into druntime. But a decision must be made, or the situation won't improve.

- Jonathan M Davis

On 2012-01-26 00:44, Jonathan M Davis wrote:
> On Thursday, January 26, 2012 00:20:52 Jesse Phillips wrote:
>> On Wednesday, 25 January 2012 at 21:44:21 UTC, Vladimir Panteleev
>>
>> wrote:
>>>> Also what is there currently isn't generating documentation.
>>>
>>> This is not true: http://dlang.org/phobos/core_thread.html
>>
>> Yes it is, there aren't doc comments on anything:
>>
>> https://github.com/D-Programming-Language/druntime/blob/master/src/core/sys/
>> windows/windows.d
>
> Pretty much none of the C bindings in druntime generate documentation - I
> believe primarily because Sean doesn't want to redo C documentation. However,
> that doesn't preclude giving them empty documentation so that you can see
> what's there, and that hasn't been done.
>
> - Jonathan M Davis

That would be good to have. A link to the C documentation would hurt either.

-- 
/Jacob Carlborg

On 2012-01-26 00:50, Jonathan M Davis wrote:
> On Wednesday, January 25, 2012 20:38:01 Don wrote:
>> Please note that the Win32 API project on dsource was created with the
>> explicit intention of including it in Phobos. See bug 317.
>
> Good to know. I would point out though that for the most part, std.c has been
> being phased out with all of its parts being moved to Phobos, so I'm not sure
> that putting the Win32 API bindings in std.c.windows makes sense.
>
> Regardless, my point is that we need to figure out what we want to do at this
> point. As it stands, we end up with pull requests such as
>
> https://github.com/D-Programming-Language/druntime/pull/139
>
> which just add more bindings which a particular programmer happens to want.
> And while that might be okay in the short term, that doesn't work in the long
> term. We need to decide where we want the Win32 API bindings, how we want them
> organized more or less, and start putting them there. The Win32 API project is
> a great start, but it's been around for years without being moved into
> druntime or Phobos, and we should start looking at moving those bindings into
> either druntime or Phobos. At this point, given that they're OS bindings, and
> the OS bindings have been going in druntime, I would expect them to go into
> druntime. But a decision must be made, or the situation won't improve.
>
> - Jonathan M Davis

It depends. If we want to keep druntime small and only contain what's necessary for the runtime it would be better to add the bindings to Phobos. But that would mean to duplicate parts of the bindings.

-- 
/Jacob Carlborg

On Wednesday, 25 January 2012 at 19:27:47 UTC, Vladimir Panteleev wrote:
> I support putting all public domain modules from the Win32 bindings project into Druntime.
>
> The entirety of the Windows API modules (headers in C) and accompanying import libraries are an integral part of The Windows SDK.

Public domain? PSDK is not public domain. Isn't "Microsoft & public domain" an oxymoron?

On Jan 25, 2012, at 11:27 AM, Vladimir Panteleev wrote:

> On Wednesday, 25 January 2012 at 19:16:34 UTC, Jonathan M Davis wrote:
>> The level of support for the Windows API in druntime and Phobos is pretty low. As I understand it, Windows users are pretty much forced to use http://www.dsource.org/projects/bindings/browser/trunk/win32 if they need comprehensive Win32 API bindings. druntime seems to be trying to define all of the OS-specific stuff like that, but on top of it missing much of it, in the case of the Win32 API, that's a _lot_ of functions, and I don't know if we want to put that much in druntime. So, the question is, how do we want to support the Win32 API in druntime and Phobos?
>> 
>> Do we want to put all of the Win32 API bindings in druntime? If not, then do we want to put them in Phobos? Or do we just want to send Windows developers to a 3rd party library like the Win32 bindings project on dsource? Given that they're OS bindings, I would _think_ that we'd want them in druntime, but I don't know.
> 
> I support putting all public domain modules from the Win32 bindings project into Druntime.
> 
> The entirety of the Windows API modules (headers in C) and accompanying import libraries are an integral part of The Windows SDK. There is no obvious finer-grained separation of modules, so including only a part of them would need to be done on mostly subjective decisions.
> 
> The only non-public-domain modules are the DirectX headers (which don't belong in Druntime either way), and WinSock modules, which are "Portions Copyright (c) 1980, 1983, 1988, 1993 The Regents of the University of California.  All rights reserved." and "Portions Copyright (c) 1993 by Digital Equipment Corporation."

For what it's worth, I tried integrating these headers ages ago (into Tango I believe) and stopped mostly because they were broken at the time.  I imagine they're sorted by now, and so long as there's little maintenance effort on my part and there are no license concerns I'm all for it.  The only other thing to be aware of is that each module linked into an executable incurs some overhead (500 bytes minimum, if I remember correctly), so moving to these new Win32 headers will increase the footprint of D apps on Windows.  I imagine it's a worthwhile tradeoff, but I did want to bring this up so no one is surprised later.

On Jan 25, 2012, at 11:58 AM, Jesse Phillips wrote:

> On Wednesday, 25 January 2012 at 19:16:34 UTC, Jonathan M Davis wrote:
>> The level of support for the Windows API in druntime and Phobos is pretty low. As I understand it, Windows users are pretty much forced to use http://www.dsource.org/projects/bindings/browser/trunk/win32 if they need comprehensive Win32 API bindings. druntime seems to be trying to define all of the OS-specific stuff like that, but on top of it missing much of it, in the case of the Win32 API, that's a _lot_ of functions, and I don't know if we want to put that much in druntime. So, the question is, how do we want to support the Win32 API in druntime and Phobos?
> 
> It seems to me that calling OS functions are expected to work out of the box. This would preclude requiring a third party library.
> 
> I don't think it makes sense in Druntime. It isn't required for the compiler to work that I know of. Maybe their are some Windows functions used within druntime? Also what is there currently isn't generating documentation.

System headers are absolutely required for Druntime to build.  This is one of the reasons core.stdc and core.sys exist.  Documentation isn't generated because these are interfaces to external APIs, so the documentation for those APIs should be referenced when coding to them.  We could add an empty doc comment to every declaration so something was generated in the Druntime docs, but I don't know why anyone would ever actually refer to them.

On Jan 25, 2012, at 1:44 PM, Vladimir Panteleev wrote:

> On Wednesday, 25 January 2012 at 19:58:40 UTC, Jesse Phillips wrote:
>> I don't think it makes sense in Druntime. It isn't required for the compiler to work that I know of.
> 
> Underlying implementation of language features trickles down to using either the DMC runtime or Windows APIs. Druntime defines the Windows APIs locally where needed, which is a hack.
> 
> Example:
> 
> https://github.com/D-Programming-Language/druntime/blob/master/src/rt/dmain2.d#L31

This was done to reduce coupling between modules and to reduce the runtime footprint in executables.  It's not terribly maintenance-friendly however, and so Druntime is slowly transition to simply importing the appropriate module.  A number of these remain mostly because it ain't broke as-is.

On Friday, January 27, 2012 10:39:41 Sean Kelly wrote:
> System headers are absolutely required for Druntime to build. This is one of the reasons core.stdc and core.sys exist. Documentation isn't generated because these are interfaces to external APIs, so the documentation for those APIs should be referenced when coding to them. We could add an empty doc comment to every declaration so something was generated in the Druntime docs, but I don't know why anyone would ever actually refer to them.

The value in having empty ddoc comments is so that if you need to use any of the functions in them for some reason (which _does_ happen upon occasion), then you know what module has what without having to go find the source code and look through it. So, ideally, all of the declarations in druntime for the C APIs should at least have empty ddoc so that they're documented. I completely agree that it's best not to try and duplicate the full documentation for them - that's best left to the C docs - but documenting what's there is still valuable, and we should probably do that at some point.

- Jonathan M Davis

On Friday, 27 January 2012 at 18:16:38 UTC, Kagamin wrote:
> On Wednesday, 25 January 2012 at 19:27:47 UTC, Vladimir Panteleev wrote:
>> I support putting all public domain modules from the Win32 bindings project into Druntime.
>>
>> The entirety of the Windows API modules (headers in C) and accompanying import libraries are an integral part of The Windows SDK.
>
> Public domain? PSDK is not public domain. Isn't "Microsoft & public domain" an oxymoron?

The files in the Windows API bindings project are translations from the MinGW headers, which were created based on the documentation (which explains the occasional errors). The MinGW headers are public domain. Considering that MinGW is not an obscure project, I think we're on safe legal ground here.

On Friday, 27 January 2012 at 18:36:28 UTC, Sean Kelly wrote:
> The only other thing to be aware of is that each module linked into an executable incurs some overhead (500 bytes minimum, if I remember correctly), so moving to these new Win32 headers will increase the footprint of D apps on Windows.  I imagine it's a worthwhile tradeoff, but I did want to bring this up so no one is surprised later.

Would it be possible to fix this overhead in the compiler (avoiding generation of such overhead for modules without classes, static ctors/dtors, etc.)?

Even though Windows API programs classically begin with #include <windows.h>, which includes most of the other headers, this isn't a requirement, and it's possible to import modules on a finer-grained basis.