Using a PRNG (pseudo random number generator) with a bias should be better
than alternating between 3 fixed ratios.
In repositories with many untestable commits it should prevent alternating
between areas where many commits are untestable. The bias should favor
commits that can give more information, so that the bisection process
should not loose much efficiency.
HPA suggested to use a PRNG and found that the best bias is to raise a
ratio between 0 and 1 given by the PRNG to the power 1.5.
An integer square root function is implemented to avoid including
<math.h> and linking with -lm.
A PRNG function is implemented to get the same number sequence on
different machines as suggested by "man 3 rand".
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We must save the pending commits that will be used during revision
walking and unparse them after, because we want to leave a clean
state for the next revision walking that will try to find the best
bisection point.
As we don't fork a process anymore to call "git rev-list", we need
to remove the use of GIT_TRACE to check how "git rev-list" is
called from the t6030 test that uses it.
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This patch adds some tests to check that "git bisect" works fine when
passing paths to "git bisect start" to reduce the number of
bisection steps.
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When the "bad" commit was also "skip"ped and when more than one
commit was skipped, the "filter_skipped" function would have
printed something like:
bisect_rev=<hash1>|<hash2>
(where <hash1> and <hash2> are hexadecimal sha1 hashes)
and this would have been evaled later as piping "bisect_rev=<hash1>"
into "<hash2>", which would have failed.
So this patch makes the "filter_skipped" function properly quote
what it outputs, so that it will print something like:
bisect_rev='<hash1>|<hash2>'
which will be properly evaled later. The caller was not stopping
properly because the scriptlet this function returned to be evaled
was not strung together with && and because of this, an error in
an earlier part of the output was simply ignored.
A test case is added to the test suite.
And while at it, we also initialize the VARS, FOUND and TRIED
variables, so that we protect ourselves from environment variables
the user may have with these names.
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The patch that allows "git bisect skip" to be passed a range of
commits using the "<commit1>..<commit2>" notation is flawed because
it introduces a regression when it was passed a simple rev or commit.
"git bisect skip <commit>" doesn't work any more, because <commit>
is quoted but not properly unquoted.
This patch fixes that and add tests cases to better check when it is
passed commits and range of commits.
While at it, this patch also properly quotes the non range arguments
using the "sq" function.
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
* cc/bisect:
bisect: remove "checkout_done" variable used when checking merge bases
bisect: only check merge bases when needed
bisect: test merge base if good rev is not an ancestor of bad rev
When one good revision is not an ancestor of the bad revision, the
merge bases between the good and the bad revision should be checked
to make sure that they are also good revisions.
A previous patch takes care of that, but it may check the merge bases
more often than really needed. In fact the previous patch did not try
to optimize this as much as possible because it is not so simple. So
this is the purpose of this patch.
One may think that when all the merge bases have been checked then
we can save a flag, so that we don't need to check the merge bases
again during the bisect process.
The problem is that the user may choose to checkout and test
something completely different from what the bisect process
suggested. In this case we have to check the merge bases again,
because there may be new merge bases relevant to the bisect
process.
That's why, in this patch, when we detect that the user tested
something else than what the bisect process suggested, we remove
the flag that says that we don't need to check the merge bases
again.
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Before this patch, "git bisect", when it was given some good revs that
are not ancestor of the bad rev, didn't check if the merge bases were
good. "git bisect" just supposed that the user knew what he was doing,
and that, when he said the revs were good, he knew that it meant that
all the revs in the history leading to the good revs were also
considered good.
But in pratice, the user may not know that a good rev is not an
ancestor of the bad rev, or he may not know/remember that all revs
leading to the good rev will be considered good. So he may give a good
rev that is a sibling, instead of an ancestor, of the bad rev, when in
fact there can be one rev becoming good in the branch of the good rev
(because the bug was already fixed there, for example) instead of one
rev becoming bad in the branch of the bad rev.
For example, if there is the following history:
A--B--C--D
\
E--F
and we launch "git bisect start D F" then only C and D would have been
considered as possible first bad commit before this patch. This could
invite user errors; F could be the commit that fixes the bug that exists
everywhere else.
The purpose of this patch is to detect when "git bisect" is passed
some good revs that are not ancestors of the bad rev, and then to first
ask the user to test the merge bases between the good and bad revs.
If the merge bases are good then all is fine, we can continue
bisecting. Otherwise, if one merge base is bad, it means that the
assumption that all revs leading to the good one are good too is
wrong and we error out. In the case where one merge base is skipped we
issue a warning and then continue bisecting anyway.
These checks will also catch the case where good and bad have been
mistaken. This means that we can remove the check that was done latter
on the output of "git rev-list --bisect-vars".
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
t6030-bisect-porcelain.sh relies on "ls" exiting with nonzero
status when asked to list nonexistent files. Unfortunately,
/bin/ls on Mac OS X 10.3 exits with exit code 0. So look at
its output instead.
Signed-off-by: Jonathan Nieder <jrnieder@uchicago.edu>
Acked-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When "git bisect" was first written, it was not possible to
checkout a detached HEAD. The detached feature appeared latter.
That's why before this patch the "git bisect" process used a
"bisect" branch to checkout new revisions to be tested (and also
a "new-bisect" one to check if the checkouts could work).
This patch makes "git bisect" checkout revisions to be tested on
a detached HEAD. This simplifies the code a bit.
The tests to check that "git bisect" does not start if a
"bisect" or a "new-bisect" branch exists are removed as they
are not relevant any more.
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Before this patch, when using "git bisect start" with mistaken revs
or when the checkout of the branch we want to test failed, we exited
after having written files like ".git/BISECT_START",
".git/BISECT_NAMES" and after having written "refs/bisect/bad" and
"refs/bisect/good-*" refs.
With this patch we trap all errors that can happen when writing the
new state and when we are in "bisect_next". So that we can try to
clean up everything in case of problems, using "bisect_clean_state".
This patch also contains a "bisect_write" cleanup to make it exit
on error and return 0 otherwise.
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Before this patch, when using for example:
$ git bisect start <stuff1> <stuff2>
with <stuff1> or <stuff2> that cannot be parsed as a revision, we
could leave a ".git/BISECT_START" file, from a previous
"git bisect start", alone.
This patch makes sure that it does not happen by removing the
"BISECT_START" file in "bisect_clean_state" and then always writing
it again at the end of "bisect_start".
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This patch adds some test cases to check that "git bisect start"
doesn't leave us in a bad state, especially when it fails.
These test cases show that "git bisect start" is not atomic when it
fails and leave some files like .git/BISECT_START, and in some
cases some refs, over.
The test failures should be fixed in latter commits.
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Before this patch no error was printed when "git rev-list --bisect-vars"
failed. This can happen when bad and good revs are mistaken.
This patch prints an error message on stderr that describe the likely
failure cause.
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
If a branch named "bisect" or "new-bisect" already was created in the
repo by other means than git bisect, doing a git bisect used to override
the branch without a warning. Now if the branch "bisect" or
"new-bisect" already exists, and it was not created by git bisect itself,
git bisect start fails with an appropriate error message. Additionally,
if checking out a new bisect state fails due to a merge problem, git
bisect cleans up the temporary branch "new-bisect".
The accidental override has been noticed by Andres Salomon, reported
through
http://bugs.debian.org/478647
Signed-off-by: Gerrit Pape <pape@smarden.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
* maint-1.5.4:
git-bisect: make "start", "good" and "skip" succeed or fail atomically
git-am: cope better with an empty Subject: line
Ignore leading empty lines while summarizing merges
bisect: squelch "fatal: ref HEAD not a symref" misleading message
builtin-apply: Show a more descriptive error on failure when opening a patch
Clarify documentation of git-cvsserver, particularly in relation to git-shell
Before this patch, when "git bisect start", "git bisect good" or
"git bisect skip" were called with many revisions, they could fail
after having already marked some revisions as "good", "bad" or
"skip".
This could be especilally bad for "git bisect start" because as
the file ".git/BISECT_NAMES" would not have been written, there
would have been no attempt to clear the marked revisions on a
"git bisect reset". That's because if there is no
".git/BISECT_NAMES" file, nothing is done to clean things up, as
the bisect session is not supposed to have started.
While at it, let's also create the ".git/BISECT_START" file, only
after ".git/BISECT_NAMES" as been created.
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
* maint-1.5.4:
bisect: fix bad rev checking in "git bisect good"
revision.c: make --date-order overriddable
Fix section about backdating tags in the git-tag docs
Document option --only of git commit
Documentation/git-request-pull: Fixed a typo ("send" -> "end")
It seems that "git bisect good" and "git bisect skip" have never
properly checked arguments that have been passed to them. As soon
as one of them can be parsed as a SHA1, no error or warning would
be given.
This is because 'git rev-parse --revs-only --no-flags "$@"' always
"exit 0" and outputs all the SHA1 it can found from parsing "$@".
This patch fix this by using, for each "bisect good" argument, the
same logic as for the "bisect bad" argument.
While at it, this patch teaches "bisect bad" to give a meaningfull
error message when it is passed more than one argument.
Note that if "git bisect good" or "git bisect skip" is given some
proper revs and then something that is not a proper rev, then the
first proper revs will still have been marked as "good" or "skip".
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The "-n" syntax is not supported by System V versions of
tail (which prefer "tail -1"). Unfortunately "tail -1" is
not actually POSIX. We had some of both forms in our
scripts.
Since neither form works everywhere, this patch replaces
both with the equivalent sed invocation:
sed -ne '$p'
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This error message is very confusing---it doesn't tell the user
anything about how to fix the situation. And the actual fix
for the situation ("git bisect reset") does a checkout of a
potentially random branch, (compared to what the user wants to
be on for the bisect she is starting).
The simplest way to eliminate the confusion is to just make
"git bisect start" do the cleanup itself. There's no significant
loss of safety here since we already have a general safety in
the form of the reflog.
Note: We preserve the warning for any cogito users. We do this
by switching from .git/head-name to .git/BISECT_START for the
extra state, (which is a more descriptive name anyway).
Signed-off-by: Carl Worth <cworth@cworth.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This error message is very confusing---it doesn't tell the user
anything about how to fix the situation. And the actual fix
for the situation ("git bisect reset") does a checkout of a
potentially random branch, (compared to what the user wants to
be on for the bisect she is starting).
The simplest way to eliminate the confusion is to just make
"git bisect start" do the cleanup itself. There's no significant
loss of safety here since we already have a general safety in
the form of the reflog.
Note: We preserve the warning for any cogito users. We do this
by switching from .git/head-name to .git/BISECT_START for the
extra state, (which is a more descriptive name anyway).
Signed-off-by: Carl Worth <cworth@cworth.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Instead of insisting on a symbolic ref, bisect now accepts detached
HEADs, too.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Before this patch, using "git bisect reset" when not bisecting
did a "git checkout master" for no good reason.
This also happened using "git bisect replay" when not bisecting
because "bisect_replay" starts by calling "bisect_reset".
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
If refs were ever packed in the middle of bisection, the bisect
refs were not removed from the "packed-refs" file.
This patch fixes this problem by using "git update-ref -d $ref $hash"
in "bisect_clean_state".
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This is incompatible with previous versions because an exit code
of 125 used to mark current commit as "bad". But hopefully this exit
code is not much used by test scripts or other programs. (126 and 127
are used by POSIX compliant shells to mean "found but not
executable" and "command not found", respectively.)
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When there are some "skip"ped revisions, we add the '--bisect-all'
option to "git rev-list --bisect-vars". Then we filter out the
"skip"ped revisions from the result of the rev-list command, and we
modify the "bisect_rev" var accordingly.
We don't always use "--bisect-all" because it is slower
than "--bisect-vars" or "--bisect".
When we cannot find for sure the first bad commit because of
"skip"ped commits, we print the hash of each possible first bad
commit and then we exit with code 2.
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
This uses "git-apply --whitespace=strip" to fix whitespace errors that have
crept in to our source files over time. There are a few files that need
to have trailing whitespaces (most notably, test vectors). The results
still passes the test, and build result in Documentation/ area is unchanged.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Instead of running rev-list and picking earlier lines using head/tail pipeline,
grab commit object name as we build commits. This also removes a non POSIX
use of tail with -linenum (more posixly-correct way to say it is "-n linenum")
Signed-off-by: Junio C Hamano <junkio@cox.net>
[jc: also fix 0a5280a9 that incorrectly changed the title of one test.]
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Nanako Shiraishi
Christian Couder
Junio C Hamano
Jonathan Nieder
Gerrit Pape
Jeff King
Carl Worth
Johannes Schindelin
Junio C Hamano