Nicolas Williams wrote: > On Thu, Jan 15, 2009 at 03:11:46PM -0700, Shawn M. Emery wrote: > >> Nicolas Williams wrote: >> >>> On Thu, Jan 15, 2009 at 02:57:13PM -0700, Shawn M. Emery wrote: >>> >>>> Two are being considered for OpenSolaris, so within each OS there could >>>> >>> Why? For interop with MIT krb5 user-land built by users? If MIT krb5 >>> received contributions from us that #ifdef solaris the choice of IPC >>> then this problem goes away. >>> >> I was considering an all or nothing approach, rather than using >> different IPC for kernel/user space. >> > > I'm in favor of using only doors for upcalls and user-land calls unless > there's a need to interop with previous installations of user-built MIT > krb5 CCAPI clients. >
The only existing CCAPI clients that I'm aware of are for Mac and Windows. Mac uses the mach IPC and Windows uses RPC (I believe). Therefore we shouldn't break compatibility. >>>> be options or accross multiple OS's. In any case, I am loath to >>>> mix-and-match IPCs across kernel and user space as we prevent code reuse >>>> and the inherit benefits that it provides (simplicity, risk, etc.). >>>> >>>> >>> Yes, me too. So why must we have two? >>> >>> >> I was really wanting to change the gssd IPC and thought doors was a good >> mechanism for handling this. What are your arguments for not using >> doors in this case? >> > > I'm not arguing that you not use doors. Quite the contrary. > Ah, good, something we agree on :) Shawn. --