Re: [ros-svn] [ekohl] 16283: Don't redefine ANSI_STRING, UNICODE_STRING and OBJECT_ATTRIBUTES if the NDK already defines them.

[Alex Ionescu]

[e-mail] wrote:
> Don't redefine ANSI_STRING, UNICODE_STRING and OBJECT_ATTRIBUTES if the NDK already defines them.
>
> Updated files:
> trunk/reactos/w32api/include/ntsecapi.h

Hi Eric,

Unforunately, we don't have the liberty to do this, since the real PSDK 
file doesn't, and MS Header compatibility is a goal. The NDK must always 
follow ntsecapi.h.

Best regards,
Alex Ionescu

[Eric Kohl]

Alex Ionescu wrote:
> Hi Eric,
>
> Unforunately, we don't have the liberty to do this, since the real PSDK 
> file doesn't, and MS Header compatibility is a goal. The NDK must always 
> follow ntsecapi.h.
>
> Best regards,
> Alex Ionescu

Okay, I'll revert this patch!

Nevertheless, the required sequence of includes

#include <windows.h>
    #include <ntsecapi.h>
    #define NTOS_MODE_USER
    #include <ntndk/ntndk.h>

looks very broken to me. Shouldn't the DDK/NDK headers get included 
first, followed by windows.h and other headers?

The MS DDK defines NT_INCLUDED right at the top of ntddk.h and the SDK 
headers skip certain sections if NT_INCLUDED is defined. Shouldn't we do 
it the same way? At least it will lead to a 'cleaner' include sequence:

#define NTOS_MODE_USER
    #include <ntndk/ntndk.h>
    #include <windows.h>
    #include <ntsecapi.h>

By the way, I planned to remove the Nt*Channel() and 
NtQueryOleDirectryFile() prototypes from include/ndk/zwfuncs.h because 
they are entirely obsolete. Do you agree?

Regards,
Eric

[Alex Ionescu]

Eric Kohl wrote:
> Alex Ionescu wrote:
> > Hi Eric,
> >
> > Unforunately, we don't have the liberty to do this, since the real 
> > PSDK file doesn't, and MS Header compatibility is a goal. The NDK 
> > must always follow ntsecapi.h.
> >
> > Best regards,
> > Alex Ionescu
>
> Okay, I'll revert this patch!
>
> Nevertheless, the required sequence of includes
> #include <windows.h>
> #include <ntsecapi.h>
> #define NTOS_MODE_USER
> #include <ntndk/ntndk.h>
>
> looks very broken to me. Shouldn't the DDK/NDK headers get included 
> first, followed by windows.h and other headers?

Well, DDK should never be followed or anywhere near windows.h! And NDK 
is an add-on to the DDK/PSDK, so it's normal that it follows. The point 
is to "Add-on" anything that isn't already provided, so a typical 
sequence would look like:

#include <windows.h>
#include <winsock2.h>
#include <ntsecapi.h>
#include <any other header you want"
#define NTOS_MODE_USER
#include <ntndk/ntndk.h>
> The MS DDK defines NT_INCLUDED right at the top of ntddk.h and the SDK 
> headers skip certain sections if NT_INCLUDED is defined. Shouldn't we 
> do it the same way? At least it will lead to a 'cleaner' include 
> sequence:
> #define NTOS_MODE_USER
> #include <ntndk/ntndk.h>
>  #include <windows.h>
> #include <ntsecapi.h>

This would mean the NDK would have to import windows.h, which I don't 
like, because some apps don't need to use windows.h (think Native Apps). 
I prefer to think of the NDK much more as an add-on, then a master 
header set.
> By the way, I planned to remove the Nt*Channel() and 
> NtQueryOleDirectryFile() prototypes from include/ndk/zwfuncs.h because 
> they are entirely obsolete. Do you agree?

Oh, yes, sure. Thanks.

OT: You'll finally get that proper Parse Routine for registry code soon.
> Regards,
> Eric

Best regards,
Alex Ionescu