[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date: Sat, 12 Sep 2009 22:12:20 +0400
From: Solar Designer <solar@...nwall.com>
To: john-users@...ts.openwall.com
Subject: Re: distro patches
Speaking of Gentoo's johntheripper-1.7.3.1-r1.ebuild:
On Thu, Sep 10, 2009 at 05:47:36AM -0600, RB wrote:
> It's called in the src_test stage that is turned off by default,
OK, so my guess was correct.
> and is actually bypassed if there aren't existing system-wide
> configuration files present from another installation.
That's very dirty. Besides, it relied on "./john --test" returning a
non-zero exit code on any failed test, which was not the case until my
very recent changes for 1.7.3.3.
While at it, I noticed that the ebuild uses "-fPIC -fPIE". This is
probably part of a Gentoo policy, but the performance hit is hardly
desirable for John the Ripper. Obviously, you need to benchmark it for
a make target and hash type where CFLAGS actually matter. You won't
notice a difference for most officially supported hash types on 32-bit
x86 builds (due to the assembly code), but you should notice it for
some other hash types and other make targets. In fact, even on 32-bit
x86 builds LM hashes may be affected somewhat (6.5% on my test build).
Then, do Gentoo's CFLAGS include -fomit-frame-pointer? Probably not.
That's a further performance hit where it matters.
And just why is -funroll-loops omitted from OPT_NORMAL?
I've just tried making those three changes to a test build, and indeed
this hurt performance a bit.
In general, most distro packages of John I've seen so far override the
CFLAGS in a way that often results in worse performance. This is not
just an arbitrary piece of software where the distro knows better; for
JtR, it's usually the other way around. Why not just disregard the
policy and drop those overrides? OK, it sometimes makes sense to add
things such as -march=..., but simply replacing all of CFLAGS with the
distro's is often a bad idea.
I guess I'll continue recommending custom builds over distro packages.
> > http://cvsweb.openwall.com/john
>
> :) I have a nightly script that tracks various OSS projects and
> syncs local copies of their SCM. Since I read that summary email
> before this one, I was at least aware of the changes if not yet what
> they did.
Great.
> Thanks! I'll re-base the patches off of those changes and submit them
> in the next round.
Would it be of any help if I release 1.7.3.3 now? (I guess so.)
Thanks,
Alexander
Powered by blists - more mailing lists
Powered by Openwall GNU/*/Linux -
Powered by OpenVZ