> On 13 May 2019, at 17:33, Joseph Myers <jos...@codesourcery.com> wrote: > > On Sun, 12 May 2019, Segher Boessenkool wrote: > >> "cc" isn't POSIX, since over a decade I think. "c99" is POSIX, and it is >> a shell script calling whatever "gcc" is first in the PATH, on most distros. > > Note that correct semantics for "c99" mean it's not a trivial wrapper; > some option reordering is needed to follow the POSIX rule that -U options > take precedence over -D options regardless of ordering on the command > line; see discussion in bug 40960 regarding support for installing variant > driver programs such as c99 with such differences in how they behave. (I > don't know if any distributions actually have wrappers that deal with > that, e.g. by using different specs in their wrapper. Another such POSIX > issue is that according to POSIX, dlopen et al should be found without > needing to special any -l options, but that could be dealt with by > adjusting the libc.so linker script, on systems using glibc.)
Darwin (Xcode release, not GCC) has stand-alone exes for c89 and c99 which appear to implement the semantics you describe. Adding these + “cc” to the Darwin GCC installation is part of what prompted my question. [the code for the Darwin c89/c99 is open-sourced and based on a freeBSD 2002 edition, so I probably don’t need to generate something new]. It seems, from this thread that there’s no specific reason for me _not_ to install a ‘cc’ for Darwin’s GCC installation - at least it will make c++/cc consistent. Iain
