If EMB will slash prices of Delphi to 1/4 of what it is today it will surely bring in more users just as its ancestors (borland) did to enter the market and broke all records!

procedure MyImpersonateUserToken(UserName : String; out UserData : PUserData);
begin
//get user token here
Win32Check(GetUserToken);
//get user information from token here and store it into UserData
Win32Check(GetTokenInformation);

Win32Check(ImpersonateLoggedOnUser(token));
…
end;

You might want to allocate moe that a single page. Consider the following:

type
TSomething = class
public
FA: array [0..9999999] of Byte;
FB: Integer;
end;

var
L: TSomething;
begin
L := nil;
L.FB := 9999;

// this might even succeed because you will acess the base pointer plus 9999999 + 1 + sizeof( Pointer ) = address 10000004. You will have the same problem with your guarded page…

What can I say. A preprocessor could make a huge difference here, because it would allow to inject information into the debug builds at compile time that don’t require extra magic in the compiler …

This way you cannot only keep a list of allocated blocks without requiring the compiler to offer that, you can even tell where (source file and line) a certain block originated, whether it was freed, from where (source file and line) it was freed and in the latter case you can mark a pseudo-freed page as reserved (i.e. not committed) and catch the exception. The nice part about it is that you get bounds checking with a little extra of preprocessor help

“This would actually help catch attempting to use invalidated pointers.”, and what about this code above? It has the same effect. It works now, we don’t have to wait.

“You’ve just created a second class of invalided pointer that will ultimately need its own “Assigned” function, which basically just puts the whole thing back where it was started (because Assigned no longer works, it might even be a step backwards)”. Well, everything it crackable if you have enough effort. If one doesn’t like it then one can switch it off. I won’t force it because it is merely an aid to find “bad design” accesses after freeing a pointer. If you create an adapted Assigned then you shoot yourself into the foot (You could have switched it off btw.). But how to make it fool proof? Easy, put it into the operating system. But then many applications won’t work anymore…so we’re stuck with it.

“Additionally, have you taken in to account that object instance pointers have negative offset members, and can have positive offset fields that are larger than your single allocated block? Will referencing them perhaps risk getting past your trap?” Well, it was a first design. I still can increase the memory block and set the GuardPointer address anywhere in the middle of the pages. So there should be enough room. My pages are 4096bytes that should be enough for at least 1000 methods pointers and variables in the vmt. Unfortunately, history repeats itself so we’re are stuck (again) on predicting the future. We need to set some value and hope that it is sufficient. Recall that it is just an aid and not a fool proof concept.

@GetNativeSystemInfo: Yes, I forgot to mention it. I added the other version for 32bit only so everybody can try it out.

This would actually help catch attempting to use invalidated pointers.

You’ve just created a second class of invalided pointer that will ultimately need its own “Assigned” function, which basically just puts the whole thing back where it was started (because Assigned no longer works, it might even be a step backwards)

Additionally, have you taken in to account that object instance pointers have negative offset members, and can have positive offset fields that are larger than your single allocated block? Will referencing them perhaps risk getting past your trap?

Oh, and if I am reading this right, GetNativeSystemInfo is only supported on WOW64 leading to a small problem with the large number of 32 bit only systems still out there. (that is assuming your GetNativeSystemInfo isn’t a front to provide similar functionality on win32 only systems)

The whole problem is getting over thought.