steps in bisecting the kernel

[Doug S]

Thanks, Doug. How did you find out which option to add to git to go forward?

Hi Lars, I didn't add any option to git. I Think git figured it out based on my answer for the "test this merge point" request above. Note that I didn't list it above but I also tried an answer of

git bisect bad

and git replied with a range of commit numbers. I suppose if that avenue needs to be pursued, a new bisection would be started with those good /bad points.

On your next posting, "no debian" sub-directory: On my computer, I have the debian sub-directory. I have no clue why I have it and you do not. However, it might be because I got to the next step, and you did not. Even though I did a "git bisect reset" and then the "git bisect start ..." again, maybe the debian directory is left over.

I also tried to compile (before I did the "git bisect reset" stuff), and it errors out right at the start. If I am even able to figure it out, it tends to take me forever at the best of times. However, I can not work on it right now.

I have been compiling the kernel a lot over the last year (mainly from Ubuntu, but also from Kernel.org), but not for a couple of months now. However, when things don't work, I tend to end up in some blackhole of getting lost and frustrated.

[Lars Noodén]

I did "git bisect reset" and, after it ran for about a minute, I got a debian directory where before I had none.

Taking the next step in the kernel build instructions gives a mess of errors, again about missing directories:

$ fakeroot debian/rules clean
/bin/bash: debian/debian.env: No such file or directory
sed: can't read /changelog: No such file or directory
/bin/bash: debian/debian.env: No such file or directory
sed: can't read /changelog: No such file or directory
/bin/bash: debian/debian.env: No such file or directory
sed: can't read /changelog: No such file or directory
/bin/bash: debian/debian.env: No such file or directory
sed: can't read /changelog: No such file or directory
debian/rules.d/0-common-vars.mk:10: *** first argument to `word' function must be greater than 0.  Stop.

and

$ fakeroot debian/rules binary-headers binary-generic/bin/bash: debian/debian.env: No such file or directory
sed: can't read /changelog: No such file or directory
/bin/bash: debian/debian.env: No such file or directory
sed: can't read /changelog: No such file or directory
/bin/bash: debian/debian.env: No such file or directory
sed: can't read /changelog: No such file or directory
/bin/bash: debian/debian.env: No such file or directory
sed: can't read /changelog: No such file or directory
debian/rules.d/0-common-vars.mk:10: *** first argument to `word' function must be greater than 0.  Stop.

Edit: It builds on another machine even though I have followed the same steps. Strange.

[Lars Noodén]

git bisect seems not to advance to the next case. When I enter 'git bisect bad' there is no output. Shouldn't it say something about the number of revisions and steps left?

$ git bisect bad
$ git bisect log
git bisect start
# bad: [6c417a4e4d15e4a77e299f5f41df24211a879dd4] UBUNTU: Ubuntu-2.6.38-16.67
git bisect bad 6c417a4e4d15e4a77e299f5f41df24211a879dd4

[Lars Noodén]

The instructions don't work at all. The directories debian and debian.master and their contents are needed to build the kernel, but doing a git bisect start ... removes them. git bisect reset creates them, but leave me (I think) unable to go forward with the bisect. How do I really get going with a bisect?

$ time git clone git://kernel.ubuntu.com/ubuntu/ubuntu-natty.git
Cloning into 'ubuntu-natty'...
remote: Counting objects: 1985139, done.
remote: Compressing objects: 100% (324883/324883), done.
remote: Total 1985139 (delta 1658653), reused 1967904 (delta 1641721)
Receiving objects: 100% (1985139/1985139), 500.91 MiB | 476 KiB/s, done.
Resolving deltas: 100% (1658653/1658653), done.
Checking out files: 100% (36408/36408), done.

real	20m46.517s
user	2m23.980s
sys	0m54.610s


$ cd ubuntu-natty


$ time git checkout -b mybisect origin/master
Branch mybisect set up to track remote branch master from origin.
Switched to a new branch 'mybisect'

real	0m0.288s
user	0m0.190s
sys	0m0.100s


$ git log --oneline Ubuntu-2.6.38-7.39..Ubuntu-2.6.38-8.40  | wc -l
704


$ git bisect start Ubuntu-2.6.38-8.40 Ubuntu-2.6.38-7.39
Bisecting: a merge base must be tested
[521cb40b0c44418a4fd36dc633f575813d59a43d] Linux 2.6.38
# no debian or debian.master directory

$ git bisect reset

# debian and debian.master directories created


$ git bisect start Ubuntu-2.6.38-8.40 Ubuntu-2.6.38-7.39
Bisecting: a merge base must be tested
[521cb40b0c44418a4fd36dc633f575813d59a43d] Linux 2.6.38
# now debian and debian.master directories disappear

$ fakeroot debian/rules clean
/usr/bin/fakeroot: line 178: debian/rules: No such file or directory

[Doug S]

Hi Lars,
I am finding the same as you. I thought, perhaps, this part:

Bisecting: a merge base must be tested
[521cb40b0c44418a4fd36dc633f575813d59a43d] Linux 2.6.38
# no debian or debian.master directory

might mean the raw kernel as from kernel.org before any changes to the Ubuntu way, which is compiled a different way. So, I thought maybe GIT set it up that way. However, I was still unable to compile, I tried two methods. One of those methods actually got about 5 minutes into the compile, which on my i7 computer is typically about 42% of the way.

By the way, I time my steps also.

My continued interest in this thread and your launchpad bug are because I have never actually manged to get this stuff work. I have had to switch to manually applying patches and backedits and such, which ends up very tedious. I was hoping to learn something.

<Begin rant:>Sometimes it seems to be very hard to try to help. It get s so frustrating and if anything goes out of the ordinary, the reference materials don't seem to cover the senario. The hours pile up, with little or no progress made.
<End rant>

[Doug S]

git bisect seems not to advance to the next case. When I enter 'git bisect bad' there is no output. Shouldn't it say something about the number of revisions and steps left?

$ git bisect bad
$ git bisect log
git bisect start
# bad: [6c417a4e4d15e4a77e299f5f41df24211a879dd4] UBUNTU: Ubuntu-2.6.38-16.67
git bisect bad 6c417a4e4d15e4a77e299f5f41df24211a879dd4

Lars, I get this when I enter "bad" at the first step:

doug@s15:~/temp-lars/ubuntu-natty$ git bisect reset
Previous HEAD position was 521cb40... Linux 2.6.38
Switched to branch 'mybisect'
doug@s15:~/temp-lars/ubuntu-natty$ git bisect start Ubuntu-2.6.38-8.40 Ubuntu-2.6.38-7.39
Bisecting: a merge base must be tested
[521cb40b0c44418a4fd36dc633f575813d59a43d] Linux 2.6.38
doug@s15:~/temp-lars/ubuntu-natty$ git bisect bad
The merge base 521cb40b0c44418a4fd36dc633f575813d59a43d is bad.
This means the bug has been fixed between 521cb40b0c44418a4fd36dc633f575813d59a43d and [36bb4f292ae74227c1ff4dc8b160781c0d02a0ae].

I assume in that senario we would start a new bisection with those parameters:

doug@s15:~/temp-lars/ubuntu-natty$ git bisect reset
Previous HEAD position was 521cb40... Linux 2.6.38
Switched to branch 'mybisect'
doug@s15:~/temp-lars/ubuntu-natty$ git bisect start 521cb40b0c44418a4fd36dc633f575813d59a43d 36bb4f292ae74227c1ff4dc8b160781c0d02a0ae
Some good revs are not ancestor of the bad rev.
git bisect cannot work properly in this case.
Maybe you mistake good and bad revs?

Oh, that seems to have issues also. I'm lost.

[Doug S]

I continued to try compile, and continue fail. The most recent thing I did was to just check out 521cb40b0c44418a4fd36dc633f575813d59a43d by itself and try to compile that:

git checkout -b dougbasetest 521cb40b0c44418a4fd36dc633f575813d59a43d
make clean
make oldconfig
... answer a bunch of question, accepting defaults ...
time make -j8 deb-pkg
... which eventually errors out, but doesn't tell me why (annoying).
make[1]: *** [deb-pkg] Error 2
make: *** [deb-pkg] Error 2
real    1m24.808s
user    9m58.337s
sys     0m38.462s

If instead, I check out, say:

git checkout -b dougbisect Ubuntu-2.6.38-7.39

Then I should be able to use the Ubuntu compile instructions, but the compile also failes.
I also followed a git bisection, giving guessed answers just to see where it lead:

doug@s15:~/temp-lars/ubuntu-natty$ git bisect start Ubuntu-2.6.38-8.40 Ubuntu-2.6.38-7.39
Bisecting: a merge base must be tested
[521cb40b0c44418a4fd36dc633f575813d59a43d] Linux 2.6.38
doug@s15:~/temp-lars/ubuntu-natty$ git bisect good
Bisecting: 351 revisions left to test after this (roughly 9 steps)
[516b498acc786576d4dd4238f7d67812a5c5df86] UBUNTU: SAUCE: hid: ntrig: New ghost-filtering event logic
doug@s15:~/temp-lars/ubuntu-natty$ head -1 debian.master/changelog
linux (2.6.37-3.11) natty; urgency=low
doug@s15:~/temp-lars/ubuntu-natty$ git bisect good
Bisecting: 175 revisions left to test after this (roughly 8 steps)
[5a5f7782e0ab5d5c93bc77179e4b471aa85abd60] UBUNTU: ubuntu: ndiswrapper -- fix interaction between __packed and packed
doug@s15:~/temp-lars/ubuntu-natty$ head -1 debian.master/changelog
linux (2.6.37-13.27) UNRELEASED; urgency=low
doug@s15:~/temp-lars/ubuntu-natty$ git bisect good
Bisecting: 87 revisions left to test after this (roughly 7 steps)
[538b2ab8ca9fca3eb02209cd6ad09724607dab99] UBUNTU: rebase fixes Bug #722689
doug@s15:~/temp-lars/ubuntu-natty$ head -1 debian.master/changelog
linux (2.6.38-5.32) UNRELEASED; urgency=low
doug@s15:~/temp-lars/ubuntu-natty$ git bisect good
Bisecting: 43 revisions left to test after this (roughly 6 steps)
[84a78d3f5a9132a3693cbe80a55b5d4cfa2f4a9d] UBUNTU: Enable common packages for armhf
doug@s15:~/temp-lars/ubuntu-natty$ head -1 debian.master/changelog
linux (2.6.38-7.36) UNRELEASED; urgency=low
doug@s15:~/temp-lars/ubuntu-natty$ git bisect good
Bisecting: 21 revisions left to test after this (roughly 5 steps)
[1018ae2e394feb81b9884178b35132a34944f7ba] UBUNTU: [Config] update configs after v2.6.38.1 rebase
doug@s15:~/temp-lars/ubuntu-natty$ head -1 debian.master/changelog
linux (2.6.38-8.40) UNRELEASED; urgency=low
doug@s15:~/temp-lars/ubuntu-natty$ git bisect bad
Bisecting: 10 revisions left to test after this (roughly 4 steps)
[7778de98450721faff3500a264aa2ccda9eb36cf] UBUNTU: Ubuntu-2.6.38-7.39
doug@s15:~/temp-lars/ubuntu-natty$ head -1 debian.master/changelog
linux (2.6.38-7.39) natty; urgency=low
doug@s15:~/temp-lars/ubuntu-natty$ git bisect good
Bisecting: 5 revisions left to test after this (roughly 3 steps)
[4a051c3c21a7e0276baba305f9ecbc377875a278] (drop after v2.6.38) HID: ntrig: apply NO_INIT_REPORTS quirk
doug@s15:~/temp-lars/ubuntu-natty$ head -1 debian.master/changelog
linux (2.6.38-8.40) UNRELEASED; urgency=low

I also tried:

doug@s15:~/temp-lars/ubuntu-natty$ git bisect reset
Previous HEAD position was 4a051c3... (drop after v2.6.38) HID: ntrig: apply NO_INIT_REPORTS quirk
Switched to branch 'mybisect'
doug@s15:~/temp-lars/ubuntu-natty$ git checkout -b dougbisect Ubuntu-2.6.38-7.39
Switched to a new branch 'dougbisect'
doug@s15:~/temp-lars/ubuntu-natty$ git bisect start Ubuntu-2.6.38-8.40 Ubuntu-2.6.38-7.39
Bisecting: a merge base must be tested
[521cb40b0c44418a4fd36dc633f575813d59a43d] Linux 2.6.38

[Lars Noodén]

I tried most of today also. I have a lot of notes but not luck in compiling. Eventually one of the changes (or something on the partition) went wrong and I could not delete files or remove subdirectories. I lost a lot of time restoring the partition from backup.

There is a new comment on the bug report, I will get to that soon.

[Doug S]

I tried the instructions/comment from the problem report. I got nothing:

doug@s15:~/temp-lars/ubuntu-natty$ git rev-list Ubuntu-2.6.38-8.40 | grep $(git log --pretty=oneline -1 Ubuntu-2.6.38-7.39 | cut -d' ' -f1)
doug@s15:~/temp-lars/ubuntu-natty$

Leaving the other stuff out of the above command, gives a complete list of commit numbers. Refering to the new link in the problem report and correlating, if we could get compiles to work, I would be tempted to jump directly to these two:

63e248e086e8a80b7345982d22d83b56d4986532
d139156bc7b509b78e9fb254c16dd2fa8d3d7f82
e6b8d2b77c7cc690015f97c771cbe9b6d1d6ef57
0ad40e118007ff770e2479f5afcdb9103782b4ba
ffaee98cdff45f84d43bd6ca0b016464cfd42984
04e8a8adfbebfab3762e494bee98521a01d10ef7
219c03274bb3ec6c74c714996461d6be4f9a8df3
7c4cec7209ffb6a718e6f15400437c0e2123127a

Because 04e8a8 is "add touchpad sysfs file".
Lars, you had mentioned in an earlier post that you did have a successful compile. Are you still able to compile on that computer?

[Lars Noodén]

I can still compile on it, but the kernel won't build without debian/rules and 'git bisect start' seems to remove that directory's contents. 'git bisect reset' seems to create the needed files. I'm not sure how to resolve that.