win32k public headers

[Timo Kreuzer]

I am currently trying to clean up our win32k headers.

The situation so far:
- ntgdi.h in psdk which should be compatible to ms version of ntgdi.h 
from the sdk
- ntgdytyp.h containing a bunch of gdi related types in 
include/reactos/win32k
- ntgdihdl.h containing handle related types and macros in 
include/reactos/win32k
- ntgdibad.h containing private and ms incompatible functions and 
structures in include/reactos/win32k
- ntuser.h containing ntuser functions and types in include/reactos/win32k
- ntusertyp.h containing ntuser related types in include/reactos/win32k
- W32CLIENTINFO is defined in both ntusertyp.h and ntgdityp.h (less 
complete).
- there is the PATRECT structure in ntuser.h, and if ntuser.h is not 
included, also in ntgdityp.h, this should be named POLYPATBLT as defined 
in ntgdi.h, but there is only an incomplete type without the structure 
definition. (IMO ms should have put this structure into wingdi.h 
together with PolyPatBlt, but they don't seem to want anyone use that 
gdi32 exported function.)

I wonder why ms put ntgdi.h into the sdk at all, as it's completely 
useless to user mode application writers, but well, ncie to have the 
prototypes ;-)

So how do we clean up this mess?

Suggestions:
- when ms puts ntgdi.h into the sdk, we could also put ntuser.h there.
- if we want to stay 100% compatible with sdk headers, we can add a 
define, like
#ifdef USE_NON_SDK
// put non sdk compatible but still windows compatible definitions here
#endif
- use a version definition, like I did with the simplecall definitions 
in ntuser.h to make it compatible to different windows versions
- keep ntgdibad.h and create ntuserbad.h for incompatible definitions
- ntgdihdl.h is used by win32k and gdi32, both also import ntgdityp.h. 
So I'd suggest merging that stuff into ntgdityp.h
- maybe merge types into ntgdi.h and ntuser.h? Some definitions are 
already in there.
- create win32ktypes.h and put ntuser/ntgdi shared stuff there, like 
W32CLIENTINFO
- maybe add POLYPATBLT and PolyPatBlt() (others?) to wingdi.h inside 
USE_NON_SDK ifdef plus a POLYPATBLT_DEFINED

I don't think it's a problem to make our sdk headers more complete than 
ms ones. This might help if we don't want a header mess or extra 
definitions inside the .c files.

Comments appreciated.

Thanks,
Timo

[James Tabor]

Hi!

On Wed, Apr 2, 2008 at 11:21 AM, Timo Kreuzer <timo.kreuzer(a)web.de> wrote:
>   I am currently trying to clean up our win32k headers.
>
>   The situation so far:
>   - ntgdi.h in psdk which should be compatible to ms version of ntgdi.h
>   from the sdk 
Yes! Should always be!!

>    - ntgdytyp.h containing a bunch of gdi related types
> in
>   include/reactos/win32k
>   - ntgdihdl.h containing handle related types and macros in
>   include/reactos/win32k
>   - ntgdibad.h containing private and ms incompatible functions and
>   structures in include/reactos/win32k 
>    - ntuser.h containing ntuser functions and types in
> include/reactos/win32k
>   - ntusertyp.h containing ntuser related types in include/reactos/win32k
>   - W32CLIENTINFO is defined in both ntusertyp.h and ntgdityp.h (less
>   complete). 
Gdi and User point to the ClientInfo area which is part of thread info
structures and like all thread structures should be place in some sort
of include/reactos/win32k/w32thread.h. This would allow Gdi and User
to access them both. I have placed #ifndef _NTUSRTYP_ so when both
headers are used there will be no compile errors.

>    - there is the PATRECT structure in ntuser.h, and if
> ntuser.h is not
>   included, also in ntgdityp.h, this should be named POLYPATBLT as defined
>   in ntgdi.h, but there is only an incomplete type without the structure
>   definition. (IMO ms should have put this structure into wingdi.h
>   together with PolyPatBlt, but they don't seem to want anyone use that
>   gdi32 exported function.)
>
>   I wonder why ms put ntgdi.h into the sdk at all, as it's completely
>   useless to user mode application writers, but well, ncie to have the
>   prototypes ;-)
>
>   So how do we clean up this mess?
>
>   Suggestions:
>   - when ms puts ntgdi.h into the sdk, we could also put ntuser.h there.
>   - if we want to stay 100% compatible with sdk headers, we can add a
>   define, like
>   #ifdef USE_NON_SDK
>   // put non sdk compatible but still windows compatible definitions here
>   #endif 
If we are compatible why do we need this? All SDK headers should be
compatible and I have pointed this out. When NDK/DDK (WDK) or PSDK
(SDK) changes are made the developer should have access to these and
compare the changes to verify them as correct. I have made changes to
the NDK with some Rtl functions and I had to make logical choices and
assumptions. Since those Rtl functions are unknown and the research
was from Wine, I did go through the steps of verifying and cross
referencing them before I moved them into NDK.

SDK and WDK are open to the public and there is nothing preventing
that developer from referencing them for research and standards.

One of my best patent guys always said,"Once someone publishes their
patent in text book form, they just gave it away." If your lawyer
doesn't know that, go find one that does!

>    - use a version definition, like I did with the
> simplecall definitions
>   in ntuser.h to make it compatible to different windows versions
>   - keep ntgdibad.h and create ntuserbad.h for incompatible definitions
>   - ntgdihdl.h is used by win32k and gdi32, both also import ntgdityp.h.
>   So I'd suggest merging that stuff into ntgdityp.h
>   - maybe merge types into ntgdi.h and ntuser.h? Some definitions are
>   already in there. 
>    - create win32ktypes.h and put ntuser/ntgdi shared
> stuff there, like
>   W32CLIENTINFO 
This will due too!

>    - maybe add POLYPATBLT and PolyPatBlt() (others?) to
> wingdi.h inside
>   USE_NON_SDK ifdef plus a POLYPATBLT_DEFINED
>
>   I don't think it's a problem to make our sdk headers more complete than
>   ms ones. This might help if we don't want a header mess or extra
>   definitions inside the .c files.
>
>   Comments appreciated.
>
>   Thanks,
>   Timo
>
>  
Thanks,
James

[Timo Kreuzer]

James Tabor schrieb:
>  
>  Gdi and User point to the ClientInfo area which is part of thread info
>  structures and like all thread structures should be place in some sort
>  of include/reactos/win32k/w32thread.h. This would allow Gdi and User
>  to access them both. I have placed #ifndef _NTUSRTYP_ so when both
>  headers are used there will be no compile errors.
>     
Placing an undocumented and partly unknown structure inside two 
different headers will lead to problems, like the two different versions 
we have now ;-)
> >    Suggestions:
> >   - when ms puts ntgdi.h into the sdk, we could also put ntuser.h there.
> >   - if we want to stay 100% compatible with sdk headers, we can add a
> >   define, like
> >   #ifdef USE_NON_SDK
> >   // put non sdk compatible but still windows compatible definitions here
> >   #endif
> >       
>  If we are compatible why do we need this? All SDK headers should be
>  compatible and I have pointed this out. When NDK/DDK (WDK) or PSDK
>  (SDK) changes are made the developer should have access to these and
>  compare the changes to verify them as correct. I have made changes to
>  the NDK with some Rtl functions and I had to make logical choices and
>  assumptions. Since those Rtl functions are unknown and the research
>  was from Wine, I did go through the steps of verifying and cross
>  referencing them before I moved them into NDK.
>
>  SDK and WDK are open to the public and there is nothing preventing
>  that developer from referencing them for research and standards.
>     
Well I'm not sure if I got you right, so I just try to make my thinking 
more clear:
There's compatible and there's compatible:
A) our sdk headers will contain 100% the same as ms sdk headers, not 
less, not more
B) our sdk headers will contain everything that is in ms sdk headers, 
but can contain more.
The ms sdk you can download is the public sdk. It's probably not what ms 
uses to compile their stuff. It's partly limited. ntgdi.h contains 
"typedef struct _POLYPATBLT  POLYPATBLT, * PPOLYPATBLT";
But where's the real structure definition? Somewhere else? Does that 
make sense? Well it does when ms uses different headers that contain all 
the complete types, but then they don't even need this typedef.. So 
where do we put it?
I suggested the #ifdef's to have compatibility A when it's not defined 
and B when it's defined.
ntuser.h doesn't belong to the ms sdk. But for what reason? Do ms people 
really have ntuser stuff somewhere else? Probably not, they just decided 
not to ship it (I think they shipped ntuser.h with DDK some time ago)

Timo

[James Tabor]

Hi!

2008/4/2 Timo Kreuzer <timo.kreuzer(a)web.de>de>:
>  
>   Placing an undocumented and partly unknown structure inside two different
>  headers will lead to problems, like the two different versions we have now
>  ;-)
>
>  
Sorry~

>    Well I'm not sure if I got you right, so I just
> try to make my thinking
>  more clear:
>   There's compatible and there's compatible:
>   A) our sdk headers will contain 100% the same as ms sdk headers, not less,
>  not more
>   B) our sdk headers will contain everything that is in ms sdk headers, but
>  can contain more.
>   The ms sdk you can download is the public sdk. It's probably not what ms
>  uses to compile their stuff. It's partly limited. ntgdi.h contains "typedef
>  struct _POLYPATBLT  POLYPATBLT, * PPOLYPATBLT"; 
BTW the structure is this:
typedef struct _POLYPATBLT
{
  LONG nXLeft;
  LONG nYLeft;
  LONG nWidth;
  LONG nHeight;
  HBRUSH hBrush;
} POLYPATBLT, * PPOLYPATBLT;

I was going to change it with the next installment of GdiBatch.

>    But where's the real structure definition?
> Somewhere else? Does that make
>  sense? Well it does when ms uses different headers that contain all the
>  complete types, but then they don't even need this typedef.. So where do we
>  put it?
>   I suggested the #ifdef's to have compatibility A when it's not defined and
>  B when it's defined.
>   ntuser.h doesn't belong to the ms sdk. But for what reason? Do ms people
>  really have ntuser stuff somewhere else? Probably not, they just decided not
>  to ship it (I think they shipped ntuser.h with DDK some time ago) 
I guess we can
add those into the our SDK. I was ranting on how we
should be compatible.

>  
>   Timo 
Do what you think is right....
Thanks,
James