On Thu, Jan 22, 2009 at 06:52:47PM +0100, Casper.Dik at Sun.COM wrote: > >That seems OK to me, but it might be nice if STDIO was smart enough to > >understand that writes to FILEs open for appending should be "atomic", > >and so to flush the stdio cache at application write boundaries. > > Hm, I think I'd still prefer writes when the buffer is full.
I had in mind flushing the buffer when it's full, but only up to an application write boundary (a fprintf or fwrite, say). > But we could make the log files line buffered. That's fine too. > >> * The behaviour Renaud saw is for a case where the file descriptor lacks > >> O_APPEND. > > > >OT: I could swear I've seen shells that don't use O_APPEND when using > >'>>', but ksh93 does it. > > sh doesn't, csh doesn't but the other I tried do (tcsh, ksh, ksh93, bash, > zsh) Good to know. Thanks! IMO they all should. Is there any good reason why they shouldn't?