From owner-cvs-all  Mon May  7 15:51:42 2001
Delivered-To: cvs-all@freebsd.org
Received: from kalaid.f2f.com.ua (kalaid.f2f.com.ua [62.149.0.33])
	by hub.freebsd.org (Postfix) with ESMTP
	id AA56B37B422; Mon,  7 May 2001 15:51:29 -0700 (PDT)
	(envelope-from sobomax@mail-in.net)
Received: from Mail-In.Net (borey.f2f.com.ua [62.149.0.24])
	by kalaid.f2f.com.ua (8.11.3/8.11.1) with ESMTP id f47Mpu813732;
	Tue, 8 May 2001 01:51:59 +0300 (EEST)
	(envelope-from sobomax@mail-in.net)
Received: from notebook.vega.com (dialup14-30.iptelecom.net.ua [212.9.229.94])
	by Mail-In.Net (8.11.3/8.H.Z) with ESMTP id f47MpH024922;
	Tue, 8 May 2001 01:51:18 +0300 (EEST)
Date: Tue, 8 May 2001 01:51:18 +0300 (EEST)
Message-Id: <200105072251.f47MpH024922@Mail-In.Net>
To: mi@misha.privatelabs.com, eric@FreeBSD.ORG
Cc: kris@obsecurity.org, knu@iDaemons.org, kris@FreeBSD.ORG,
	cvs-committers@FreeBSD.ORG, cvs-all@FreeBSD.ORG
From: Maxim Sobolev <sobomax@mail-in.net>
Reply-To: sobomax@FreeBSD.ORG
Subject: Re: port policies
X-Mailer: Pygmy (v0.5.5)
Sender: owner-cvs-all@FreeBSD.ORG
Precedence: bulk
X-Loop: FreeBSD.ORG

On Mon, 7 May 2001 15:44:03 -0400 (EDT), Mikhail Teterin wrote:
> On  4 May, Eric Melville wrote:
> >> Hmm..don't know of any ports which do this actually. It's likely to be
> >> safer (than using high -O values) though, because the author
> >> presumably tested the build actually works with that value, and -j is
> >> more of a deterministic thing (except for race conditions in the
> >> build) than -O is.
> > 
> > These days, make -j (>1) may well be safer than gcc -O(>1), but it still
> > causes headaches for people with slower hardware and less memory. Yes, the
> > list of ports that do this is very small, but it's still fairly lame. I do
> > like proposed idea of a tunable -j number for usage across the tree.
> 
> I'd love to see the example pioneered by the qt port to become part of
> the bsd.port.mk. And, the default number of jobs should be something
> like double the number of processors. available.

No way! The default should be exctly 1 job, no matter how
many processors one has. If you think that you have enough
CPU/memory you can always increase that limit.

We don't officially support -jN even on the /usr/src, which is
usually much better maintained and tested in various conditions
that /usr/ports.

> The port should then
> need to define:
> 	PARALLEL_SAFE=	yes
> 
> I'd reject the "slower hardware" part of the argument, leaving only the
> "less memory" part :) The ports (I try to add -j in my ports wherever
> possible) build faster for more users that way, and _still build_ on
> low memory machines.

One big company from Redmond also presumes that "all users
want what majority mants" and we all know what the result is.
Please don't follow bad practice.

In summary: I'm all for adding PARALLEL_SAFE but think that
default number of jobs should be 1.

-Maxim

To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe cvs-all" in the body of the message