Earnie
earni****@users*****
Fri Jun 29 23:18:14 JST 2018
On 6/29/2018 6:10 AM, Keith Marshall wrote: > On 29/06/18 07:26, Eli Zaretskii wrote: >>> From: David Gressett <DGres****@amli-*****> >>> Date: Thu, 28 Jun 2018 21:52:07 -0500 >>> ../../../src/gcc-7.3.0/gcc/ada/rtinit.c:195:44: error: >>> '_O_U16TEXT' was not declared in this scope >>> __gnat_current_ccs_encoding = _O_U16TEXT; >>> >>> The other two undefined items were _O_WTEXT and _O_U8TEXT. >>> For all three of these, the compiler suggested _O_TEXT as an >>> alternative. >>> >>> rtinit.c includes <fcntl.h> which defines the three undefined items: >>> >>> #if (__MSVCRT_VERSION__ >= 0x0800) >>> #define _O_WTEXT 0x10000 >>> #define _O_U16TEXT 0x20000 >>> #define _O_U8TEXT 0x40000 >>> #endif >>> >>> fcntl.h includes _mingw.h which includes mscrtver.h, which defines >>> a default value for __MSVCRT_VERSION__ >>> >>> The default value is 0x0600, so unless an overriding #define >>> produces a larger value, these constants will be undefined. >> >> AFAIU, the above means that the Ada compiler has dropped support of >> Windows versions before Vista, and the compiler you produce will be >> unable to run on those older versions as it does on Windows 7 and >> later. Unless, that is, the source has some fallback for those older >> versions (and _O_TEXT sounds like a good idea to me up front), but I >> didn't look at the sources to see if they do have such a fallback, and >> I don't really understand what that will do, since UTF-16 text will >> not be supported that way. >> >> If you never intend to have your compiler to run on older systems, >> just define __MSVCRT_VERSION__ to the higher number, I'd say. >> >> Keith will probably have a better insight. > > Well, I really cannot condone any suggestion to arbitrarily define > __MSVCRT_VERSION__; its purpose is to declare intent to link with one of > Microsoft's non-free (not freely distributable) runtimes, such as > MSVCR80.DLL or later, *instead* of the OS-provided MSVCRT.DLL. As such, > the *only* way it should *ever* be defined to a non-default value is via > a custom GCC specs file, which *replaces* MSVCRT.DLL with the > corresponding non-free alternative, in the list of default libraries > passed to the linker. > > As David notes, the default value for __MSVCRT_VERSION__ is specified as > 0x0600; this, or any lesser value, is interpreted as a declaration of > intent to maintain compatibility with default MSVCRT.DLL linking > behaviour. Comments within <msvcrtver.h>, (for which I assume David's > reference to <mscrtver.h> is a typo), explain this. > > That said, if Vista and later versions of MSVCRT.DLL now expose APIs to > which these __MSVCRT_VERSION__ restricted constants relate, then the > restriction in <fcntl.h> should reflect that, with a _WIN32_WINNT > minimum version test to complement the __MSVCRT_VERSION__ restriction. > > Would anyone care to propose a test case for such version specific > support? I no longer use Windows for any daily purpose, but I do still > have the ability to test compatibility for WinXP and Win7, running > either as a virtual machine. Would anyone else be able to assist with > testing on other (esp. legacy) Windows versions? > The 2nd example from https://docs.microsoft.com/en-us/cpp/c-runtime-library/reference/setmode should suffice for a test case. However, I don't have any existence of any version of Windows less than 10. My assumption here is that since this affects the standard API set such as _setmode then support for these exists in the delivered MSVCRT.DLL for all newer OS versions. Therefore the __MSVCRT_VERSION__ macro is not the correct guard and _WIN32_WINNT macro should be used. However, that would mean as distributed, we would only support _WIN32_WINNT > VISTA as the build would require it. -- Earnie
Fork