Post by Michael ShellOn Thu, 12 Jul 2018 08:41:23 +0100
Post by Ken MoffatAnd the generic command is probably 'git revert 7744ccdbc16f' but
since I'm not currently bisecting, I'm not sure what state that
would leave things in.
Ken,
I should have realized that since Hazel was working with the full git tree,
that he could easily revert any commit within git. Duh!
Hmmmm ... I think what I was thinking was that I didn't trust the results
of the bisection within git (because the identified problem code does not
even seem be active within Hazel's config, and Hazel said he was new to
the bisection process). So, I'm not so sure the commit that is believed
to be the offender really is the guilty party. I was thinking that Hazel
would take the 4.15 source tree (from the official release tarball) and
then manually backout the suspect commit - if that works/boots, then it
is virtually certain we found the problem area (as we didn't depend on
git).
At the risk of teaching people to suck eggs ...
If someone has the git tree (either Greg's stable tree for a
particular minor version, or Linus's tree for .0) the suspected
commit can be viewed with
git show 7744ccdbc16f
(in stable trees, backported commits will have a different hash).
Assuming that command does show the suspected commit,
git show 7744ccdbc16f >/tmp/suspect1
then view /tmp/suspect1 to confirm it, and
cat /tmp/suspect1 | patch -p1 -R --dry-run
(other invocations are available, but I like pipes)
to see if it will revert cleanly. If it does, revert it, rebuild
the kernel, see if it fixes the problem.
Post by Michael ShellAnd, if we find the problem commit, the next step might be to manually
change things in the source until we home in on the offending *line*,
if possible. So, we'd be manually tweaking the source at some point
anyway.
With regard to reversing a patch (without any use of git), is it for
certain that the -R option of patch can be used to reverse *any*
patch file?
https://www.drupal.org/patch/reverse
No. For context diffs in the kernel, there have been several
instances where a hunk (of an update, but the principle is the same)
gets applied to similar code elsewhere in the file. Those cases
were with git, but I'm sure patch will do the same. But things like
that are uncommon.
Backing up the file before reversing (modern patch will often do
this, creating a .orig version - I think it depends on the amount of
fuzz) and then diffing the two and comparing to the patch, and
perhaps looking at more details in the before and after files, should
be able to catch this.
Also, if too much has changed since the patch was created then it
cannot be reversed. But I'm sure you knew that. In those cases
(e.g. a function got changed, or extra code added in the middle of
what is being changed) it needs to be fixed up manually.
Unfortunately, that is fairly common when trying to apply our own,
or distro, patches in BLFS or beyond-BLFS.
Post by Michael ShellIf it is not universally possible to create a reverse patch using only
the information in a patch file, then I'd say that that is an oversight
in the design of diff.
https://stackoverflow.com/questions/3902388/permanently-reversing-a-patch-file
http://cyberelk.net/tim/software/patchutils/
( simple standard install: ./configure --prefix=/usr
configure looks for perl and xmlto )
interdiff file.patch /dev/null > reversed.patch
The resulting reverse patch looks good to me, but when I tried a dry
patch --dry-run -p1 -i ../tip-x86-mm-x86-mm-Add-Secure-Memory-Encryption-SME-support_reverse.patch
checking file arch/x86/Kconfig
Reversed (or previously applied) patch detected! Assume -R? [n]
I don't understand why it would think the patch had been already applied
as the patch is supposed to *delete* code that is indeed in my Kconfig file.
I think the problem might be because the specific kernel tree I tried it on
has the context lines at 1436 rather than the 1415 specified by the patch.
Dunno, but in beyond-BLFS, or trying to apply a BLFS patch to a
newer version, I often get that. Sometimes I can see differences
which make me go "yes, changed code", other times it all looks
unaltered but still rejects. In either case, manually apply the
.rej items with any necessary changes, then remove the .orig and
.rej, rediff, and test to see if it builds (and if it does, test to
see if it works).
In theory, just moving a block of code by a few lines should only
increse the fuzz. But recent versions of patch seem to be more
sensitive.
Post by Michael ShellI've attached a copy of the interdiff created
tip-x86-mm-x86-mm-Add-Secure-Memory-Encryption-SME-support_reverse.patch
in case anyone wants to look at/try it.
Anyway, for the record, in the stackoverflow post, Lie Ryan's script
is not quite right. It reverses the --- and the +++ alright, but
later on it reverses the - and + at the start of the lines which also
ends up affecting the --- and +++ lines and so we end up getting -++
and +--. If he went through all that trouble to create that script,
didn't he take the time to inspect the results of it? Sigh.
Cheers,
Mike
If somebody has the original patch which they want to revert, I don't
understand why they would try to manipulate it like that. But this
is one of those cases where I can't immediately see why the Kconfig
change is treated as a revert.
ĸen
--
Keyboard not found, Press F1 to continue
--
http://lists.linuxfromscratch.org/listinfo/lfs-support
FAQ: http://www.linuxfromscratch.org/blfs/faq.html
Unsubscribe: See the above information page
Do not top post on this list.
A: Because it messes up the order in which people normally read text.
Q: Why is top-posting such a bad thing?
A: Top-posting.
Q: What is the most annoying thing in e-mail