This approach was used in the ATL/WTL (Active Template Library, Windows Template Library) in the early 2000-s. It was a bad idea, because you need to generate executable code, interfering with NX-bit memory protection.
I'm genuinely surprised Microsoft's attitude towards "wndprocs don't have a context pointer" was "let's JIT compile a trampoline to hold the context pointer" and not to add support for a five-parameter wndproc into USER.dll, or have a wrapper that grabs GWLP_USERDATA and copies it to the register this lives in.
> I'm genuinely surprised Microsoft's attitude towards "wndprocs don't have a context pointer"
They designed windows classes to be reusable, and assumed many developers going to reuse windows classes across windows.
Consider the following use case. Programmer creates a window class for a custom control, registers the class. Designs a dialog template with multiple of these custom controls in a single dialog. Then creates the dialog by calling DialogBoxW or similar.
These custom controls are created automatically multiple at once, hard to provide context pointers for each control.
The stdcall calling convention used APIs and API callbacks on Windows x86 doesn't use registers at all, all parameters are passed on the stack. MSVC does support thiscall/fastcall/vectorcall conventions that pass some values in registers, but the system APIs and COM interfaces all use stdcall.
Windows x64 and ARM64 do use register passing, with 4 registers for x64 (rcx/rdx/r8/r9) and 8 registers for ARM64 (x0-x7). Passing an additional parameter on the stack would be cheap compared to the workarounds that everyone has to do now.
This somewhat reminds me of the old MakeProcInstance mechanism in Win16, which was quickly rendered obsolete by someone who made an important realisation: https://www.geary.com/fixds.html
Another seemingly underutilised feature closely related to {Get,Set}WindowLong is cbClsExtra/cbWndExtra which lets you allocate additional data associated with a window, and store whatever you want there. The indices to the GWL/SWL function are quite revealing of how this mechanism works:
> This is more work than going through GWLP_USERDATA
Indeed, aside from a party trick, why build an executable trampoline at runtime when you can store and retrieve the context, or a pointer to the context, with SetWindowLong() / GetWindowLong() [1]?
Slightly related: in my view Win32 windows are a faithful implementation of the Actor Model. The window proc of a window is mutable, it represents the current behavior, and can be changed in response to any received message. While I haven't personally seen this used in Win32 programs it is a powerful feature as it allows for implementing interaction state machines in a very natural way (the same way that Miro Samek promotes in his book.)
There's an annoying corner case when using SetWindowLongPtr/GetWindowLongPtr() -- Windows sends WM_GETMINMAXINFO before WM_NCCREATE. This can be worked around with a thread local, but a trampoline inherently handles it. Trampolines are also useful for other Win32 user functions that don't have an easy way to store context data, such as SetWindowsHookEx(). They're also slightly faster, though GetWindowLongPtr() at least seems able to avoid a syscall.
The code as written, though, is missing a call to FlushInstructionCache() and might not work in processes that prohibit dynamic code generation. An alternative is to just pregenerate an array of trampolines in a code segment, each referencing a mutable pointer in a parallel array in the data segment. These can be generated straightforwardly with a little template magic. This adds size to the executable unlike an empty RWX segment, but doesn't run afoul of any dynamic codegen restrictions or require I-cache flushing. The number of trampolines must be predetermined, but the RWX segment has the same limitation.
I wasn't aware of the thread local trick, I solve this problem by not setting WS_VISIBLE and calling SetWindowPos & ShowWindow after CreateWindow returns (this solves some other problems as well..)
I'm not convinced this is always guaranteed for a Windows x64 program. When running on bare x64 hardware, FlushInstructionCache() does seem to be an (inefficient) noop on Windows 11 x64, but when running in emulation on Windows 11 ARM64, it's running a significantly larger amount of ARM64 native code -- it looks like it might be ensuring that stale JIT code is flushed.
Hah! I usually allocate trampolines at runtime, as the article suggests, but reserving R/W space for them within the application's memory space is a cute trick.
Probably not useful for most of my use cases (I'm usually injecting a payload, so I'd still have the pointer-distance issue between the executable and my payload), but it's still potentially handy. Will have to keep that around!
You can't turn a capturing C++ lambda into a WNDPROC, which is an ordinary function pointer. You'd still have to ferry the lambda via a context pointer, which is what this blog post and the other solutions in the comments are all about.
If you mean that you can call a C++ lambda from a static C callback via a context pointer, of course you can do that, it's not very mind-boggling. Rust FFI libraries similarly have to do that trick all the time to turn a closure into a C callback. The primary problem with WNDPROC is how to get that context pointer in the first place, which is the part that OP and everyone in the comments are talking about.
I hate to say it (and I know a lot of C apologists will downvote it), but there is no native closure in C, all you have is a function pointer in C, and you need to manually add the "context" pointer to make it a closure, in the strict (textbook) sense. That's because C does not have the concept of "data ownership", only automatic memory (that is on stack or register) or manual memory (in the sense of malloc/sbrk'd blocks), but a (again, textbook definition of) closure requires you to have access to the data of caller/"parent"/upper layer [^1].
And that's why I generally don't see C to have closures, and requires a JIT/dynamic code generation approach as this article has actually done (using shadow stacks). There is also a hack in GNU C which introduce local function lambda, but it is not in ISO C, and obviously won't in the next decade or so.
This approach was used in the ATL/WTL (Active Template Library, Windows Template Library) in the early 2000-s. It was a bad idea, because you need to generate executable code, interfering with NX-bit memory protection.
Windows actually had a workaround in its NX-bit implementation that recognized the byte patterns of these trampolines from the fault handler: https://web.archive.org/web/20090123222148/http://support.mi...
I'm genuinely surprised Microsoft's attitude towards "wndprocs don't have a context pointer" was "let's JIT compile a trampoline to hold the context pointer" and not to add support for a five-parameter wndproc into USER.dll, or have a wrapper that grabs GWLP_USERDATA and copies it to the register this lives in.
> I'm genuinely surprised Microsoft's attitude towards "wndprocs don't have a context pointer"
They designed windows classes to be reusable, and assumed many developers going to reuse windows classes across windows.
Consider the following use case. Programmer creates a window class for a custom control, registers the class. Designs a dialog template with multiple of these custom controls in a single dialog. Then creates the dialog by calling DialogBoxW or similar.
These custom controls are created automatically multiple at once, hard to provide context pointers for each control.
Doesn't x32 only have four registers available in the calling convention, AX-DX?
The stdcall calling convention used APIs and API callbacks on Windows x86 doesn't use registers at all, all parameters are passed on the stack. MSVC does support thiscall/fastcall/vectorcall conventions that pass some values in registers, but the system APIs and COM interfaces all use stdcall.
Windows x64 and ARM64 do use register passing, with 4 registers for x64 (rcx/rdx/r8/r9) and 8 registers for ARM64 (x0-x7). Passing an additional parameter on the stack would be cheap compared to the workarounds that everyone has to do now.
The x32 ABI uses the same fastcall convention as the regular x86-64 ABI. It's mostly syscall numbers that are affected by the shrunk pointers.
It was also used by Delphi in 90s.
This somewhat reminds me of the old MakeProcInstance mechanism in Win16, which was quickly rendered obsolete by someone who made an important realisation: https://www.geary.com/fixds.html
Another seemingly underutilised feature closely related to {Get,Set}WindowLong is cbClsExtra/cbWndExtra which lets you allocate additional data associated with a window, and store whatever you want there. The indices to the GWL/SWL function are quite revealing of how this mechanism works:
https://learn.microsoft.com/en-us/windows/win32/api/winuser/...
All my window classes uses cbWndExtra, and I leave GWLP_USERDATA for the user who is creating windows.
> This is more work than going through GWLP_USERDATA
Indeed, aside from a party trick, why build an executable trampoline at runtime when you can store and retrieve the context, or a pointer to the context, with SetWindowLong() / GetWindowLong() [1]?
Slightly related: in my view Win32 windows are a faithful implementation of the Actor Model. The window proc of a window is mutable, it represents the current behavior, and can be changed in response to any received message. While I haven't personally seen this used in Win32 programs it is a powerful feature as it allows for implementing interaction state machines in a very natural way (the same way that Miro Samek promotes in his book.)
[1] https://learn.microsoft.com/en-us/windows/win32/api/winuser/...
There's an annoying corner case when using SetWindowLongPtr/GetWindowLongPtr() -- Windows sends WM_GETMINMAXINFO before WM_NCCREATE. This can be worked around with a thread local, but a trampoline inherently handles it. Trampolines are also useful for other Win32 user functions that don't have an easy way to store context data, such as SetWindowsHookEx(). They're also slightly faster, though GetWindowLongPtr() at least seems able to avoid a syscall.
The code as written, though, is missing a call to FlushInstructionCache() and might not work in processes that prohibit dynamic code generation. An alternative is to just pregenerate an array of trampolines in a code segment, each referencing a mutable pointer in a parallel array in the data segment. These can be generated straightforwardly with a little template magic. This adds size to the executable unlike an empty RWX segment, but doesn't run afoul of any dynamic codegen restrictions or require I-cache flushing. The number of trampolines must be predetermined, but the RWX segment has the same limitation.
I wasn't aware of the thread local trick, I solve this problem by not setting WS_VISIBLE and calling SetWindowPos & ShowWindow after CreateWindow returns (this solves some other problems as well..)
FlushInstructionCache isn't needed on x86_64. I-cache and D-cache are coherent.
I'm not convinced this is always guaranteed for a Windows x64 program. When running on bare x64 hardware, FlushInstructionCache() does seem to be an (inefficient) noop on Windows 11 x64, but when running in emulation on Windows 11 ARM64, it's running a significantly larger amount of ARM64 native code -- it looks like it might be ensuring that stale JIT code is flushed.
The combination `GWLP_USERDATA` to pass state and `GWLP_WNDPROC` to update the actual wnd procedure is what I used in my [rust wrapper](https://github.com/timokroeger/winmsg-executor/blob/main/src...).
This two step approach is the only way I found to use rust closures for wndproc without double allocation and additional indirection.
Hah! I usually allocate trampolines at runtime, as the article suggests, but reserving R/W space for them within the application's memory space is a cute trick.
Probably not useful for most of my use cases (I'm usually injecting a payload, so I'd still have the pointer-distance issue between the executable and my payload), but it's still potentially handy. Will have to keep that around!
> Taking this idea further, I’d like to generate these new functions on demand at run time akin to a JIT compiler
This is cool, but isn’t runtime code generation pretty frowned upon nowadays?
Or I don't know, just use C++ lambdas instead?
You can't turn a capturing C++ lambda into a WNDPROC, which is an ordinary function pointer. You'd still have to ferry the lambda via a context pointer, which is what this blog post and the other solutions in the comments are all about.
You kind of can, that is one of their design points, naturally you need to move the context into the body and know what to cast back from.
I guess I need to prove a point on my Github during next week.
If you mean that you can call a C++ lambda from a static C callback via a context pointer, of course you can do that, it's not very mind-boggling. Rust FFI libraries similarly have to do that trick all the time to turn a closure into a C callback. The primary problem with WNDPROC is how to get that context pointer in the first place, which is the part that OP and everyone in the comments are talking about.
No need for context pointer,
https://github.com/pjmlp/LambdaWndProc/blob/main/LambdaWndPr...
I assume by "move the context into the body" you mean using GetWindowLongPtr? Why not just use a static wndproc at that point?
I mean using a static C++ lambda that moves the context into the lambda body via capture specifier.
C++ lambdas are basically old style C++ functors that are compiled generated, with the calling address being the operator().
That doesn't sound like a valid wndproc
It certainly is,
https://github.com/pjmlp/LambdaWndProc/blob/main/LambdaWndPr...
It looks like you missed the part where you "move the context into the lambda body via capture specifier."
I hate to say it (and I know a lot of C apologists will downvote it), but there is no native closure in C, all you have is a function pointer in C, and you need to manually add the "context" pointer to make it a closure, in the strict (textbook) sense. That's because C does not have the concept of "data ownership", only automatic memory (that is on stack or register) or manual memory (in the sense of malloc/sbrk'd blocks), but a (again, textbook definition of) closure requires you to have access to the data of caller/"parent"/upper layer [^1].
And that's why I generally don't see C to have closures, and requires a JIT/dynamic code generation approach as this article has actually done (using shadow stacks). There is also a hack in GNU C which introduce local function lambda, but it is not in ISO C, and obviously won't in the next decade or so.
[^1]: https://en.wikipedia.org/wiki/Closure_(computer_programming)
[dead]