TODO

Last updated: 2012/06/05

* Review Win32::API::Struct

  1) make code style consistent
  2) remove warnings
  3) implement test cases for the various structure types,
     including compound structures, struct containing arrays,
	 and types non aligned to 4 bytes
  4) implement a 64-bit flag, so that all word sizes are automatically
     calculated as 8 bytes instead of hardwiring 4 bytes


bulk88's comments

1.  callback's in parameters are similar to api's out parameter and
    callback's out parameter is similar to api's in parameter, merge them
2.  vararg prototype processing, whether to supply the stack pointer as an IV
    or implement a Perl class with c lib va_list API IDK
3.  Callback.xs needs to be eliminated, its too small to put in its own DLL
4.  Use Win32's LoadLibrary and FreeLibrary instead of our own to save DLL
    space?
5.  *DONE* APIPARAM needs to be a union
6.  create 1 time use intra-process DLLs with Callback's function pointers in
    the DLL export table?
7.  create standalone DLLs that start a perl interps in the process and have
    Callback's function pointers in the DLL export table?
8.  stop using unpackstring in Callback, start using unpackstring in API?
9.  Register user supplied callbacks for type packing/unpacking, probably
    it should inspect caller() to make sure one packages/module's type callbacks
    dont collide with the type callbacks of another
10. Void ** doesn't parse.
11. Callback needs more logic to be moved from runtime to constructor time.
    Currently a bad type letter isn't caught until runtime, with
    a croak/longjmp through the foreign C lib.
12. Multi OS threading catching stderr message isn't tested properly by
    the test suite.
13. Certain calling conventions on certain platforms, pass by copy silently
    becomes pass by pointer when the struct exceeds a certain size.
    API/Callback's C parsing don't do this at the moment.
14. *DONE* Alot of GV lookups could be cached to their pointers and kept inside
    START_MY_CXT globals. On clone the pointers can be refreshed from the new
    interp. I (bulk88) have done this before.
15. *DONE* When using hv_fetch and hv_store, a slight performance optimization can
    be obtained by the char * being passed, being exactly equal to the PV *
    in the SV thats part of the HEK using the shared string table system,
    and by precalculating the hash number for that string per interp.
    I (bulk88) have done this before.
16. *DONE* Stop installing Test.pm on users computers, delete existing copies.
17. If someone asks for fastcall/thiscall on x86, add it.
18. If someone asks for MSVC C++ name demangling and supplies a patch, add it.
    No reason other it being extremely extremely difficult that C++ can't be
    supported.
19. *DONE* On x86, add a stack pointer checker similar to runtime checks by visual
    studio, requires asm code, will catch stdcall funcs called with wrong
    prototype, or very broken cdecl funcs written in asm, maybe setjmp/longjmp
    is of use for this idea
20. XP and newer has "Manifests" and "Activation Contexts", maybe some of that
    API will have to be exposed one day in Win32::API, some DLLs such as late
    2000s/2010s MS CRTs explode with error messages if they are loaded into a
    process that isn't manifest compliant.
21. Enums, unions and bitfields aren't supported by Win32::API::Struct
22. convert SEH exceptions (access vio and friends) into Perl exceptions? GCC
    SEH is a problem, SEH is completly different between Win32/Win64, SEH is
    completly different between GCC and VC, so a total of 3 different
    implementations required, GCC 64, GCC 32, and VC
23. Whats the calling convetion on WinCE or WOA? Will Perl ever run again/on
    those 2 platforms?
24. Win32::API::Struct type name system is messy, struct foo * doesn't work,
    struct foo doesn't work, pass by copy structs don't work
25. always UTF8 vs always bytes vs always perl scalar "internal format"
    semantics aren't defined, I've never seen a force bytes or force utf8
    in Win32::API, the current behaviour of Win32::API needs to be documented
    on what actually is passed to the C funcs, and maybe a discussion what
    should the behaviour should be (feeding UTF8 to Win32 'A' functions never
    does what is intended)
26. C types aren't sanitized for whitespace in many places, non matches or
    other security type bugs can come up by exploiting the lack of type name
    sanitization
27. Use DeviceIOControl to copy a APIPARAM array to a NT Driver that comes with
    Win32::API. NT Driver will run the kernel mode function and return the
    results in a APIPARAM. NT Driver is Perl License just like ::API.
28. Implement a proper OOP class for HMODULEs that is fork ok, similar to
    implementation of Win32::API::Callback::HeapBlock, current fork fix for
    DLLs is a hack.