support for macOS and Windows

[gerwin3]

• added CI
• macOS support
• Windows

Note that the MSVC ABI is NOT supported for Windows. Just the default mingw included ABI

[BanacialPC2]

Got it to compile on windows without specifying Windows MSVC using .version = "2.16.1-4", and zig "0.15.0-dev.77+aa8aa6625";

I actually had to include "nasmlib/ilog2.c", again, and rewrite two macros in include\nasmint.h
#define INT64_C(x) x##LL // instead of x ## i64
#define UINT64_C(x) x##ULL // instead of x ## ui64

These Flags I have set to 1, else is null or the default / linux thing (from ._XOPEN_SOURCE onwards)
.CFLAGS_STD_C17 = 1,
.HAVE_ACCESS = 1,
.HAVE_DECL_STRICMP = 1,
.HAVE_DECL_STRNICMP = 1,
.HAVE_FCNTL_H = 1,
.HAVE_IO_H = 1,
.HAVE_MEMPCPY = 1,
.HAVE_SNPRINTF = 1,
.HAVE_STDINT_H = 1,
.HAVE_STDLIB_H = 1,
.HAVE_STRICMP = 1,
.HAVE_STRING_H = 1,
.HAVE_STRNICMP = 1,
.HAVE_STRNLEN = 1,
.HAVE_SYS_STAT_H = 1,
.HAVE_SYS_TYPES_H = 1,
.HAVE_VSNPRINTF = 1,
.HAVE__CHSIZE = 1,
.HAVE__FILENO = 1,
.HAVE__FSEEKI64 = 1,
.HAVE__FSTATI64 = 1,
.HAVE__FULLPATH = 1,
.HAVE__STATI64 = 1,
.HAVE__FILELENGTHI64 = 1,
.PACKAGE_BUGREPORT = "",
.PACKAGE_NAME = "",
.PACKAGE_STRING = "",
.PACKAGE_TARNAME = "",
.PACKAGE_URL = "",
.PACKAGE_VERSION = "",
.STDC_HEADERS = 1,
._ALL_SOURCE = 1,
._DARWIN_C_SOURCE = 1,
.EXTENSIONS = 1,
._GNU_SOURCE = 1,
._HPUX_ALT_XOPEN_SOCKET_API = 1,
._MINIX = null,
._NETBSD_SOURCE = 1,
._OPENBSD_SOURCE = 1,
._POSIX_SOURCE = null,
._POSIX_1_SOURCE = null,
._POSIX_PTHREAD_SEMANTICS = 1,
.STDC_WANT_IEC_60559_ATTRIBS_EXT = 1,
.STDC_WANT_IEC_60559_BFP_EXT = 1,
.STDC_WANT_IEC_60559_DFP_EXT = 1,
.STDC_WANT_IEC_60559_FUNCS_EXT = 1,
.STDC_WANT_IEC_60559_TYPES_EXT = 1,
.STDC_WANT_LIB_EXT2 = 1,
.STDC_WANT_MATH_SPEC_FUNCS = 1,
._TANDEM_SOURCE = 1,

[gerwin3]

Nice. Do the code changes required not break Linux/mac?

[Banacial]

Nice. Do the code changes required not break Linux/mac?

I used your build system, removed the windows depended usage of nasm/ilog2.c, removed the msvc dependence check and changed the config.h flags as told above.

Well, compiling without errors might just mean the compilation didn't encounter known errors, not that the build programm actually works. But I had no problems compiling with :

zig build -Dtarget=x86_64-linux -Doptimize=ReleaseSafe
zig build -Dtarget=x86_64-windows -Doptimize=ReleaseSafe
zig build -Dtarget=x86_64-macos -Doptimize=ReleaseSafe
zig build -Dtarget=aarch64-linux -Doptimize=ReleaseSafe

What bugs me is that I had to make changes to a nasm-native file: include/nasmint.h

zig cc -dM -E - < NUL (on my windows machine tells me that on my zig x86-64 windows) clang has for example predefined:

#define LLONG_WIDTH 64
#define LONG_LONG_MAX 9223372036854775807LL
#define LONG_MAX 2147483647L
#define LONG_WIDTH 32
and some more, but the nasmint.h defines under certain conditions

#define INT64_C(x) x ## i64
#define UINT64_C(x) x ## ui64

which clang cannot expand. So I with my limited compiler knowledge have no idea why that macro-branch is used instead of the other which define.

#define INT64_C(x) x ## L
#define UINT64_C(x) x ## UL

[Banacial]

oh and I have no idea what a yalm or .ci file is used for: Good old manual browser download boy, no fancy git

[Banacial]

Nice. Do the code changes required not break Linux/mac?

https://github.com/Banacial/Files_I_Need/blob/main/build_nasm.zig

[gerwin3]

Nice. Do the code changes required not break Linux/mac?

I used your build system, removed the windows depended usage of nasm/ilog2.c, removed the msvc dependence check and changed the config.h flags as told above.

Well, compiling without errors might just mean the compilation didn't encounter known errors, not that the build programm actually works. But I had no problems compiling with :

zig build -Dtarget=x86_64-linux -Doptimize=ReleaseSafe zig build -Dtarget=x86_64-windows -Doptimize=ReleaseSafe zig build -Dtarget=x86_64-macos -Doptimize=ReleaseSafe zig build -Dtarget=aarch64-linux -Doptimize=ReleaseSafe

What bugs me is that I had to make changes to a nasm-native file: include/nasmint.h

zig cc -dM -E - < NUL (on my windows machine tells me that on my zig x86-64 windows) clang has for example predefined:

#define LLONG_WIDTH 64 #define LONG_LONG_MAX 9223372036854775807LL #define LONG_MAX 2147483647L #define LONG_WIDTH 32 and some more, but the nasmint.h defines under certain conditions

#define INT64_C(x) x ## i64 #define UINT64_C(x) x ## ui64

which clang cannot expand. So I with my limited compiler knowledge have no idea why that macro-branch is used instead of the other which define.

#define INT64_C(x) x ## L #define UINT64_C(x) x ## UL

Yeah bugs me too. Seems that it thinks MSVC is active even though x86_64-windows defaults to the GNU ABI is what I thought...

[gerwin3]

ziglang/zig#9536
seems to imply it should not be defined 🤔

[gerwin3]

Okay I think what happened was:

namsint.h checks _I64_MAX to for the MSVC defines. Mingw libc does NOT define _MSC_VER but it does define _I64_MAX since it is expected on Windows. This leads nasmint.h to define a bunch of constants that will not work on non-MSVC. I fixed it by changing the line to #if defined(_MSC_VER) && _I64_MAX == 9223372036854775807 which only triggers for MSVC compilers.

It now compiles :)

[gerwin3]

@andrewrk

• This PR adds support for macOS and Windows (MSVC ABI not supported).
• It also adds a GitHub CI file that builds on each platform.

One thing that I'm not so sure about is if maybe some of the compilation flags may affect NASM output. Since there are no test cases in this repo there is no way to know for sure if there are any regressions. Maybe we could run the test suite in CI. It seems though we'd need to check in the "golden" files which are required to run the test cases against. I'm not even sure how to get those for Windows (probably by grabbing the NASM official binary).

[gerwin3]

Note that this PR also unlocks Windows and macOS support for the ffmpeg package :)