Sergey Poznyakoff g...@gnu.org.ua skribis:
FWIW I think it’s very important that tar --sort=name remains the
default, as it guarantees that archive production is deterministic.
Actually, the default is not to sort directory entries at all: tar
stores them in the same order as returned by the
Hi,
Sergey Poznyakoff g...@gnu.org.ua skribis:
According to [1], using directory lists ordered by inode can speed up
tar archivation by up to 40% on ext[34] filesystems (confirmed by my own
measurements). Please take a look at the attached patch. Is it OK
to apply?
FWIW I think it’s very
Hi Ludovic,
FWIW I think it’s very important that tar --sort=name remains the
default, as it guarantees that archive production is deterministic.
Actually, the default is not to sort directory entries at all: tar
stores them in the same order as returned by the underlying system calls.
I don't
Hi,
Any objections to pushing the latest version of this patch to git head?
Regards,
Sergey
On 02/13/2014 10:30 AM, Sergey Poznyakoff wrote:
Any objections to pushing the latest version of this patch to git head?
Thanks, it looks good to me. Maybe also fold in the attached minor
change, which simplifies the code and which I don't think will hurt
performance significantly. But it's
Hi Paul,
It simplifies things by removing the obsolescent function
fdsavedir, and omitting the static variable that stores internal
state.
Great, thank you. It passed the test successfully. While at it,
it appeared to me that we can also avoid some memory reallocations
and one iteration
92ddbfb74f05e5cded933f760cddd0ce007001ae Mon Sep 17 00:00:00 2001
From: Sergey Poznyakoff g...@gnu.org.ua
Date: Tue, 11 Feb 2014 10:52:44 +0200
Subject: [PATCH] savedir: optionally produce ordered directory list
New function streamsavedir_ordered returns directory
entries ordered by names or inode
Thanks for doing that. How about the attached (untested) patch
instead? It simplifies things by removing the obsolescent function
fdsavedir, and omitting the static variable that stores internal state.
The downside is that it changes the API, but that's OK in gnulib. It
fixes a couple of