From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from merlin.resmel.bhp.com.au ([134.18.1.6]) by hawkwind.utcs.toronto.edu with SMTP id <2688>; Sun, 21 Mar 1993 22:33:21 -0500 Received: from cerberus.bhpese.oz.au by merlin.resmel.bhp.com.au with SMTP id AA20197 (5.65c/IDA-1.4.4 for ); Mon, 22 Mar 1993 13:32:40 +1000 Received: from localhost by cerberus.bhpese.oz.au with SMTP id AA10340; Mon, 22 Mar 1993 13:32:23 +1000; sendmail 5.67a/Sm3.6RMPSU (from Sm@cerberus.bhpese.oz.au for rc@hawkwind.utcs.toronto.edu) Message-Id: <199303220332.AA10340@cerberus.bhpese.oz.au> To: byron@netapp.com (Byron Rakitzis) Cc: rc@hawkwind.utcs.toronto.edu Subject: Re: informal survey regarding <{} In-Reply-To: Your message of "Sun, 21 Mar 93 22:12:26 EST." <9303220312.AA12650@netapp.netapp.com> X-Face: '82~l%BnDBWVn])DV^cl_%bla$T]kNbRN&]>v{ED9[" >I was wondering if anyone in rc-land is using the <{} feature >with fifos. > >Specifically, I've been thinking about ditching fifos in favor >of synchronous operation with tmp files. > >Advantages: > > o You can seek on a tmp file, so diff will work. > o tmp files are much closer in semantics to real pipes, > so I expect fewer bugs. > o fifos aren't present everywhere, files are. > o fifos don't work everywhere (e.g., the notorious > tmpfs bug), files do. > >Disadvantages: > > o Existing users might depend on fifos. > o Advantage of pipe in the kernel is lost; command must > wait on all the tmp files being written first. > >I could even argue, based on the first item in "Advantages", >that one might want to use tmp files even over /dev/fd, but >I don't want to risk starting a flame war. > >Anyway, reply to me and I can summarize to the list. I'd be interested in the tmp file approach. Fifos don't work under risc/os 4.52. I tried to write a /dev/fd file system for the kernel, but am getting some kernel crashes because of undocumented interfaces. Is there anything wrong with having tmp files as the default, and fifos & /dev/fd as options ? Sm