git bisect <subcommand> <options>
The command takes various subcommands, and different options depending
on the subcommand:
git bisect help
git bisect start [<bad> [<good>...]] [--] [<paths>...]
git bisect bad [<rev>]
git bisect good [<rev>...]
git bisect skip [(<rev>|<range>)...]
git bisect reset [<branch>]
git bisect visualize
git bisect replay <logfile>
git bisect log
git bisect run <cmd>...
This command uses git rev-list --bisect to help drive the
binary search process to find which change introduced a bug, given an
old "good" commit object name and a later "bad" commit object name.
Use "git bisect" to get a short usage description, and "git bisect
help" or "git bisect -h" to get a long usage description.
Basic bisect commands: start, bad, good
$ git bisect start
$ git bisect bad # Current version is bad
$ git bisect good v2.6.13-rc2 # v2.6.13-rc2 was the last version
# tested that was good
When you give at least one bad and one good versions, it will bisect
the revision tree and say something like:
Bisecting: 675 revisions left to test after this
and check out the state in the middle. Now, compile that kernel, and
boot it. Now, let’s say that this booted kernel works fine, then just
$ git bisect good # this one is good
Bisecting: 337 revisions left to test after this
and you continue along, compiling that one, testing it, and depending
on whether it is good or bad, you say "git bisect good" or "git bisect
bad", and ask for the next bisection.
Until you have no more left, and you’ll have been left with the first
bad kernel rev in "refs/bisect/bad".
Oh, and then after you want to reset to the original head, do a
to get back to the original branch, instead of being on the bisection
commit ("git bisect start" will do that for you too, actually: it will
reset the bisection state).
During the bisection process, you can say
to see the currently remaining suspects in gitk. visualize is a bit
too long to type and view is provided as a synonym.
If DISPLAY environment variable is not set, git log is used
instead. You can even give command line options such as -p and
Bisect log and bisect replay
The good/bad input is logged, and
shows what you have done so far. You can truncate its output somewhere
and save it in a file, and run
$ git bisect replay that-file
if you find later you made a mistake telling good/bad about a
Avoiding to test a commit
If in a middle of bisect session, you know what the bisect suggested
to try next is not a good one to test (e.g. the change the commit
introduces is known not to work in your environment and you know it
does not have anything to do with the bug you are chasing), you may
want to find a near-by commit and try that instead.
It goes something like this:
$ git bisect good/bad # previous round was good/bad.
Bisecting: 337 revisions left to test after this
$ git bisect visualize # oops, that is uninteresting.
$ git reset --hard HEAD~3 # try 3 revs before what
# was suggested
Then compile and test the one you chose to try. After that, tell
bisect what the result was as usual.
Instead of choosing by yourself a nearby commit, you may just want git
to do it for you using:
$ git bisect skip # Current version cannot be tested
But computing the commit to test may be slower afterwards and git may
eventually not be able to tell the first bad among a bad and one or
more "skip"ped commits.
You can even skip a range of commits, instead of just one commit,
using the "<commit1>..<commit2>" notation. For example:
$ git bisect skip v2.5..v2.6
would mean that no commit between v2.5 excluded and v2.6 included
can be tested.
Note that if you want to also skip the first commit of a range you can
use something like:
$ git bisect skip v2.5 v2.5..v2.6
and the commit pointed to by v2.5 will be skipped too.
Cutting down bisection by giving more parameters to bisect start
You can further cut down the number of trials if you know what part of
the tree is involved in the problem you are tracking down, by giving
paths parameters when you say bisect start, like this:
$ git bisect start -- arch/i386 include/asm-i386
If you know beforehand more than one good commits, you can narrow the
bisect space down without doing the whole tree checkout every time you
give good commits. You give the bad revision immediately after start
and then you give all the good revisions you have:
$ git bisect start v2.6.20-rc6 v2.6.20-rc4 v2.6.20-rc1 --
# v2.6.20-rc6 is bad
# v2.6.20-rc4 and v2.6.20-rc1 are good
If you have a script that can tell if the current source code is good
or bad, you can automatically bisect using:
$ git bisect run my_script
Note that the "run" script (my_script in the above example) should
exit with code 0 in case the current source code is good. Exit with a
code between 1 and 127 (inclusive), except 125, if the current
source code is bad.
Any other exit code will abort the automatic bisect process. (A
program that does "exit(-1)" leaves $? = 255, see exit(3) manual page,
the value is chopped with "& 0377".)
The special exit code 125 should be used when the current source code
cannot be tested. If the "run" script exits with this code, the current
revision will be skipped, see git bisect skip above.
You may often find that during bisect you want to have near-constant
tweaks (e.g., s/#define DEBUG 0/#define DEBUG 1/ in a header file, or
"revision that does not have this commit needs this patch applied to
work around other problem this bisection is not interested in")
applied to the revision being tested.
To cope with such a situation, after the inner git bisect finds the
next revision to test, with the "run" script, you can apply that tweak
before compiling, run the real test, and after the test decides if the
revision (possibly with the needed tweaks) passed the test, rewind the
tree to the pristine state. Finally the "run" script can exit with
the status of the real test to let the "git bisect run" command loop to
determine the outcome.