git bisect start [--term-{old,good}=<term> --term-{new,bad}=<term>]

	  [--no-checkout] [<bad> [<good>...]] [--] [<paths>...]

git bisect (bad|new|<term-new>) [<rev>]

git bisect (good|old|<term-old>) [<rev>...]

git bisect terms [--term-good | --term-bad]

git bisect skip [(<rev>|<range>)...]

git bisect reset [<commit>]

git bisect (visualize|view)

git bisect replay <logfile>

git bisect log

git bisect run <cmd>...

git bisect help

As an example, suppose you are trying to find the commit that broke a feature that was known to work in version v2.6.13-rc2 of your project. You start a bisect session as follows:

Once you have specified at least one bad and one good commit, git bisect selects a commit in the middle of that range of history, checks it out, and outputs something similar to the following:

You should now compile the checked-out version and test it. If that version works correctly, type

If that version is broken, type

Then git bisect will respond with something like

Keep repeating the process: compile the tree, test it, and depending on whether it is good or bad run git bisect good or git bisect bad to ask for the next commit that needs testing.

Eventually there will be no more revisions left to inspect, and the command will print out a description of the first bad commit. The reference refs/bisect/bad will be left pointing at that commit.

After a bisect session, to clean up the bisection state and return to the original HEAD, issue the following command:

By default, this will return your tree to the commit that was checked out before git bisect start. (A new git bisect start will also do that, as it cleans up the old bisection state.)

With an optional argument, you can return to a different commit instead:

For example, git bisect reset bisect/bad will check out the first bad revision, while git bisect reset HEAD will leave you on the current bisection commit and avoid switching commits at all.

Sometimes you are not looking for the commit that introduced a breakage, but rather for a commit that caused a change between some other "old" state and "new" state. For example, you might be looking for the commit that introduced a particular fix. Or you might be looking for the first commit in which the source-code filenames were finally all converted to your company’s naming standard. Or whatever.

In such cases it can be very confusing to use the terms "good" and "bad" to refer to "the state before the change" and "the state after the change". So instead, you can use the terms "old" and "new", respectively, in place of "good" and "bad". (But note that you cannot mix "good" and "bad" with "old" and "new" in a single session.)

In this more general usage, you provide git bisect with a "new" commit that has some property and an "old" commit that doesn’t have that property. Each time git bisect checks out a commit, you test if that commit has the property. If it does, mark the commit as "new"; otherwise, mark it as "old". When the bisection is done, git bisect will report which commit introduced the property.

To use "old" and "new" instead of "good" and bad, you must run git bisect start without commits as argument and then run the following commands to add the commits:

to indicate that a commit was before the sought change, or

to indicate that it was after.

To get a reminder of the currently used terms, use

You can get just the old (respectively new) term with git bisect terms --term-old or git bisect terms --term-good.

If you would like to use your own terms instead of "bad"/"good" or "new"/"old", you can choose any names you like (except existing bisect subcommands like reset, start, …​) by starting the bisection using

For example, if you are looking for a commit that introduced a performance regression, you might use

Or if you are looking for the commit that fixed a bug, you might use

Then, use git bisect <term-old> and git bisect <term-new> instead of git bisect good and git bisect bad to mark commits.

To see the currently remaining suspects in gitk, issue the following command during the bisection process (the subcommand view can be used as an alternative to visualize):

If the DISPLAY environment variable is not set, git log is used instead. You can also give command-line options such as -p and --stat.

After having marked revisions as good or bad, issue the following command to show what has been done so far:

If you discover that you made a mistake in specifying the status of a revision, you can save the output of this command to a file, edit it to remove the incorrect entries, and then issue the following commands to return to a corrected state:

If, in the middle of a bisect session, you know that the suggested revision is not a good one to test (e.g. it fails to build and you know that the failure does not have anything to do with the bug you are chasing), you can manually select a nearby commit and test that one instead.

For example:

Then compile and test the chosen revision, and afterwards mark the revision as good or bad in the usual manner.

Instead of choosing a nearby commit by yourself, you can ask Git to do it for you by issuing the command:

However, if you skip a commit adjacent to the one you are looking for, Git will be unable to tell exactly which of those commits was the first bad one.

You can also skip a range of commits, instead of just one commit, using range notation. For example:

This tells the bisect process that no commit after v2.5, up to and including v2.6, should be tested.

Note that if you also want to skip the first commit of the range you would issue the command:

This tells the bisect process that the commits between v2.5 and v2.6 (inclusive) should be skipped.

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 specifying path parameters when issuing the bisect start command:

If you know beforehand more than one good commit, you can narrow the bisect space down by specifying all of the good commits immediately after the bad commit when issuing the bisect start command:

If you have a script that can tell if the current source code is good or bad, you can bisect by issuing the command:

Note that the script (my_script in the above example) should exit with code 0 if the current source code is good/old, and exit with a code between 1 and 127 (inclusive), except 125, if the current source code is bad/new.

Any other exit code will abort the bisect process. It should be noted that a program that terminates via exit(-1) leaves $? = 255, (see the exit(3) manual page), as the value is chopped with & 0377.

The special exit code 125 should be used when the current source code cannot be tested. If the script exits with this code, the current revision will be skipped (see git bisect skip above). 125 was chosen as the highest sensible value to use for this purpose, because 126 and 127 are used by POSIX shells to signal specific error status (127 is for command not found, 126 is for command found but not executable—​these details do not matter, as they are normal errors in the script, as far as bisect run is concerned).

You may often find that during a bisect session you want to have temporary modifications (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 another 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, the script can apply the patch before compiling, run the real test, and afterwards decide if the revision (possibly with the needed patch) passed the test and then rewind the tree to the pristine state. Finally the script should exit with the status of the real test to let the git bisect run command loop determine the eventual outcome of the bisect session.

$ git bisect start --term-old broken --term-new fixed

$ git bisect fixed

$ git bisect broken HEAD~10

Use git bisect to get a short usage description, and git bisect help or git bisect -h to get a long usage description.