Newsgroups: comp.lang.perl
Path: utzoo!utgpu!news-server.csri.toronto.edu!mailrus!usenet.ins.cwru.edu!ncoast!allbery
From: allbery@NCoast.ORG (Brandon S. Allbery KB8JRR)
Subject: Re: (f)flush?
Reply-To: allbery@ncoast.ORG (Brandon S. Allbery KB8JRR)
Organization: North Coast Public Access *NIX, Cleveland, OH
Date: Thu, 2 Aug 90 22:54:10 GMT
Message-ID: <1990Aug2.225410.2587@NCoast.ORG>
Followup-To: comp.lang.perl
References: <7193@star.cs.vu.nl> <8936@jpl-devvax.JPL.NASA.GOV>

As quoted from <8936@jpl-devvax.JPL.NASA.GOV> by lwall@jpl-devvax.JPL.NASA.GOV (Larry Wall):
+---------------
| In article <7193@star.cs.vu.nl> maart@cs.vu.nl (Maarten Litmaath) writes:
| : Is (f)flush such a bad idea that it isn't to be included in `libperl.a'?
| : Trivial to write, of course, but shouldn't it be STDPerl?  (I don't want
| : immediate writes all the time, you know...)
| 
| Good grief.
+---------------

Larry, let me correct this suggestion, invert it, and justify it:

I would like to have a per-filehandle immediate-flush flag to replace the
current $| variable.  There are times when I want buffered output on one
filehandle but not the others.

Additionally, I'd like a "setbuf" function, again per-filehandle.

Reason?  I sat down and wrote a cpio in Perl in response to the tar in Perl
that was posted a few weeks ago, including a library to do tape buffering...
and then remembered that Perl uses stdio and its default buffers.  I can turn
off buffering with $|, but it then goes character at a time instead of print
statement at a time, no?  And I don't necessarily want to unbuffer stdout
while I'm doing this, and setting $| every time I want to do tape I/O is a
major pain.  Worse is to have a setbuf and have it apply to all filehandles
via $|; the standard tape block size on our machine is 512K bytes....

This buffering problem is the reason I haven't tried to post my cpio library;
as is, it has a lot of excess baggage via the nonfunctional tape-buffering
routines, and it won't write tapes with compatible block sizes.  Worse, the
cpio-reading part will break on machines with stupid tape drivers which deal
with reading a small chunk of a large block (i.e. stdio-buffered read of a
20-blocked cpio tape) by returning the requested amount of data and discarding
the rest of the block.  And yes, there are machines which do that!
Mercifully, I've forgotten which one it was, but it screwed me up for a while
back when I was free-lance consulting.

++Brandon
-- 
Me: Brandon S. Allbery			VHF: KB8JRR on 224.50 (Lake County, OH)
Internet: allbery@NCoast.ORG		Delphi: ALLBERY
uunet!usenet.ins.cwru.edu!ncoast!allbery America OnLine: KB8JRR