Sender:	nate AT cartsys DOT com
Message-ID:	<35BD5A3D.3B5C56E1@cartsys.com>
Date:	Mon, 27 Jul 1998 21:57:33 -0700
From:	Nate Eldredge <nate AT cartsys DOT com>
MIME-Version:	1.0
To:	Martin Str|mberg <ams AT ludd DOT luth DOT se>
CC:	Eli Zaretskii <eliz AT is DOT elta DOT co DOT il>,
	DJGPP-WORKERS <djgpp-workers AT delorie DOT com>
Subject:	Re: A call to `sync' inside `spawn'
References:	<199807271748 DOT TAA29763 AT father DOT ludd DOT luth DOT se>

Martin Str|mberg wrote:
> 
> According to Eli Zaretskii:
> >
> > On Sun, 26 Jul 1998, Martin Str|mberg wrote:
> >
> > > I suppose that there is no way to flush the _write_ cache and not the
> > > read cache?
> 
> [Klippa, klapp, kluppit DOZE mess.]
> 
> > In any case, the call to `sync' before running a child program doesn't
> > even need to flush the write cache.  It just needs to make sure output
> > to the same file/device doesn't appear out-of-order, for which the DOS
> > CommitFile function is enough.  IMHO, this is simply a DOS bug (it
> > should have done that in our stead, inside its Exec function) which we
> > are working around.  I can hardly believe any Unix box calls `sync',
> > since the `sync' system call is typically prohibitively expensive on
> > Unix (it takes several seconds even on fast machines).
> 
> No, I was thinking the sync call. It's such a waste to forget the read
> cache, when it's not necessary.
> 
> And sync on my machine (Linux) doesn't take several seconds. Less than
> one, actually, and it's not idle.

I think it will depend on what the system is doing.  On Unix, it's
traditional for the `update' daemon to call `sync' every 30 seconds or
so, so if it's been fairly idle since then, there won't be much to do. 
If the system is heavily loaded, on the other hand (try `cd
/usr/src/linux; make -j 9999 &'), it will be slow.
-- 

Nate Eldredge
nate AT cartsys DOT com

- Raw text -