Subject: Re: Removing tmpfs' experimental status
To: None <tls@rek.tjls.com>
From: Elad Efrat <elad@NetBSD.org>
List: tech-kern
Date: 10/22/2006 20:30:54
Thor Lancelot Simon wrote:
> On Sun, Oct 22, 2006 at 06:37:42PM +0200, Tobias Nygren wrote:
>> If you put pkgsrc's wrkobjdir on a tmpfs filesystem, like for example ...
>>
>> WRKOBJDIR=/tmp/pkgsrc-obj
>>
>> .. and then try to make install in pkgsrc/devel/autoconf, it will fail with
>> the output below. This does not occur when wrkobjdir is on an ffs
>> filesystem. I've reproduced this on i386 and alpha.
> 
> And I have reproduced it on amd64.  When I looked at the autoconf tests
> that failed, they had clearly been corrupted in memory.  This is why I
> don't use tmpfs on any of my systems: I may only _notice immediately_ when
> data corruption causes autoconf to fail, but that doesn't mean that other
> data isn't being corrupted!
> 
> Not good.

Yes, so I guess it stays experimental. :)

On a different note, somewhat off-topic, I'm going to enable by default
both fileassoc(9) and veriexec(9) in GENERIC. Reasons being that unless
Veriexec is actually *used* it doesn't hurt to have it there, and having
it enabled by default makes them more accessible.

Here are differences in kernel sizes. Base kernel is GENERIC with
DEBUG/DIAGNOSTIC/DDB/LOCKDEBUG:

   text    data     bss     dec     hex filename
8042325  332064  434008 8808397  8667cd base
8045261  332064  434072 8811397  867385 base+fileassoc
8056685  332096  434136 8822917  86a085 base+fileassoc+veriexec

Are there any objections for this?

-e.

-- 
Elad Efrat