A type-change diff is always split into a patch to delete old,
immediately followed by a patch to create new. check_patch()
routine noticed that the path to be created already exists in
the working tree and/or in the index when looking at the
creation patch and mistakenly thought it to be an error.
When creating a new file where a directory used to be (or the user had
an empty directory) the code did not check the result from lstat() closely
enough, and mistakenly thought the path already existed in the working tree.
This does not fix the problem where you have a patch that creates a file
at "foo" and removes a file at "foo/bar" (which presumably is the last file
in "foo/" directory in the original). For that, we would need to restructure
write_out_results() loop.
typechange tests for git apply (currently failing)
I've found that git apply is incapable of handling patches
involving object type changes to the same path.
Of course git itself is perfectly capable of making commits that
generate these changes, as it only tracks trees states. It's
just that the diffs between them are less useful if they can't
be applied.
Some of these are rare, but I've hit one of them (file becoming
a symlink) recently in real-world usage, and was inspired to
find more potential breakages :)
I'm not sure when I'll have time to fix these myself and I'm not
very familiar with the apply code. So if someone could get
some or all of these cases working, they would be my hero :)
Some of these are what I would refer to as corner-cases from
hell. Most (if not all) other systems fail some of these. In
fact, they aren't even capable of representing most of these
changes in their histories; much less being able to handle
patches to that effect.
Signed-off-by: Eric Wong <normalperson@yhbt.net> Signed-off-by: Junio C Hamano <junkio@cox.net>
The core function used in show-branch, join_revs(), was supposed
to be exactly the same algorithm as merge_bases(), except that
it was a version enhanced for use with more than two heads.
However, it needed to mark and keep a list of all the commits it
has seen, because it needed them for its semi-graphical output.
The function to implement this list, mark_seen(), stupidly used
insert_by_date(), when it did not need to keep the list sorted
during its processing. This made "show-branch --merge-base"
more than 20x slower compared to "merge-base --all" in some
cases (e.g. between b5032a5 and 48ce8b0 in the Linux 2.6 kernel
archive). The performance of "show-branch --independent"
suffered from the same reason.
This patch sorts the resulting list after the list traversal
just once to fix these problems.
Documentation/urls.txt: Use substitution to escape square brackets
This changes "[user@]" to use {startsb} and {endsb} to insert [ and ],
similar to how {caret} is used in git-rev-parse.txt.
[jc: Removed a well-intentioned comment that broke the final
formatting from the original patch. While we are at it,
updated the paragraph that claims to be equivalent to the
section that was updated earlier without making matching
changes.]
Signed-off-by: Jonas Fonseca <fonseca@diku.dk> Signed-off-by: Junio C Hamano <junkio@cox.net>
Fix "git-fetch --tags" exit status when nothing has been changed
After commit 55b7835e1b81a6debc7648149d2b8a4c5c64ddba git-fetch --tags
exits with status 1 when no tags have been changed, which breaks calling
git-fetch from scripts.
Signed-off-by: Sergey Vlasov <vsu@altlinux.ru> Signed-off-by: Junio C Hamano <junkio@cox.net>
Use .git/info/exclude in the example in git-ls-files.txt,
instead of .git/ignore, and update the list of commands looking
at .git/info/exclude in repository-layout.txt.
Signed-off-by: Matthias Lederhofer <matled@gmx.net> Signed-off-by: Junio C Hamano <junkio@cox.net>
format-patch: Generate a newline between the subject header and the message body
format-patch previously didn't generate a newline after a subject. This
caused the diffstat to not be displayed in messages with only one line
for the commit message.
This patch fixes this by adding a newline after the headers if a body
hasn't been added. Signed-off-by: Robert Shearman <rob@codeweavers.com> Signed-off-by: Junio C Hamano <junkio@cox.net>
This changes one test commit in the sequence to have more than
one lines of commit log. A few output formats (--pretty=email
aka format-patch and --pretty=oneline) need to behave
differently on single and multi-line log, and this change will
help catching breakages.
Display help for Git mode after pressing `h' or `?' in *git-status*
Add bindings for "h" and "?" in git-status-mode to display help about the mode,
including keymap via (describe-function 'git-status-mode), like in PCL-CVS.
Signed-off-by: Jakub Narebski <jnareb@gmail.com> Signed-off-by: Junio C Hamano <junkio@cox.net>
Remove TYPE_* constant macros and use object_type enums consistently.
This updates the type-enumeration constants introduced to reduce
the memory footprint of "struct object" to match the type bits
already used in the packfile format, by removing the former
(i.e. TYPE_* constant macros) and using the latter (i.e. enum
object_type) throughout the code for consistency.
Eventually we can stop passing around the "type strings"
entirely, and this will help - no confusion about two different
integer enumeration.
Signed-off-by: Linus Torvalds <torvalds@osdl.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
It's entirely possible that we should just make that whole
if (ret == ENOENT)
go away. Yes, it's the right error code if a subdirectory is missing, and
yes, POSIX requires it, and yes, WXP is probably just a horrible piece of
sh*t, but on the other hand, I don't think git really has any serious
reason to even care.
* ml/trace:
test-lib: unset GIT_TRACE
GIT_TRACE: fix a mixed declarations and code warning
GIT_TRACE: show which built-in/external commands are executed
git-repack: avoid redirecting stderr into git-pack-objects
We are trying to catch error condition of git-rev-list and cause
the downstream pack-objects to barf, but if you run rev-list
with anything that mucks with its stderr (such as GIT_TRACE),
any stderr output would cause the pipeline to fail.
[jc: originally from Matthias Lederhofer, with a reworded error message.]
We never talk about "xyz commit", except when we end up talking
about a commit as a branch head (notably, I would say "the HEAD
commit", or possibly "the top-of-master commit", but here
$commit is a SHA1 name, not anything else).
Signed-off-by: Linus Torvalds <torvalds@osdl.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
* lt/merge-tree:
Improved three-way blob merging code
Prepare "git-merge-tree" for future work
xdiff: generate "anti-diffs" aka what is common to two files
gitweb: Make command invocations go through the git wrapper
This patch makes invocations of core git commands go through the 'git'
binary itself, which improves readability and might help system
administrators lock down their CGI environment for security.
Signed-off-by: Alp Toker <alp@atoker.com> Signed-off-by: Junio C Hamano <junkio@cox.net>
This helps users tell one 'git' bookmark apart from the other in their
browser and improves the indexing of gitweb sites in Web search engines.
The title defaults to the SERVER_NAME environment variable, often given
by the webserver.
Signed-off-by: Alp Toker <alp@atoker.com> Signed-off-by: Junio C Hamano <junkio@cox.net>
gitweb: Send XHTML as 'application/xhtml+xml' where possible
"The 'text/html' media type [RFC2854] is primarily for HTML, not for
XHTML. In general, this media type is NOT suitable for XHTML."
This patch makes gitweb use content negotiation to conservatively send
pages as Content-Type 'application/xhtml+xml' when the user agent
explicitly claims to support it.
It falls back to 'text/html' even if the user agent appears to
implicitly support 'application/xhtml+xml' due to a '*/*' glob, working
around an insidious bug in Internet Explorer where sending the correct
media type prevents the page from being displayed.
Signed-off-by: Alp Toker <alp@atoker.com> Signed-off-by: Junio C Hamano <junkio@cox.net>
This way we don't have to remember to set it for each test; and
if we forget, we won't cause interactive editors to be spawned
for non-interactive tests.
Signed-off-by: Eric Wong <normalperson@yhbt.net> Signed-off-by: Junio C Hamano <junkio@cox.net>
When git-fetch updates a reference record in the associated reflog
what type of update took place and who caused it (git-fetch or
git-pull, by invoking git-fetch).
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
Record the type of commit operation in the reflog.
If committing a merge (.git/MERGE_HEAD exists), an initial tree
(no HEAD) or using --amend to amend the prior commit then denote
the subtype of commit in the reflog. This helps to distinguish
amended or merge commits from normal commits.
In the case of --amend the prior sha1 is probably the commit which
is being thrown away in favor of the new commit. Since it is likely
that the old commit doesn't have any ref pointing to it anymore
it can be interesting to know why that the commit was replaced
and orphaned.
In the case of a merge the prior sha1 is probably the first parent
of the new merge commit. Consequently having its prior sha1 in the
reflog is slightly less interesting but its still informative to
know the commit was the result of a merge which had to be completed
by hand.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
Allow user.name and user.email to drive reflog entry.
Apparently calling setup_ident() after git_config causes the
user.name and user.email values read from the config file to be
replaced with the data obtained from the host. This means that
users who have setup their email address in user.email will instead
be writing reflog entries with their hostname.
Moving setup_ident() to before git_config in update-ref resolves
this ordering problem.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
On one of my systems, the linker is not intelligent enough to link with
pager.o (in libgit.a) when only the variable pager_in_use is needed. The
consequence is that the linker complains about an undefined variable. So,
put the variable into environment.o, where it is linked always.
Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de> Signed-off-by: Junio C Hamano <junkio@cox.net>
Avoid C99 comments, use old-style C comments instead.
This doesn't make the code uglier or harder to read, yet it makes the
code more portable. This also simplifies checking for other potential
incompatibilities. "gcc -std=c89 -pedantic" can flag many incompatible
constructs as warnings, but C99 comments will cause it to emit an error.
Signed-off-by: Pavel Roskin <proski@gnu.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
The only visible change is that git-blame doesn't understand
"--compability" anymore, but it does accept "--compatibility" instead,
which is already documented.
Signed-off-by: Pavel Roskin <proski@gnu.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
gitweb.cgi: Teach "a=blob" action to know the blob/file mime type
Now action "blob" knows the file type: if the file type is
not "text/*" then action "blob" defaults to "blob_plain",
i.e. the file is downloaded raw for the browser to interpret.
If the file type is "text/*", then "blob" defaults to the
current "cat -n"-like output, from which you can click
"plain", to get the "blob_plain" output.
Signed-off-by: Luben Tuikov <ltuikov@yahoo.com> Signed-off-by: Junio C Hamano <junkio@cox.net>
* ew/diff:
templates/hooks--update: replace diffstat calls with git diff --stat
diff: do not use configuration magic at the core-level
Update diff-options and config documentation.
diff.c: --no-color to defeat diff.color configuration.
diff.c: respect diff.renames config option
* ew/svn:
Fix some doubled word typos
Typofix in Makefile comment.
Makefile: export NO_SVN_TESTS
git-svn: migrate out of contrib (follow-up)
git-svn: migrate out of contrib
This removes the "contaminate the well even more" approach
taken in the current merge-base postprosessing code. Instead,
when there are more than one merge-base results, we compute the
merge-base between them and see if one is a fast-forward of the
other, in which case the ancestor is removed from the result.
This actually removes the objects to be pruned, unless you specify "-n"
(at which point it will just tell you which files it would prune).
This doesn't do the pack-file pruning that the shell-script used to do,
but if somebody really wants to, they could add it easily enough. I wonder
how useful it is, though, considering that "git repack -a -d" is just a
lot more efficient and generates a better end result.
Signed-off-by: Linus Torvalds <torvalds@osdl.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
Add "raw" output option to blobs in "tree" view format
Add a "raw" output option to blobs in "tree" view format, so that the
user doesn't have to click on "blob", wait for the (binary) file to be
uploaded and shown in "blob" mode, and then click on "plain" to
download the (binary) file.
This is useful when the file is clearly binary and we don't want the
browser to upload and display it in "blob" mode, but we just want to
download it. Case in point: pdf files, wlg.
Note: the "raw" format is equivalent to the blob->plain view, not
blob->head view. I.e. the view has the hash of the file as listed
by git-ls-tree, not just "HEAD".
Signed-off-by: Luben Tuikov <ltuikov@yahoo.com> Signed-off-by: Junio C Hamano <junkio@cox.net>
diff: do not use configuration magic at the core-level
The Porcelainish has become so much usable as the UI that there
is not much reason people should be using the core programs by
hand anymore. At this point we are better off making the
behaviour of the core programs predictable by keeping them
unaffected by the configuration variables. Otherwise they will
become very hard to use as reliable building blocks.
For example, "git-commit -a" internally uses git-diff-files to
figure out the set of paths that need to be updated in the
index, and we should never allow diff.renames that happens to be
in the configuration to interfere (or slow down the process).
The UI level configuration such as showing renamed diff and
coloring are still honored by the Porcelainish ("git log" family
and "git diff"), but not by the core anymore.
Even if the standard output is connected to a tty, do not
colorize the diff if we are talking to a dumb terminal when
diff.color configuration variable is set to "auto".