[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <5064f88a-952a-6f3f-f280-e76e09603c1b@suse.de>
Date: Mon, 20 Mar 2023 22:30:00 +0100
From: "Bernhard M. Wiedemann" <bwiedemann@...e.de>
To: Ian Rogers <irogers@...gle.com>, Ben Hutchings <benh@...ian.org>
Cc: Arnaldo Carvalho de Melo <acme@...nel.org>,
linux-perf-users@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH] perf jevents: Sort list of input files
On 20/03/2023 21.48, Ian Rogers wrote:
> I think this may already be addressed by sorting prior to output:
> https://lore.kernel.org/r/20220812230949.683239-5-irogers@google.com
>
> Could you confirm?
Hi Ian,
I was testing on 6.2.6 which includes that patch and it was still affected.
The trouble with sorting at the end is, that there can be influences of
ordering in earlier processing steps, that don't get ironed out by the
sort later.
Some more experimenting showed that only the ftw scandir needed sorting,
which allows to further simplify the patch to
"""Replicate the directory/file walking behavior of C's ...
- for item in os.scandir(path):
+ for item in sorted(os.scandir(path), key=lambda e: e.name):
action(parents, item)
Without the patch, a random diff in pmu-events.c starts with
-static const struct compact_pmu_event pme_amdzen2[] = {
+static const struct compact_pmu_event pme_silvermont[] = {
While I'm testing on scratch ext4 filesystems where dirindex causes
randomness, you could also use the disorderfs FUSE-filesystem with its
shuffle mode to give you random order.
Ciao
Bernhard M.
Download attachment "OpenPGP_signature" of type "application/pgp-signature" (237 bytes)
Powered by blists - more mailing lists