[Linux-ha-dev] FIFOs, buffering, bugs, etc.

[Alan Robertson]

bmartin@penguincomputing.com wrote:
> 
> On Tue, Mar 06, 2001 at 07:24:03PM -0800, bmartin@penguincomputing.com wrote:
> > On Tue, Mar 06, 2001 at 08:06:53PM -0700, Alan Robertson wrote:
> > > Hi,
> > >
> > > I renamed one of the FIFOs, and put lots more comments on what they are and
> > > how they work in the code.  I believe that I put the appropriate buffering
> > > patch into the code according to the current theory.  It now reads like
> > > this:
> > >
> > > #if 0
> > >         /* Probably a bad idea ;-) */
> > >         setvbuf(pi->ReplyFIFO, ReplyFdBuf, _IOLBF, sizeof(ReplyFdBuf));
> > > #else
> > >         setvbuf(pi->ReplyFIFO, NULL, _IONBF, 0);
> > > #endif
> > >
> > > I left it as a #if just to make it clearer and highlight the change for
> > > now.  If it works, then we'll remove the #if 0 construction...
> > >
> > > This is all in CVS now (as is a nice fix for the Baytech RPC3a from Juri
> > > Haberland).
> > >
> > Awesome.  I was just in the process of trying to figure what was going on :)
> >
> > If I understand this setvbuf stuff correctly, then previously clients should
> > have been line buffered by default on the input, which is most certainly not
> > the case in my current straces.  I'll do an update and see what happens.
> >
> Okey Dokey.
> 
> First off, the fix is working.  My straces are now much bigger as there is
> one read for every character.

You mean the BUG IS FINALLY FIXED??

	HURRAY!!

> My only remaining confusion about the topic is about the line buffering.
> The setvbuf manpage doesn't really explain the difference between line
> and full buffering.

On output, I understand it completely.  It means that every time you send a
'\n' it does a flush.  On input, I think I'm confused too..  I put it in
probably out of wishful thinking, and knowing I didn't want full buffering,
and not thinking too clearly.
> 
> I assumed it meant that it would call read() up until a new line.  In
> retrospect that doesn't make as much sense.  How would you call read like
> that?  You would have to call read a bunch of times with size = 1, which
> what not what was happening.
> 
> Anyhow, thanks for all the help, I don't think I could have squashed this
> one on my own.

It was my bug.  You did all the hard work.  Sorry for the bad naming of the
FIFOs, and lack of documentation.

	Thanks again, A WHOLE LOT!

Because of your valiant efforts, these fixes will be in the next stable
release!!

	-- Alan Robertson
	   alanr@unix.sh