On Thu, Jan 30, 2014 at 03:48:33PM +0000, Nick wrote: > On Thu, Jan 30, 2014 at 03:35:25PM +0000, sin wrote: > > I think this is probably the best course of action > > at this point. > > ... > > Support for character and block devices is optional in POSIX. > > We cannot guarantee that this will work correctly and it is not > > portable, so just drop support for it in tar(1). > > I'm not sure I agree. We shouldn't be tethered to POSIX if it means > losing useful functionality for no practical reason. Are there any > systems out there that don't have support for character and block > devices? Haiku isn't a compelling argument to me. In any case, I > prefer the #ifdef solution. #ifdefs are dangerous, but I see nothing > wrong with how you used them here.
BTW, I will be preparing a README for sbase and I'd like to briefly mention on which OS'es it has been tested. I've tried a bunch of different compilers (tcc, pcc, gcc, clang, nwcc) on various combinations of OS'es and architectures (*BSD, Solaris, Linux, Haiku). On Haiku, everything works except nice(1) and renice(1) - they don't have getpriority()/setpriority() (that's their fault.) If anyone has access to some special oldschool UNIX systems, I'd like to see if sbase compiles and works on those. cheers, sin