MSVCRT-LIGHT - A REPLACEMENT FOR THE MSVC STANDARD LIBRARY USING MSVCRT.DLL
How and when to use these libraries:
By default, the linker includes the C / C++ standard library.
As a result, executable size is at least 40 KByte.
Or you believe to be smart, switch to the DLL version (like msvcrp80.dll),
but this will still bloat your EXE file by some kByte,
and you have to distribute the DLL file too.
By disabling the default library with /NOD or /NODEFAULTLIB switch,
useful routines are missing, especially for float-point support.
I assume that the user knows that entry points change to
int CALLBACK WinMainCRTStartup(), int CALLBACK _mainCRTStartup(),
and for a DLL to int CALLBACK _DllMainCRTStartup(HINSTANCE,DWORD,LPVOID).
On each system, an MSVCRT.DLL file is preinstalled and available.
This is the clue for these import libraries.
If your program should run on every Win32 platform, use "msvcrt-light.lib"
It has most limitations.
So when you get a linker error, you have to feed the missing function or
variable yourself, because it's simply not exported by every MSVCRT.DLL.
If your program requires NT based platforms (typically as a Unicode executable),
you can use "msvcrt-light-w2k.lib" with some more functions,
but that's only useful when you have to feed at least one function name otherwise.
Last but not least, if you compile for AMD64 platform (it was introduced with XP),
you use the "msvcrt-light-x64.lib" import library.
To inform MSVC to generate msvcrt library calls without stubs,
ensure that you link to .DLL file in your project settings.
This will define a "_DLL" preprocessor macro.
All run-time checks must be disabled.
Stack usage is limited to 4 KByte per function.
Static constructors cannot be used.
MSVCRT.DLL has no support for reading/writing UTF-8 files,
i.e. the "css=<encoding>" flag and BOM detection (see fopen()) will not work.
Float-point support needs some extra work to compile&link properly:
* Define a symbol _fltused somewhere, typically:
** char _fltused; // C
** extern "C" char _fltused; char _fltused; // C++
** EXTERN_C char _fltused; char _fltused; // both
* Avoid typecasts to int and write your own function, and use the x87 built-in
rounding feature instead of the brain-dead (int)floor(x+0.5) construct:
__forceinline __int64 rndint64(double f) {
__int64 i;
_asm fld f
_asm fistp i // This rounds 0.5 to 0 and 1.5 to 2 (Banker's Rounding)
return i; // whereas floor(0.5+0.5) rounds to 1 (Kaufmännisches Runden)
}
** It's possible to generate this sequence on each (int) cast
using the command-line option /QIfist, but this will make your source code
unreadable because the casts won't round towards zero.
And using _controlfp() will affect other operations.
** For AMD64, you need an external assembler, like ml64 oder yasm:
rndint64:
cvtpd2dq xmm0,xmm0
movq rax,xmm0
ret
* Using intrinsic forms with command-line options /Oi, /Og, and /fp:fast
(all switches necessary, but /O1 includes /Og) avoids library calls
for some functions but that's not a requirement.
<stdio> support (console applications) needs some extra work
when using MSVC newer than version 6:
_CRTIMP FILE _iob[3];
#undef stdin
#define stdin (_iob+0)
#undef stdout
#define stdout (_iob+1)
#undef stderr
#define stderr (_iob+2)
The structure for the runtime library was heavily changed with Visual Studio 2015.
You usually get errors like "unresolved symbol __acrt_iob_func".
Defining _NO_CRT_STDIO_INLINE reverts the headers to previous behaviour,
so you should define it in your project settings.
Good luck with your fat-reduced executables!
Similar applies to "htmlhelp.lib" which directly imports from hhctrl.ocx
and avoids dependencies to __security_check_cookie and other debugging symbols.
To avoid creating stubs, change your "htmlhelp.h" file prepending
_declspec(dllimport) in front of HWND WINAPI HtmlHelpA(...) resp. HtmlHelpW.
The same applies to "hidpi.h" and "hidsdi.h" file, among some others.
h#s
| Vorgefundene Kodierung: UTF-8 | 0
|