t4200: skip gc-rerere test on systems with non GNU date.
Quite nonstandard "date -d @11111111 +%s" does not even fail on
OpenBSD but gives the current date in "seconds since epoch"
format, which is useless for the purpose of this test. We want
to make sure that this returns exactly the same input before
proceeding.
* 'ml/gitk' (early part):
gitk: Use show-ref instead of ls-remote
Make gitk work reasonably well on Cygwin.
gitk - remove trailing whitespace from a few lines.
Not on all platforms are size_t and unsigned long equivalent.
Since I do not know how portable %z is, I play safe, and just
cast the respective variables to unsigned long.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Signed-off-by: Junio C Hamano <junkio@cox.net>
When we remove the last reflog in a directory, opendir() would
succeed and we would iterate over its dirents, expecting retval
to be initialized to zero and setting it to non-zero only upon
seeing an error. If the directory is empty, oops!, we do not
have anybody that touches retval.
The problem is because we initialize retval to errno even on
success from opendir(), which would leave the errno unmolested.
Merge branch 'master' of git://repo.or.cz/git/fastimport
* 'master' of git://repo.or.cz/git/fastimport:
Add a Tips and Tricks section to fast-import's manual.
Don't crash fast-import if the marks cannot be exported.
Dump all refs and marks during a checkpoint in fast-import.
Teach fast-import how to sit quietly in the corner.
Teach fast-import how to clear the internal branch content.
Minor timestamp related documentation corrections for fast-import.
git-clone --reference: work well with pack-ref'ed reference repository
Earlier we only used loose refs to anchor already existing
objects. When cloning from a repository that forked relatively
long time ago from the reference repository, this made the
want/have exchange by fetch-pack to do unnecessary work.
Add a Tips and Tricks section to fast-import's manual.
There has been some informative lessons learned in the gfi user
community, and these really should be written down and documented
for future generations of frontend developers.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Don't crash fast-import if the marks cannot be exported.
Apparently fast-import used to die a horrible death if we
were unable to open the marks file for output. This is
slightly less than ideal, especially now that we dump
the marks as part of the `checkpoint` command.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Dump all refs and marks during a checkpoint in fast-import.
If the frontend asks us to checkpoint (via the explicit checkpoint
command) its probably because they are afraid the current import
will crash/fail/whatever and want to make sure they can pickup from
the last checkpoint. To do that sort of recovery, we will need the
current tip of every branch and tag available at the next startup.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Teach fast-import how to sit quietly in the corner.
Often users will be running fast-import from within a larger frontend
process, and this may be a frequent periodic tool such as a future
edition of `git-svn fetch`. We don't want to bombard users with our
large stats output if they won't be interested in it, so `--quiet`
is now an option to make gfi be more silent.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Teach fast-import how to clear the internal branch content.
Some frontends may not be able to (easily) keep track of which files
are included in the branch, and which aren't. Performing this
tracking can be tedious and error prone for the frontend to do,
especially if its foreign data source cannot supply the changed
path list on a per-commit basis.
fast-import now allows a frontend to request that a branch's tree
be wiped clean (reset to the empty tree) at the start of a commit,
allowing the frontend to feed in all paths which belong on the branch.
This is ideal for a tar-file importer frontend, for example, as
the frontend just needs to reformat the tar data stream into a gfi
data stream, which may be something a few Perl regexps can take
care of. :)
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Merge branch 'master' of git://repo.or.cz/git/fastimport
* 'master' of git://repo.or.cz/git/fastimport: (81 commits)
S_IFLNK != 0140000
Don't do non-fastforward updates in fast-import.
Support RFC 2822 date parsing in fast-import.
Minor fast-import documentation corrections.
Remove unnecessary null pointer checks in fast-import.
Correct fast-import timezone documentation.
Correct minor style issue in fast-import.
Correct compiler warnings in fast-import.
Remove --branch-log from fast-import.
Initial draft of fast-import documentation.
Don't support shell-quoted refnames in fast-import.
Reduce memory usage of fast-import.
Include checkpoint command in the BNF.
Accept 'inline' file data in fast-import commit structure.
Reduce value duplication in t9300-fast-import.
Create test case for fast-import.
Support delimited data regions in fast-import.
Remove unnecessary options from fast-import.
Use fixed-size integers when writing out the index in fast-import.
Always use struct pack_header for pack header in fast-import.
...
If fast-import is being used to update an existing branch of
a repository, the user may not want to lose commits if another
process updates the same ref at the same time. For example, the
user might be using fast-import to make just one or two commits
against a live branch.
We now perform a fast-forward check during the ref updating process.
If updating a branch would cause commits in that branch to be lost,
we skip over it and display the new SHA1 to standard error.
This new default behavior can be overridden with `--force`, like
git-push and git-fetch.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Since some frontends may be working with source material where
the dates are only readily available as RFC 2822 strings, it is
more friendly if fast-import exposes Git's parse_date() function
to handle the conversion. This way the frontend doesn't need
to perform the parsing itself.
The new --date-format option to fast-import can be used by a
frontend to select which format it will supply date strings in.
The default is the standard `raw` Git format, which fast-import
has always supported. Format rfc2822 can be used to activate the
parse_date() function instead.
Because fast-import could also be useful for creating new, current
commits, the format `now` is also supported to generate the current
system timestamp. The implementation of `now` is a trivial call
to datestamp(), but is actually a whole whopping 3 lines so that
fast-import can verify the frontend really meant `now`.
As part of this change I have added validation of the `raw` date
format. Prior to this change fast-import would accept anything
in a `committer` command, even if it was seriously malformed.
Now fast-import requires the '> ' near the end of the string and
verifies the timestamp is formatted properly.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Corrected a couple of header markup lines which were shorter than the
actual header, and made the `data` commands two formats into a named
list, which matches how we document the two formats of the `M` command
within a commit.
Also tried to simplify the language about our decimal integer format;
Linus pointed out I was probably being too specific at the cost of
reduced readability.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Remove unnecessary null pointer checks in fast-import.
There is no need to check for a NULL pointer before invoking free(),
the runtime library automatically performs this check anyway and
does nothing if a NULL pointer is supplied.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Andy Parkins and Linus Torvalds both noticed that the description
of the timezone was incorrect. Its not expressed in minutes.
Its more like "hhmm", where "hh" is the number of hours and "mm"
is the number of minutes shifted from GMT/UTC.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
This allows you to set up mothership-satellite configuration
more symmetrically and naturally by allowing the globbing
wildcard refspec for git-push. On your satellite machine:
You would say "git fetch mothership" to update your tracking
branches under mothership/ to keep track of the progress on the
mothership side, and when you are done working on the satellite
machine, you would "git push mothership" to update their
tracking branches under satellite/. Corresponding configuration
on the mothership machine can be used to make "git fetch satellite"
update its tracking branch under satellite/. on the mothership.
Junio noticed that I was using a different style in fast-import
for returned pointers than the rest of Git. Before merging this
code into the main git.git tree I'd like to make it consistent,
as this style variation was not intentional.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Junio noticed these warnings/errors in fast-import when compiling
with `-Werror -ansi -pedantic`. A few changes are to reduce compiler
warnings, while one (in cmd_merge) is a bug fix.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
The --branch-log option and its associated code hasn't been used in
several months, as its not really very useful for debugging fast-import
or a frontend. I don't plan on supporting it in this state long-term,
so I'm killing it now before it gets distributed to a wider audience.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
bash: Support git-rebase -m continuation completion.
Apparently `git-rebase -m` uses a metadata directory within .git
(.git/.dotest-merge) rather than .dotest used by git-am (and
git-rebase without the -m option). This caused the completion code
to not offer --continue, --skip or --abort when working within a
`git-rebase -m` session.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
This is a first pass at the manpage for git-fast-import.
I have tried to cover the input format in extreme detail, creating a
reference which is more detailed than the BNF grammar appearing in
the header of fast-import.c. I have also covered some details about
gfi's performance and memory utilization, as well as the average
learning curve required to create a gfi frontend application (as it
is far lower than it might appear on first glance).
The documentation still lacks real example input streams, which may
turn out to be difficult to format in asciidoc due to the blank lines
which carry meaning within the format.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Don't support shell-quoted refnames in fast-import.
The current implementation of shell-style quoted refnames and
SHA-1 expressions within fast-import contains a bad memory leak.
We leak the unquoted strings used by the `from` and `merge`
commands, maybe others. Its also just muddling up the docs.
Since Git refnames cannot contain LF, and that is our delimiter
for the end of the refname, and we accept any other character
as-is, there is no reason for these strings to support quoting,
except to be nice to frontends. But frontends shouldn't be
expecting to use funny refs in Git, and its just as simple to
never quote them as it is to always pass them through the same
quoting filter as pathnames. So frontends should never quote
refs, or ref expressions.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
It used to be ls-remote on self was the only easy way to grab
the ref information. Now we have show-ref which does not
involve fork and IPC, so use it.
The gitk gui layout was completely broken on Cygwin. If gitk was started
without previous geometry in ~/.gitk, the user could drag the window sashes
to get a useable layout. However, if ~/.gitk existed, this was not possible
at all.
The fix was to rewrite makewindow, changing the toplevel containers and
the particular geometry information saved between sessions. Numerous bugs
in both the Cygwin and the Linux Tk versions make this a delicate
balancing act: the version here works in both but many subtle variants
are competely broken in one or the other environment.
Three user visible changes result:
1 - The viewer is fully functional under Cygwin.
2 - The search bar moves from the bottom to the top of the lower left
pane. This was necessary to get around a layout problem on Cygwin.
3 - The window size and position is saved and restored between sessions.
Again, this is necessary to get around a layout problem on Cygwin.
Fix longstanding mismerge of ALL_CFLAGS vs BASIC_CFLAGS
The earlier commit d7b6c3c0 (Aug 15, 2006) introduced this
mismerge when most of the CFLAGS were renamed to BASIC_CFLAGS.
Not that it matters right now, since we do not compile XS
Perl extensions which wanted non GNU subset of ALL_CFLAGS for
compilation, but we should make things consistent.
If you resize the terminal while less is waiting for input, less
will exit entirely without even showing the output. This is very
noticeable if you do something like "git diff" on a big and
cold-cache tree and git takes a few seconds to think, and then
you resize the window while it's preparing. Boom. No output AT
ALL.
The way to reproduce the problem is to do some pager operation
that takes a while in git, and resizing the window while git is
thinking about the output. Try
git diff --stat v2.6.12..
in the kernel tree to do something where it takes a while for git to start
outputting information.
(3) set up remotes/jbf-um/HEAD to point at jbf-um/master so
that later I can say "git log jbf-um"
Or I could do
$ git remote add -t 'ap/*' andy /home/andy/git.git
to make Andy's topic branches kept track of under refs/remotes/andy/ap/.
Other possible improvements I considered but haven't implemented
(hint, hint) are:
* reject wildcard letters other than a trailing '*' to the -t
parameter;
* make -m optional and when the first -t parameter does not
have the trailing '*' default to that value (so the above
example does not need to say "-m master");
* if -m is not given, and -t parameter ends with '*' (i.e. the
above defaulting did not tell us where to point HEAD at), and
if we did the fetch with -f, check if 'master' was fetched
and make HEAD point at it.
Use pretend_sha1_file() in git-blame and git-merge-recursive.
git-merge-recursive wants an null tree as the fake merge base
while producing the merge result tree. The null tree does not
have to be written out in the object store as it won't be part
of the result, and it is a prime example for using the new
pretend_sha1_file() function.
git-blame needs to register an arbitrary data to in-core index
while annotating a working tree file (or standard input), but
git-blame is a read-only application and the user of it could
even lack the privilege to write into the object store; it is
another good example for pretend_sha1_file().
The new interface allows an application to temporarily hash a
small number of objects and pretend that they are available in
the object store without actually writing them.
git-blame: no rev means start from the working tree file.
Warning: this changes the semantics.
This makes "git blame" without any positive rev to start digging
from the working tree copy, which is made into a fake commit
whose sole parent is the HEAD.
It also adds --contents <file> option to pretend as if the
working tree copy has the contents of the named file. You can
use '-' to make the command read from the standard input.
If you want the command to start annotating from the HEAD
commit, you need to explicitly give HEAD parameter.
git-blame: an Emacs minor mode to view file with git-blame output.
Here's another version of git-blame.el that automatically tries to
create a sensible list of colors to use for both light and dark
backgrounds. Plus a few minor fixes.
To use:
1) Load into emacs: M-x load-file RET git-blame.el RET
2) Open a git-controlled file
3) Blame: M-x git-blame-mode
hg-to-git.py is able to convert a Mercurial repository into a git one,
and preserves the branches in the process (unlike tailor)
hg-to-git.py can probably be greatly improved (it's a rather crude
combination of shell and python) but it does already work quite well for
me. Features:
- supports incremental conversion
(for keeping a git repo in sync with a hg one)
- supports hg branches
- converts hg tags
Signed-off-by: Stelian Pop <stelian@popies.net> Signed-off-by: Junio C Hamano <junkio@cox.net>
If the repository directory name is in non-ascii, $project needs to be
converted from perl internal to utf-8 because it will be used as
title, page path, and snapshot filename.
We now offer completion support for git-bisect's subcommands,
as well as ref name completion on the good/bad/reset subcommands.
This should make interacting with git-bisect slightly easier on
the fingers.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
We've recently added --add as an argument to git-config, but I
missed putting it into the earlier round of git-config updates
within the bash completion.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
Don't offer resolve as a possible subcommand completion. If you
read the top of the script, there is a big warning about how it
will go away soon in the near future. People should not be using it.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
Apparently nobody really makes use of git-diff-stages, as nobody
has complained that it is not supported by the git-diff frontend.
Since its likely this will go away in the future, we should not
offer it as a possible subcommand completion.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
I just realized I did not support ref name completion for git-cherry.
This tool is just too useful to contributors who submit patches
upstream by email; completion support for it is very handy.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
Show an example of deleting commits with git-rebase.
This particular use of git-rebase to remove a single commit or a
range of commits from the history of a branch recently came up on
the mailing list. Documenting the example should help other users
arrive at the same solution on their own.
It also was not obvious to the newcomer that git-rebase is able to
accept any commit for --onto <newbase> and <upstream>. We should
at least minimally document this, as much of the language in
git-rebase's manpage refers to 'branch' rather than 'committish'.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
We at least know that the test as written has a problem in an
environment where "touch '$p'; ls | fgrep '$p'" fails, and have
a clear understand why it fails.
This tests if the filesystem has that particular issue we know "git
add" has a problem with, and skips the test in such an environment.
This way, we might catch issues "git add" might have in other environments.
Some structs are allocated rather frequently, but were using integer
types which were far larger than required to actually store their
full value range.
As packfiles are limited to 4 GiB we don't need more than 32 bits to
store the offset of an object within that packfile, an `unsigned long`
on a 64 bit system is likely a 64 bit unsigned value. Saving 4 bytes
per object on a 64 bit system can add up fast on any sizable import.
As atom strings are strictly single components in a path name these
are probably limited to just 255 bytes by the underlying OS. Going
to that short of a string is probably too restrictive, but certainly
`unsigned int` is far too large for their lengths. `unsigned short`
is a reasonable limit.
Modes within a tree really only need two bytes to store their whole
value; using `unsigned int` here is vast overkill. Saving 4 bytes
per file entry in an active branch can add up quickly on a project
with a large number of files.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Back when I committed this, it seemed to be a good idea. People
who always use remote tracking branches can optionally use the
local name they happen to use to specify what to merge, which meant
that I did not have to teach them why we use the name at the remote
side every time they are confused.
But allowing it seems to break other people's scripts. The real
solution is not to allow more ways to express the same thing, but
to educate people to use the right syntax.
* np/dreflog:
show-branch -g: default to the current branch.
Let git-checkout always drop any detached head
Enable HEAD@{...} and make it independent from the current branch
scan reflogs independently from refs
add reflog when moving HEAD to a new branch
create_symref(): do not assume pathname from git_path() persists long enough
add logref support to git-symbolic-ref
move create_symref() past log_ref_write()
add reflog entries for HEAD when detached
enable separate reflog for HEAD
lock_ref_sha1_basic(): remember the original name of a ref when resolving it
make reflog filename independent from struct ref_lock
git-clone --reference: saner handling of borrowed symrefs.
When using --reference to borrow objects from a neighbouring
repository while cloning, we copy the entire set of refs under
temporary "refs/reference-tmp/refs" space and set up the object
alternates. However, a textual symref copied this way would not
point at the right place, and causes later steps to emit error
messages (which is harmless but still alarming). This is most
visible when using a clone created with the separate-remote
layout as a reference, because such a repository would have
refs/remotes/origin/HEAD with 'ref: refs/remotes/origin/master'
as its contents.
Although we do not create symbolic-link based refs anymore, they
have the same problem because they are always supposed to be
relative to refs/ hierarchy (we dereference by hand, so it only
is good for HEAD and nothing else).
In either case, the solution is simply to remove them after
copying under refs/reference-tmp; if a symref points at a true
ref, that true ref itself is enough to ensure that objects
reachable from it do not needlessly get fetched.
format-patch/log/whatchanged all take --not and --all as options
to the internal revlist process. So these should be supported
as possible completions.
gitk takes anything rev-list/log/whatchanged takes, so we should
use complete_revlist to handle its options.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
Because our use of -o nospace prevents bash from adding a trailing space
when a completion is unique and has been fully completed, we need to
perform this addition on our own. This (large) change converts all
existing uses of compgen to our wrapper __gitcomp which attempts to
handle this by tacking a trailing space onto the end of each offered
option.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
In many cases we know a completion will be unique, but we've disabled
bash's automatic space addition (-o nospace) so we need to do it
ourselves when necessary.
This change adds additional support for new configuration options
added in 1.5.0, as well as some extended completion support for
the color.* family of options.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
Most of these commands are not ones you want to invoke from the
command line on a frequent basis, or have been renamed in 1.5.0 to
more friendly versions, but the old names are being left behind to
support existing scripts in the wild.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
bash: Add space after unique command name is completed.
Because we use the nospace option for our completion function for
the main 'git' wrapper bash won't automatically add a space after a
unique completion has been made by the user. This has been pointed
out in the past by Linus Torvalds as an undesired behavior. I agree.
We have to use the nospace option to ensure path completion for
a command such as `git show` works properly, but that breaks the
common case of getting the space for a unique completion. So now we
set IFS=$'\n' (linefeed) and add a trailing space to every possible
completion option. This causes bash to insert the space when the
completion is unique.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
The new --interactive mode of git-add can be very useful, so users
will probably want to have completion for it.
Likewise the new git-add--interactive executable is actually a
plumbing command. Its invoked by `git add --interactive` and is
not intended to be invoked directly by the user. Therefore we
should hide it from the list of available Git commands.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
Now that git-show is capable of displaying any file content from any
revision and is the approved Porcelain-ish level method of doing so,
cat-file should no longer be classified as a user-level utility by
the bash completion package.
I'm also classifying the new git-reflog command as plumbing for the
time being as there are no subcommands which are really useful to
the end-user. git-gc already invokes `git reflog expire --all`,
which makes it rather unnecessary for the user to invoke it directly.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
bash: Remove short option completions for branch/checkout/diff.
The short options (-l, -f, -d) for git-branch are rather silly to
include in the completion generation as these options must be fully
typed out by the user and most users already know what the options
are anyway, so including them in the suggested completions does
not offer huge value. (The same goes for git-checkout and git-diff.)
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
Now we have a separate reflog on HEAD, show-branch -g without an explicit
parameter defaults to the current branch, or HEAD when it is detached
from branches.
We used to refuse leaving a detached HEAD when it wasn't matching an
existing ref so not to lose any commit that might have been performed
while not on any branch (unless -f was provided).
But this protection was completely bogus since it was still possible
to move to HEAD^ while still remaining detached but losing the last
commit anyway if there was one.
Now that we have a proper reflog for HEAD it is best to simply remove
that bogus (and admitedly annoying) protection and simply display the
last HEAD position instead. If one wants to recover a lost detached
state then it can be retrieved from the HEAD reflog.
Signed-off-by: Nicolas Pitre <nico@cam.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
This is to resolve conflicts early in preparation for possible
inclusion of "reflog on detached HEAD" series by Nico, as having
it in 1.5.0 would really help us remove confusion between
detached and attached states.
Its really nice to be able to run a test with -v and automatically
see the "debugging" dump from merge-recursive, especially if we
are actually trying to debug merge-recursive.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
My earlier fix (8371234e) to delete renamed tracked files from the
working directory also caused merge-recursive to delete untracked
files that were in the working directory.
The problem here is merge-recursive is deleting the working directory
file without regard for which branch it was associated with. What we
really want to do during a merge is to only delete files that were
renamed by the branch we are merging into the current branch,
and that are still tracked by the current branch. These files
definitely don't belong in the working directory anymore.
Anything else is either a merge conflict (already handled in other
parts of the code) or a file that is untracked by the current branch
and thus is not even participating in the merge. Its this latter
class that must be left alone.
For this fix to work we are now assuming that the first non-base
argument passed to git-merge-recursive always corresponds to the
working directory. This is already true for all in-tree callers
of merge-recursive. This assumption is also supported by the
long time usage message of "<base> ... -- <head> <remote>", where
"<head>" is implied to be HEAD, which is generally assumed to be
the current tree-ish.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
Jakub Narebski pointed out the positional notation in git-remote's
documentation was very confusing, especially now that we have 3
supported subcommands. Instead of referring to subcommands by
position, refer to them by name.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
Even when --unified=0 is given, the main loop to show the
combined textual diff needs to handle a line that is unchanged
but has lines that were deleted relative to a parent before it
(because that is where the lost lines hang). However, such a
line should not be emitted in the final output.
Why is it bad to rewind a branch that has already been pushed out?
I was reading the tutorial and noticed that we say this:
Also, don't use "git reset" on a publicly-visible branch that
other developers pull from, as git will be confused by history
that disappears in this way.
I do not think this is a good explanation. For example, if we
do this:
(1) I build a series and push it out.
---o---o---o---j
(2) Alice clones from me, and builds two commits on top of it.
---o---o---o---j---a---a
(3) I rewind one and build a few, and push them out.
---o---o---o...j
\
h---h---h---h
(4) Alice pulls from me again:
---o---o---o---j---a---a---*
\ /
h---h---h---h
Contrary to the description, git will happily have Alice merge
between the two branches, and never gets confused.
Maybe I did not want to have 'j' because it was an incomplete
solution to some problem, and Alice may have fixed it up with
her changes, while I abandoned that approach I started with 'j',
and worked on something completely unrelated in the four 'h'
commits. In such a case, the merge Alice would make would be
very sensible, and after she makes the merge if I pull from her,
the world will be perfect. I started something with 'j' and
dropped the ball, Alice picked it up and perfected it while I
went on to work on something else with 'h'. This would be a
perfect example of distributed parallel collaboration. There is
nothing confused about it.
The case the rewinding becomes problematic is if the work done
in 'h' tries to solve the same problem as 'j' tried to solve in
a different way. Then the merge forced on Alice would make her
pick between my previous attempt with her fixups (j+a) and my
second attempt (h). If 'a' commits were to fix up what 'j'
started, presumably Alice already studied and knows enough about
the problem so she should be able to make an informed decision
to pick between what 'j+a' and 'h' do.
A lot worse case is if Alice's work is not at all related to
what 'j' wanted to do (she did not mean to pick up from where I
left off -- she just wanted to work on something different).
Then she would not be familiar enough with what 'j' and 'h'
tried to achieve, and I'd be forcing her to pick between the
two. Of course if she can make the right decision, then again
that is a perfect example of distributed collaboration, but that
does not change the fact that I'd be forcing her to clean up my
mess.
This allows git-cherry-pick and git-revert to properly identify
themselves in the resulting reflog entries. Earlier they were
recorded as what git-commit has done.
Even when -l is not given from the command line, the repository
may have the configuration variable core.logallrefupdates set,
or an old-timer might have done ": >.git/logs/refs/heads/new"
before running "git branch new". In these cases, the code gave
an uninitialized msg[] from the stack to be written out as the
reflog message.
This also passes a different message when '-f' option is used.
Saying "git branch -f branch some-commit" is a moral equilvalent
of doing "git-reset some-commit" while on the branch.
Currently, the search for all reflogs depends on the existence of
corresponding refs under the .git/refs/ directory. Let's scan the
.git/logs/ directory directly instead.
Signed-off-by: Nicolas Pitre <nico@cam.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
Don't leak file descriptors from unavailable pack files.
If open_packed_git failed it may have been because the packfile
actually exists and is readable, but some sort of verification
did not pass. In this case open_packed_git left pack_fd filled
in, as the file descriptor is valid. We don't want to leak the
file descriptor, nor do we want to allow someone in the future
to use this packed_git.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
Don't find objects in packs which aren't available anymore.
Matthias Lederhofer identified a race condition where a Git reader
process was able to locate an object in a packed_git index, but
was then preempted while a `git repack -a -d` ran and completed.
By the time the reader was able to seek in the packfile to get the
object data, the packfile no longer existed on disk.
In this particular case the reader process did not attempt to
open the packfile before it was deleted, so it did not already
have the pack_fd field popuplated. With the packfile itself gone,
there was no way for the reader to open it and fetch the data.
I'm fixing the race condition by teaching find_pack_entry to ignore
a packed_git whose packfile is not currently open and which cannot
be opened. If none of the currently known packs can supply the
object, we will return 0 and the caller will decide the object is
not available. If this is the first attempt at finding an object,
the caller will reprepare_packed_git and try again. If it was
the second attempt, the caller will typically return NULL back,
and an error message about a missing object will be reported.
This patch does not address the situation of a reader which is
being starved out by a tight sequence of `git repack -a -d` runs.
In this particular case the reader will try twice, probably fail
both times, and declare the object in question cannot be found.
As it is highly unlikely that a real world `git repack -a -d` can
complete faster than a reader can open a packfile, so I don't think
this is a huge concern.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>