ls-files: honour per-directory ignore file from higher directories.
When git-ls-files -o --exclude-per-directory=.gitignore is run
from a subdirectory, it did not read from .gitignore from its
parent directory. Reading from them makes output from these two
commands consistent:
It tried to "restore" GIT_AUTHOR_EMAIL environment variable but
the variable started out as unset, so ended up setting it to an
empty string. This is now caught as an error.
This is not really part of the proposed updates for CE_VALID,
but with this change, ls-files -t shows CE_VALID paths with
lowercase tag letters instead of the usual uppercase. Useful
for checking out what is going on.
"Assume unchanged" git: do not set CE_VALID with --refresh
When working with automatic assume-unchanged mode using
core.ignorestat, setting CE_VALID after --refresh makes things
more cumbersome to use. Consider this scenario:
(1) the working tree is on a filesystem with slow lstat(2).
The user sets core.ignorestat = true.
(2) "git checkout" to switch to a different branch (or initial
checkout) updates all paths and the index starts out with
"all clean".
(3) The user knows she wants to edit certain paths. She uses
update-index --no-assume-unchanged (we could call it --edit;
the name is inmaterial) to mark these paths and starts
editing.
(4) After editing half of the paths marked to be edited, she
runs "git status". This runs "update-index --refresh" to
reduce the false hits from diff-files.
(5) Now the other half of the paths, since she has not changed
them, are found to match the index, and CE_VALID is set on
them again.
For this reason, this commit makes update-index --refresh not to
set CE_VALID even after the path without CE_VALID are verified
to be up to date. The user still can run --really-refresh to
force lstat() to match the index entries to the reality.
This is a workaround for filesystems that do not have lstat()
that is quick enough for the index mechanism to take advantage
of. On the paths marked as "assumed to be unchanged", the user
needs to explicitly use update-index to register the object name
to be in the next commit.
You can use two new options to update-index to set and reset the
CE_VALID bit:
These forms manipulate only the CE_VALID bit; it does not change
the object name recorded in the index file. Nor they add a new
entry to the index.
When the configuration variable "core.ignorestat = true" is set,
the index entries are marked with CE_VALID bit automatically
after:
- update-index to explicitly register the current object name to the
index file.
- when update-index --refresh finds the path to be up-to-date.
- when tools like read-tree -u and apply --index update the working
tree file and register the current object name to the index file.
The flag is dropped upon read-tree that does not check out the index
entry. This happens regardless of the core.ignorestat settings.
Index entries marked with CE_VALID bit are assumed to be
unchanged most of the time. However, there are cases that
CE_VALID bit is ignored for the sake of safety and usability:
- while "git-read-tree -m" or git-apply need to make sure
that the paths involved in the merge do not have local
modifications. This sacrifices performance for safety.
- when git-checkout-index -f -q -u -a tries to see if it needs
to checkout the paths. Otherwise you can never check
anything out ;-).
- when git-update-index --really-refresh (a new flag) tries to
see if the index entry is up to date. You can start with
everything marked as CE_VALID and run this once to drop
CE_VALID bit for paths that are modified.
Most notably, "update-index --refresh" honours CE_VALID and does
not actively stat, so after you modified a file in the working
tree, update-index --refresh would not notice until you tell the
index about it with "git-update-index path" or "git-update-index
--no-assume-unchanged path".
This version is not expected to be perfect. I think diff
between index and/or tree and working files may need some
adjustment, and there probably needs other cases we should
automatically unmark paths that are marked to be CE_VALID.
But the basics seem to work, and ready to be tested by people
who asked for this feature.
http-fetch: Abort requests for objects which arrived in packs
In fetch_object, there's a call to release an object request if the
object mysteriously arrived, say in a pack. Unfortunately, the fetch
attempt for this object might already be in progress, and we'll leak the
descriptor. Instead, try to tidy away the request.
Signed-off-by: Mark Wooding <mdw@distorted.org.uk> Signed-off-by: Junio C Hamano <junkio@cox.net>
format-patch: Remove last vestiges of --mbox option
Don't mention it in docs or --help output.
Remove mbox, date and author variables from git-format-patch.sh.
Use DESCRIPTION text from man-page to update LONG_USAGE output. It's
a bit silly to have two texts saying the same thing in different words,
and I'm too lazy to update both.
Signed-off-by: Andreas Ericsson <ae@op5.se> Signed-off-by: Junio C Hamano <junkio@cox.net>
Introduce --only flag to allow the new "partial commit"
semantics when paths are specified. The default is still the
traditional --include semantics. Once peoples' fingers and
scripts that want the traditional behaviour are updated to
explicitly say --include, we could change it to either default
to --only, or refuse to operate without either --only/--include
when paths are specified.
This also fixes a couple of bugs in the previous round. Namely:
- forgot to save/restore index in some cases.
- forgot to use the temporary index to show status when '--only
paths...' semantics was used.
- --author did not take precedence when reusing an existing
commit.
- "git commit" without _any_ parameter keeps the traditional
behaviour. It commits the current index.
We commit the whole index even when this form is run from a
subdirectory.
- "git commit --include paths..." (or "git commit -i paths...")
is equivalent to:
git update-index --remove paths...
git commit
- "git commit paths..." acquires a new semantics. This is an
incompatible change that needs user training, which I am
still a bit reluctant to swallow, but enough people seem to
have complained that it is confusing to them. It
1. refuses to run if $GIT_DIR/MERGE_HEAD exists, and reminds
trained git users that the traditional semantics now needs
-i flag.
2. refuses to run if named paths... are different in HEAD and
the index (ditto about reminding). Added paths are OK.
3. reads HEAD commit into a temporary index file.
4. updates named paths... from the working tree in this
temporary index.
5. does the same updates of the paths... from the working
tree to the real index.
6. makes a commit using the temporary index that has the
current HEAD as the parent, and updates the HEAD with this
new commit.
- "git commit --all" can run from a subdirectory, but it updates
the index with all the modified files and does a whole tree
commit.
- In all cases, when the command decides not to create a new
commit, the index is left as it was before the command is
run. This means that the two "git diff" in the following
sequence:
$ git diff
$ git commit -a
$ git diff
would show the same diff if you abort the commit process by
making the commit log message empty.
This commit also introduces much requested --author option.
$ git commit --author 'A U Thor <author@example.com>'
In a workflow that employs relatively long lived topic branches,
the developer sometimes needs to resolve the same conflict over
and over again until the topic branches are done (either merged
to the "release" branch, or sent out and accepted upstream).
This commit introduces a new command, "git rerere", to help this
process by recording the conflicted automerge results and
corresponding hand-resolve results on the initial manual merge,
and later by noticing the same conflicted automerge and applying
the previously recorded hand resolution using three-way merge.
If the first part uses quoted-printable to protect iso8859-1
name in the commit log, and the second part was plain ascii text
patchfile without even Content-Transfer-Encoding subheader, we
incorrectly tried to decode the patch as quoted printable.
Docs: move git url and remotes text to separate sections
The sections on git urls and remotes files in the git-fetch,
git-pull, and git-push manpages seem long enough to be worth a
manpage section of their own.
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu> Signed-off-by: Junio C Hamano <junkio@cox.net>
The push and pull man pages include a bunch of shared text from
pull-fetch-param.txt. This simplifies maintenance somewhat, but
there's actually quite a bit of text that applies only to one or the
other.
So, separate out the push- and pull/fetch-specific text into
pull-fetch-param.txt and git-push.txt, then include the largest chunk
of common stuff (the description of protocols and url's) from
urls.txt. That cuts some irrelevant stuff from the man pages without
making us duplicate too much.
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu> Signed-off-by: Junio C Hamano <junkio@cox.net>
combine-diff: do not punt on removed or added files.
When we remove a file, the parents' contents are all removed so
it is not that interesting to show all of them, but the fact it
was removed when all parents had it *is* unusual. When we add a
file, similarly the fact it was added when no parent wanted it
*is* unusual, and in addition the result matters, so show it.
gitk: Use git-diff-tree --cc for showing the diffs for merges
This replaces a lot of code that used the result from several 2-way
diffs to generate a combined diff for a merge. Now we just use
git-diff-tree --cc and colorize the output a bit, which is a lot
simpler, and has the enormous advantage that if the diff doesn't
show quite what someone thinks it should show, I can deflect the
blame to someone else. :)
Apparently this simplifies things for the parser/compiler and makes
it go slightly faster (since without the braces, it potentially has
to do two levels of substitutions rather than one).
* jc/daemon:
daemon: extend user-relative path notation.
daemon: Set SO_REUSEADDR on listening sockets.
daemon: do not forbid user relative paths unconditionally under --base-path
* mw/http:
http-fetch: Tidy control flow in process_alternate_response
http: Turn on verbose Curl messages if GIT_CURL_VERBOSE set in environment
http-fetch: Fix message reporting rename of object file.
http-fetch: Fix object list corruption in fill_active_slots().
git-diff-tree --stdin ignored second and subsequent parents when
fed git-rev-list --parents output. Update diff_tree_commit()
function to take a commit object, and pass a fabricated commit
object after grafting the fake parents from diff_tree_stdin().
Earlier we had a workaround to avoid misspelled revision name to
be taken as a filename when "--no-revs --no-flags" are in
effect. This cleans up the logic.
Earlier, we made --base-path to automatically forbid
user-relative paths, which was probably a mistake. This
introduces --user-path (or --user-path=path) option to control
the use of user-relative paths independently. The latter form
of the option can be used to restrict accesses to a part of each
user's home directory, similar to "public_html" some webservers
supports.
If we're invoked with --user-path=FOO option, then a URL of the
form git://~USER/PATH/... resolves to the path HOME/FOO/PATH/...,
where HOME is USER's home directory.
[jc: This is much reworked by me so bugs are mine, but the
original patch was done by Mark Wooding.]
http-fetch: Fix message reporting rename of object file.
move_temp_to_file returns 0 or -1. This is not a good thing to pass to
strerror(3). Fortunately, someone already reported the error, so don't
worry too much.
Signed-off-by: Mark Wooding <mdw@distorted.org.uk> Signed-off-by: Junio C Hamano <junkio@cox.net>
http-fetch: Fix object list corruption in fill_active_slots().
In fill_active_slots() -- if we find an object which has already arrived,
say as part of a pack, /don't/ remove it from the list. It's already been
prefetched and someone will ask for it later. Just label it as done and
carry blithely on. (As it was, the code would dereference a freed object
to continue through the list anyway.)
Signed-off-by: Mark Wooding <mdw@distorted.org.uk> Signed-off-by: Junio C Hamano <junkio@cox.net>
This is essentially 'git whatchanged -n1 --always --cc "$@"'.
Just like whatchanged takes default flags from
whatchanged.difftree configuration, this uses show.difftree
configuration.
It could be made later to show unmerged state nicer than the
default as we did for diff-files later, but this would suffice
for now. We would like to make --cc the default for 'git diff'.
It _might_ make sense for certain users like gitk and gitview if
we had a single tool that gives --pretty and its diff even if
the diff is empty. Having said that, the flag --cc -m is too
specific. If some uses want to see the commit log even for an
empty diff, that flag should not be something only --cc honors.
Using "--verify" together with "--no-flags" makes perfect sense, but
git-rev-parse would complain about it when it saw a flag, even though it
would never actually use/output that flag.
This fixes it.
Signed-off-by: Linus Torvalds <torvalds@osdl.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
As far as I can see, create_subdirectories() in apply.c just
duplicates the functionality of safe_create_leading_directories() from
sha1_file.c. The former has a warm, fuzzy const parameter, but that's
not important.
The potential problem with EEXIST and creating directories should
never occur here, but will be removed by future
safe_create_leading_directories() changes. Other uses of EEXIST in
apply.c should be fine barring intentionally malicious behavior.
Signed-off-by: Jason Riedy <ejr@cs.berkeley.edu> Signed-off-by: Junio C Hamano <junkio@cox.net>
daemon: Provide missing argument for logerror() call.
Could cause a crash if --base-path set. Unlikely to be a security the
concern: message doesn't go to the client, so we can't leak anything
(except by dumping core), and we've already forked, so it's not a denial
of service.
Signed-off-by: Mark Wooding <mdw@distorted.org.uk> Signed-off-by: Junio C Hamano <junkio@cox.net>
get_sha1_1: allow octopus^12 to be properly parsed.
We probably thought anybody who does more than 9 parents in an
Octopus is insane when this was initially done, but there is no
inherent reason to limit the number of independent topic
branches that happen to mature at the same time.
Our commit-tree allows up to 16 already, so at least we should
prepare to handle what we can produce, if only to be consistent.
combine-diff: finishing touches to git-diff-tree --cc
This updates the output format to make administrative lines more
consistent with the traditional diffs.
The "index" line shows blob object names from each parents
(separated by commas), double dots and the object name of the
resulting blob.
The hunk header line begins with N+1 '@' characters for N-way
diff, the line number L of the first line in the hunk and line
count C from the parent in "-L,C" format for each parents and
then the line number of the first line in the hunk and line
count from the resulting file in "+L,C" format, and finally
N+1 '@' characters (earlier versions had the line numbers from
the resulting file at the beginning).
The flag on the surviving lines meant "this parent is not
different" while the parent_map flag on the lost lines meant
"this parent is different", which was confusing. So swap the
meaning of on-bit in the flag. Also more heavily comment the
code.
Provide a more meaningful initial "From " line when using --compose in git-send-email.
git-send-email, when used with --compose, provided the user with a mbox-format
file to edit. Some users, however, were confused by the leading, blank, "From
" line, so this change puts the value that will appear on the From: line of the
actual email on this line, along with a note that the line is ignored.
Signed-off-by: Ryan Anderson <ryan@michonline.com> Signed-off-by: Junio C Hamano <junkio@cox.net>
Documentation: git-diff-tree --cc also omits empty commits
A misguided attempt to show logs at all time was inserted only to
the documentation of this flag. Worse yet, it was not even implemented,
causing more confusion. Drop it.
We might want to have an option to show --pretty even when there is no
diff output, but that is applicable to all forms of diff, not just --cc.
The code misplaced a raw hunk that consists of solely deleted
lines by one line. This showed e.g. Len's 12-way octopus
(9fdb62af in the linux-2.6), kernel/power/disk.c, hunk starting
at line 95, incorrectly.
The earlier change implemented "only two version" check but
without checking if the change rewrites from all the parents.
This implements a check to make sure that a change introduced
by the merge from all the parents is caught to be interesting.
Earlier logic was discarding hunks that has difference from only
one parent or the same difference from all but one parent. This
changes it to check if the differences on all lines are from the
same sets of parents. This discards more uninteresting hunks
and seems to match expectations more naturally.
When dealing with an insanely large Octopus, it is possible to
optimize by noticing that more than one parents have the same
blob and avoid running diff between a parent and the merge
result by reusing an earlier result.
So far, e.g. git-update-index --refresh was basically uninterruptable
by ctrl-c, since it hooked the SIGINT handler, but that handler would
only unlink the lockfile but not actually quit. This makes it propagate
the signal to the default handler.
Note that I expected it to work without resetting the signal handler to
SIG_DFL, but without that it ended in an infinite loop of tgkill()s -
is my glibc violating SUS or what?
Signed-off-by: Petr Baudis <pasky@suse.cz> Signed-off-by: Junio C Hamano <junkio@cox.net>
Showing the same parent more than once for a commit does not
make much sense downstream, so stop it.
This can happen with an incorrectly made merge commit that
merges the same parent twice, but can happen in an otherwise
sane development history while squishing the history by taking
into account only commits that touch specified paths.
where 5 independent development tracks, only two of which have
changes in the specified paths since they forked. The last
change for the other three development tracks was done by the
same commit before they forked, and we were showing that three
times.
rev-list: allow -n<n> as shorthand for --max-count=<n>
Both -n<n> and -n <n> are supported. POSIX versions of head(1) and
tail(1) allow their line limits to be parsed this way. I find
--max-count to be a commonly used option, and also similar in spirit to
head/tail, so I decided to make life easier on my worn out (and lazy :)
fingers with this patch.
Signed-off-by: Eric Wong <normalperson@yhbt.net> Signed-off-by: Junio C Hamano <junkio@cox.net>
Here's some changes to the cvs-migration.txt. As usual, in my attempt
to make things clearer someone may have found I've made them less so, or
I may have just gotten something wrong; so any review is welcomed.
I can break up this sort of thing into smaller steps if preferred, the
monolothic patch is just a bit simpler for me for this sort of
thing.
I moved the material describing shared repository management from
core-tutorial.txt to cvs-migration.txt, where it seems more appropriate,
and combined two sections to eliminate some redundancy.
I also revised the earlier sections of cvs-migration.txt, mainly trying
to make it more concise.
I've left the last section of cvs-migration.txt (on CVS annotate
alternatives) alone for now.
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu> Signed-off-by: Junio C Hamano <junkio@cox.net>
[PATCH] pre-commit sample hook: do not barf on the initial import
The example hook barfs on the initial import. Ideally it should
produce a diff from an empty tree, but for now let's stop at
squelching the bogus error message. Often an initial import
involves tons of badly formatted files from foreign SCM, so not
complaining about them like this patch does might actually be a
better idea than enforcing the "Perfect Patch" format on them.