On Sat, Feb 17, 2007 at 01:58:04PM +0000, Richard W.M. Jones wrote: > > Would anyone object to a patch using __attribute__((constructor)) > throughout libvirt rather than all this if(!initialised) initialise() stuff? I don't think that annotation is widely supported enough by constructors for us to be able to use it - in particular I think it'd be a problem for the Solaris guys using libvirt unless they switched to GCC instead of the Sun compilers/linkers.

Mark McLoughlin wrote: >On Thu, 2007-02-22 at 20:00 +0000, Daniel P. Berrange wrote: >>On Sat, Feb 17, 2007 at 01:58:04PM +0000, Richard W.M. Jones wrote: >>>Would anyone object to a patch using __attribute__((constructor)) >>>throughout libvirt rather than all this if(!initialised) initialise() >>>stuff? >>I don't think that annotation is widely supported enough by constructors >>for us to be able to use it - in particular I think it'd be a problem for >>the Solaris guys using libvirt unless they switched to GCC instead of the >>Sun compilers/linkers. > > A better reason[1], IMHO is that it just makes code more obtuse for >little gain - e.g. what order are they called in? And you'd have weird >stuff where the handler set by virSetErrorFunc() would not be called if >an error occurred in a constructor, requiring the library user to also >use a constructor in order to call virSetErrorFunc() ... Actually this is an argument for using constructors, because the linker should order them in a sensible order taking into account module dependencies. Anyhow, I take the point about people using non-GCC compilers (and I'm a bit surprised too ...) Does anyone know if the Solaris compiler supports C99 idioms? The remote patch currently uses C99 idioms all over the place ...