Fix C89 compliance

[beutlich]

POSIX is not availabe if C-Sources are compiled with -ansi (or -std=c89).

[sjoelund]

The -ansi flag is not the same as being pure C89 without POSIX. POSIX functions are still available and there is no way to disable them on a standard GCC install on Linux as far as I know. That's why the _POSIX_VERSION macro is enabled. Almost nothing in the Linux system headers changes depending on the __STRICT_ANSI__ macro being present. Some macros change from being x ? y : z to if (x) y; else z;, disables inlining of some math functions. But nothing really major.

How did you find this in the first place?

[beutlich]

How did you find this in the first place?

I found this when compiling C-sources with -std=c89 and got a compiler warning:

libtool: compile:  gcc "-DPACKAGE_NAME=\"Modelica Standard Library Tables\"" -DPACKAGE_TARNAME=\"libmodelicastandardtables\" -DPACKAGE_VERSION=\"4.0.0\" "-DPACKAGE_STRING=\"Modelica Standard Library Tables 4.0.0\"" -DPACKAGE_BUGREPORT=\"https://github.com/modelica/ModelicaStandardLibrary\" -DPACKAGE_URL=\"https://modelica.org\" -DPACKAGE=\"libmodelicastandardtables\" -DVERSION=\"4.0.0\" -DSTDC_HEADERS=1 -DHAVE_SYS_TYPES_H=1 -DHAVE_SYS_STAT_H=1 -DHAVE_STDLIB_H=1 -DHAVE_STRING_H=1 -DHAVE_MEMORY_H=1 -DHAVE_STRINGS_H=1 -DHAVE_INTTYPES_H=1 -DHAVE_STDINT_H=1 -DHAVE_UNISTD_H=1 -DHAVE_DLFCN_H=1 -DLT_OBJDIR=\".libs/\" -DHAVE_DIRENT_H=1 -DHAVE_LOCALE_H=1 -DHAVE_TIME_H=1 -I. -DDUMMY_FUNCTION_USERTAB=1 -DTABLE_SHARE=1 -g -O2 -fno-delete-null-pointer-checks -std=c89 -MT ../../C-Sources/ModelicaInternal.lo -MD -MP -MF ../../C-Sources/.deps/ModelicaInternal.Tpo -c ../../C-Sources/ModelicaInternal.c -o ../../C-Sources/ModelicaInternal.o
../../C-Sources/ModelicaInternal.c: In function ‘ModelicaInternal_getTime’:
../../C-Sources/ModelicaInternal.c:1339:12: warning: assignment makes pointer from integer without a cast [-Wint-conversion]
     tlocal = localtime_r(&calendarTime, &tres); /* Time fields in local time zone */
            ^

localtime_r of time.h is not available when compiling with -std=c89.

[HansOlsson]

How did you find this in the first place?

I found this when compiling C-sources with -std=c89 and got a compiler warning:

libtool: compile:  gcc "-DPACKAGE_NAME=\"Modelica Standard Library Tables\"" -DPACKAGE_TARNAME=\"libmodelicastandardtables\" -DPACKAGE_VERSION=\"4.0.0\" "-DPACKAGE_STRING=\"Modelica Standard Library Tables 4.0.0\"" -DPACKAGE_BUGREPORT=\"https://github.com/modelica/ModelicaStandardLibrary\" -DPACKAGE_URL=\"https://modelica.org\" -DPACKAGE=\"libmodelicastandardtables\" -DVERSION=\"4.0.0\" -DSTDC_HEADERS=1 -DHAVE_SYS_TYPES_H=1 -DHAVE_SYS_STAT_H=1 -DHAVE_STDLIB_H=1 -DHAVE_STRING_H=1 -DHAVE_MEMORY_H=1 -DHAVE_STRINGS_H=1 -DHAVE_INTTYPES_H=1 -DHAVE_STDINT_H=1 -DHAVE_UNISTD_H=1 -DHAVE_DLFCN_H=1 -DLT_OBJDIR=\".libs/\" -DHAVE_DIRENT_H=1 -DHAVE_LOCALE_H=1 -DHAVE_TIME_H=1 -I. -DDUMMY_FUNCTION_USERTAB=1 -DTABLE_SHARE=1 -g -O2 -fno-delete-null-pointer-checks -std=c89 -MT ../../C-Sources/ModelicaInternal.lo -MD -MP -MF ../../C-Sources/.deps/ModelicaInternal.Tpo -c ../../C-Sources/ModelicaInternal.c -o ../../C-Sources/ModelicaInternal.o
../../C-Sources/ModelicaInternal.c: In function ‘ModelicaInternal_getTime’:
../../C-Sources/ModelicaInternal.c:1339:12: warning: assignment makes pointer from integer without a cast [-Wint-conversion]
     tlocal = localtime_r(&calendarTime, &tres); /* Time fields in local time zone */
            ^

localtime_r of time.h is not available when compiling with -std=c89.

We have to be clear regarding ISO (and ANSI) compliance. ANSI and ISO standards without a year normally refer to the latest released one (as far as I understand the previous standards are even superseded by the new ones), and as far as I understand the planned ISO standard nicknamed C23 (or C2x) is planning to add localtime_r. I don't know if it works with -std=c2x

Thus we should call this C89 compliance, not ISO Standard compliance; and ensure that it works with C23.

[sjoelund]

If you #include <unistd.h>, you will get some POSIX stuff available (but not things like putenv either; see -Werror=implicit-function-declaration). Most problematic is that it defines _POSIX_VERSION, which the standards say should only be available if that's the POSIX version available. It feels wrong to look for some GCC internal flag to be able to figure out if we actually have POSIX or not...

I guess the correct way on Linux/glibc would be to #include <features.h> instead of unistd.h. But then you need to know if that header is available (it's not on Mac). Also, on Mac localtime_r is available when STRICT_ANSI is active.

The following seems to work in both Linux and Mac (unix is not defined if STRICT_ANSI is given; could also be !defined(__STRICT_ANSI)):

#if (defined(__unix__) && (!defined(__linux__) || defined(unix))) || defined(__APPLE_CC__)
  #include <unistd.h>
#endif
#if !defined(_POSIX_) && defined(_POSIX_VERSION)
  #define _POSIX_ 1
#endif

That seems slightly better since we won't load unistd.h if -ansi is given

[sjoelund]

Or perhaps we could look for unix and not GLIBC with STRICT_ANSI. Might capture more OSes...

[beutlich]

What about defining _POSIX_C_SOURCE just as we already define _GNU_SOURCE?

See also https://stackoverflow.com/q/21867897/8520615.

[sjoelund]

What about defining _POSIX_C_SOURCE just as we already define _GNU_SOURCE?

See also https://stackoverflow.com/q/21867897/8520615.

That's also an option and would hopefully give us the correct value for _POSIX_VERSION from the headers

[beutlich]

OK, it was easier than expected. Defining _GNU_SOURCE in ModelicaInternal.c fixes the issue with localtime_r and also enables realpath.

[sjoelund]

This still does not compile with C89 because zlib requires snprintf (C99 or #define _POSIX_C_SOURCE 200112L in gzwrite.c and gzlib.c)

[beutlich]

This still does not compile with C89 because zlib requires snprintf (C99 or #define _POSIX_C_SOURCE 200112L in gzwrite.c and gzlib.c)

I disabled snprintf in case of C89. No need to worry, since gzwrite and gzlib are not used at all.

[sjoelund]

Then maybe we should remove gzlib and gzwrite sources? :)

[HansOlsson]

Looks good.
I found a separate issue with C99-testing, but that deserves another PR.