git-svn: don't minimize-url when doing an init that tracks multiple paths
I didn't have a chance to test the off-by-default minimize-url
stuff enough before, but it's quite broken for people passing
the --trunk/-T, --tags/-t, --branches/-b switches to "init" or
"clone" commands.
Additionally, follow-parent functionality seems broken when we're
not connected to the root of the repository.
Default behavior for "traditional" git-svn users who only track
one directory (without needing follow-parent) should be
reasonable, as those users started using things before
minimize-url functionality existed.
Behavior for users more used to the git-svnimport-like command
line will also benefit from a more-flexible command-line than
svnimport given the assumption they're working with
non-restrictive read permissions on the repository.
I hope to properly fix these bugs when I get a chance to in the
next week or so, but I would like to get this stopgap measure of
reverting to the old behavior as soon as possible.
Signed-off-by: Eric Wong <normalperson@yhbt.net> Signed-off-by: Junio C Hamano <junkio@cox.net>
git-svn: avoid crashing svnserve when creating new directories
When sorting directory names by depth (slash ("/") count) and
closing the deepest directories first (as the protocol
requires), we failed to put the root baton (with an empty string
as its key "") after top-level directories (which did not have
any slashes).
This resulted in svnserve being in a situation it couldn't
handle and caused a segmentation fault on the remote server.
This bug did not affect users of DAV and filesystem repositories.
Signed-off-by: Eric Wong <normalperson@yhbt.net> Confirmed-by: Matthieu Moy <Matthieu.Moy@imag.fr> Signed-off-by: Junio C Hamano <junkio@cox.net>
The todo list at the end of the user manual says that something must be
said about .gitignore. Also, there seems to be a lack of documentation
on how to choose between the various types of ignore files (.gitignore
vs. .git/info/exclude, etc.).
This patch adds a section on ignoring files which try to introduce how
to tell git about ignored files, and how the different strategies
complement eachother.
The syntax of exclude patterns is explained in a simplified manner, with
a reference to git-ls-files(1) which already contains a more thorough
explanation.
I don't really want to look like we're encouraging the shared repository
thing. Take down some of the argument for using purely
single-developer-owned repositories and collaborating using patches and
pulls instead.
Signed-off-by: "J. Bruce Fields" <bfields@citi.umich.edu>
The embarassing history of this tutorial is that I started it without
really understanding the index well, so I avoided mentioning it.
And we all got the idea that "index" was a word to avoid using around
newbies, but it was reluctantly mentioned that *something* had to be
said. The result is a little awkward: the discussion of the index never
actually uses that word, and isn't well-integrated into the surrounding
material.
Let's just go ahead and use the word "index" from the very start, and
try to demonstrate its use with a minimum of lecturing.
Also, remove discussion of using git-commit with explicit filenames.
We're already a bit slow here to get people to their first commit, and
I'm not convinced this is really so important.
Signed-off-by: "J. Bruce Fields" <bfields@citi.umich.edu>
Merge branch 'maint' of git://linux-nfs.org/~bfields/git into maint
* 'maint' of git://linux-nfs.org/~bfields/git:
user-manual: reorganize public git repo discussion
user-manual: listing commits reachable from some refs not others
user-manual: introduce git
user-manual: add a "counting commits" example
user-manual: move howto/using-topic-branches into manual
user-manual: move howto/make-dist.txt into user manual
Documentation: remove howto's now incorporated into manual
user-manual: move quick-start to an appendix
glossary: expand and clarify some definitions, prune cross-references
user-manual: revise birdseye-view chapter
Add a birdview-on-the-source-code section to the user manual
git-rev-list(1) talks about patterns as values for the
--grep, --committed etc. parameters, without going into detail.
This patch mentions that these patterns are actually regexps.
Signed-off-by: Petr Baudis <pasky@suse.cz> Signed-off-by: Junio C Hamano <junkio@cox.net>
Fix crlf attribute handling to match documentation
gitattributes.txt says, of the crlf attribute:
Set::
Setting the `crlf` attribute on a path is meant to mark
the path as a "text" file. 'core.autocrlf' conversion
takes place without guessing the content type by
inspection.
That is to say that the crlf attribute does not force the file to have
CRLF line endings, instead it removes the autocrlf guesswork and forces
the file to be treated as text. Then, whatever line ending is defined
by the autocrlf setting is applied.
However, that is not what convert.c was doing. The conversion to CRLF
was being skipped in crlf_to_worktree() when the following condition was
true:
action == CRLF_GUESS && auto_crlf <= 0
That is to say conversion took place when not in guess mode (crlf attribute
not specified) or core.autocrlf set to true. This was wrong. It meant
that the crlf attribute being on for a given file _forced_ CRLF
conversion, when actually it should force the file to be treated as
text, and converted accordingly. The real test should simply be
auto_crlf <= 0
That is to say, if core.autocrlf is falsei (or input), conversion from
LF to CRLF is never done. When core.autocrlf is true, conversion from
LF to CRLF is done only when in CRLF_GUESS (and the guess is "text"), or
CRLF_TEXT mode.
Similarly for crlf_to_worktree(), if core.autocrlf is false, no conversion
should _ever_ take place. In reality it was only not taking place if
core.autocrlf was false _and_ the crlf attribute was unspecified.
Signed-off-by: Andy Parkins <andyparkins@gmail.com> Signed-off-by: Junio C Hamano <junkio@cox.net>
git-archive: convert archive entries like checkouts do
As noted by Johan Herland, git-archive is a kind of checkout and needs
to apply any checkout filters that might be configured.
This patch adds the convenience function convert_sha1_file which returns
a buffer containing the object's contents, after converting, if necessary
(i.e. it's a combination of read_sha1_file and convert_to_working_tree).
Direct calls to read_sha1_file in git-archive are then replaced by calls
to convert_sha1_file.
Since convert_sha1_file expects its path argument to be NUL-terminated --
a convention it inherits from convert_to_working_tree -- the patch also
changes the path handling in archive-tar.c to always NUL-terminate the
string. It used to solely rely on the len field of struct strbuf before.
archive-zip.c already NUL-terminates the path and thus needs no such
change.
Signed-off-by: Rene Scharfe <rene.scharfe@lsrfire.ath.cx> Signed-off-by: Junio C Hamano <junkio@cox.net>
user-manual: reorganize public git repo discussion
Helping a couple people set up public repos recently, I wanted to point
them at this piece of the user manual, but found it wasn't as helpful as
it could be:
- It starts with a big explanation of why you'd want a public
repository, not necessary in their case since they already knew
why they wanted that. So, separate that out.
- It skimps on some of the git-daemon details, and puts the http
export information first. Fix that.
Also group all the public repo subsections into a single section, and do
some miscellaneous related editing.
Signed-off-by: "J. Bruce Fields" <bfields@citi.umich.edu>
user-manual: move howto/using-topic-branches into manual
Move howto/using-topic-branches into the user manual as an example for
the "sharing development" chapter. While we're at it, remove some
discussion that's covered in earlier chapters, modernize somewhat (use
separate-heads setup, remotes, replace "whatchanged" by "log", etc.),
and replace syntax we'd need to explain by syntax we've already covered
(e.g. old..new instead of new ^old).
The result may not really describe what Tony Luck does any more.... Hope
that's not annoying.
Cc: Tony Luck <tony.luck@intel.com> Signed-off-by: "J. Bruce Fields" <bfields@citi.umich.edu>
user-manual: move howto/make-dist.txt into user manual
There seems to be a perception that the howto's are bit-rotting a
little. The manual might be a more visible location for some of them,
and make-dist.txt seems like a good candidate to include as an example
in the manual.
For now, incorporate much of it verbatim. Later we may want to update
the example a bit.
Signed-off-by: "J. Bruce Fields" <bfields@citi.umich.edu>
Documentation: remove howto's now incorporated into manual
These two howto's have both been copied into the manual. I'd rather not
maintain both versions if possible, and I think the user-manual will be
more visible than the howto directory. (Though I wouldn't mind some
duplication if people really like having them here.)
Signed-off-by: "J. Bruce Fields" <bfields@citi.umich.edu>
The quick start interrupts the flow of the manual a bit. Move it to
"appendix A" but add a reference to it in the preface. Also rename the
todo chapter to "appendix B", and revise the preface a little.
Signed-off-by: "J. Bruce Fields" <bfields@citi.umich.edu>
glossary: expand and clarify some definitions, prune cross-references
Revise and expand some of the definitions in the glossary, based in part
on a recent thread started by a user looking for help with some of the
jargon. I've borrowed some of the language from Linus's email on that
thread. (I'm assuming standing permission to plagiarize Linus's
email....)
Also start making a few changes to mitigate the appearance of
"circularity" mentioned in that thread:
- feel free to use somewhat longer definitions and to explain
some things more than once instead of relying purely on
cross-references
- don't use cross-references when they're redundant: eliminate
self-references and repeated references to the same entry.
Cc: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: "J. Bruce Fields" <bfields@citi.umich.edu>
gitweb: Allow arbitrary strings to be dug with pickaxe
Currently, there are rather draconian restrictions on the strings accepted
by the pickaxe search, which degrades its usefulness for digging in code
significantly. This patch remedies mentioned limitation.
Signed-off-by: Petr Baudis <pasky@suse.cz> Signed-off-by: Junio C Hamano <junkio@cox.net>
The 'grep' type of search greps the currently selected tree for given
regexp and shows the results in a fancy table with links into blob view.
The number of shown matches is limited to 1000 and the whole feature
can be turned off (grepping linux-2.6.git already makes repo.or.cz a bit
unhappy).
This second revision makes it in documentation explicit that grep accepts
regexps, and makes grep accept extended regexps instead of basic regexps.
Signed-off-by: Petr Baudis <pasky@suse.cz> Signed-off-by: Junio C Hamano <junkio@cox.net>
Currently, searchbar font was as big as the page heading font, because
font-size was made relative - but to the parent element, which was for some
reason indeed page_header. Since that seems to be illogical to me, I just
moved the div.search outside of div.page_header. I'm no CSS/DOM expert but
no adverse effects were observed by me.
Signed-off-by: Petr Baudis <pasky@suse.cz> Signed-off-by: Junio C Hamano <junkio@cox.net>
gitweb: Fix error in git_patchset_body for deletion in merge commit
Checking if $diffinfo->{'status'} is equal 'D' is no longer the way to
check if the file was deleted in result. For merge commits
$diffinfo->{'status'} is reference to array of statuses for each
parent. Use the fact that $diffinfo->{'to_id'} is all zeros as sign
that file was deleted in result.
Signed-off-by: Jakub Narebski <jnareb@gmail.com> Signed-off-by: Junio C Hamano <junkio@cox.net>
git-gui: Gracefully handle bad TCL_PATH at compile time
Petr Baudis pointed out the main git.git repository's Makefile dies
now if git-gui 0.7.0-rc1 or later is being used and TCL_PATH was not
set to a working tclsh program path. This breaks people who may have
a working build configuration today and suddenly upgrade to the latest
git release.
The tclIndex is required for git-gui to load its associated lib files,
but using the Tcl auto_load procedure to source only the files we need
is a performance optimization. We can emulate the auto_load by just
source'ing every file in that directory, assuming we source class.tcl
first to initialize our crude class system.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
gitweb: Empty patch for merge means trivial merge, not no differences
Earlier commit 4280cde95fa4e3fb012eb6d0c239a7777baaf60c made gitweb
show "No differences found" message for empty diff, for the HTML
output. But for merge commits, either -c format we use or --cc format,
empty diff doesn't mean no differences, but trivial merge.
Show "Trivial merge" message instead of "No differences found" for
merges.
While at it reword conditional in the code for easier reading.
Signed-off-by: Jakub Narebski <jnareb@gmail.com> Signed-off-by: Junio C Hamano <junkio@cox.net>
* maint:
format-patch: add MIME-Version header when we add content-type.
Fixed link in user-manual
import-tars: Use the "Link indicator" to identify directories
git name-rev writes beyond the end of malloc() with large generations
Documentation/branch: fix small typo in -D example
Separate search text, which is saved in $searchtext global variable,
and is used in links, as default value for the textfield in search
form, and for pickaxe search, from search regexp, which is saved in
$search_regexp global variable, and is used as parameter to --grep,
--committer or --author options to git-rev-list, and for searching
commit body in gitweb. For now $search_regexp is unconditionallt
equal to quotemeta($searchtext), meaning that we always search for
fixed string.
This fixes bug where 'next page' links for 'search' view didn't work
for searchtext containing quotable characters, like `@'.
Signed-off-by: Jakub Narebski <jnareb@gmail.com> Signed-off-by: Junio C Hamano <junkio@cox.net>
David Kågedal proposed that gitweb should explicitely request
being somewhat smaller than normal, because it has good use for
long lines. However gitweb presents a table with several
columns, so having wider line is OK for it. Therefore explicit
'font-size: small' would make sense. Apparently many people on
the list seem to agree with him.
import-tars: Use the "Link indicator" to identify directories
Earlier, we used the mode to determine if a name was associated with
a directory. This fails, since some tar programs do not set the mode
correctly. However, the link indicator _has_ to be set correctly.
Noticed by Chris Riddoch.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Acked-by: Junio C Hamano <junkio@cox.net> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Ensure return value from xread() is always stored into an ssize_t
This patch fixes all calls to xread() where the return value is not
stored into an ssize_t. The patch should not have any effect whatsoever,
other than putting better/more appropriate type names on variables.
Signed-off-by: Johan Herland <johan@herland.net> Signed-off-by: Junio C Hamano <junkio@cox.net>
The return value from xread() is ssize_t.
Paolo Teti <paolo.teti@gmail.com> pointed out that in this case, the
signed return value was assigned to an unsigned type (size_t). This patch
fixes that.
Signed-off-by: Johan Herland <johan@herland.net> Signed-off-by: Junio C Hamano <junkio@cox.net>
This allocation is only sufficient if the generation number is
less than 5 digits, in my case generation was 13432. In reality
parent_number can be up to 16 so that also can require two digits,
reducing us to 3 digits before we are at risk of blowing this
allocation.
This patch introduces a decimal_length() which approximates the
number of digits a type may hold, it produces the following:
Type Longest Value Len Est
---- ------------- --- ---
unsigned char 256 3 4
unsigned short 65536 5 6
unsigned long 4294967296 10 11
unsigned long long 18446744073709551616 20 21
char -128 4 4
short -32768 6 6
long -2147483648 11 11
long long -9223372036854775808 20 21
This is then used to size the new_name.
Signed-off-by: Andy Whitcroft <apw@shadowen.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
Optimized cvsexportcommit: calling 'cvs status' once instead of once per touched file.
Runtime is now independent of the number of modified files.
The old implementation executed 'cvs status' for each file touched by the patch
to be applied. The new code calls 'cvs status' only once with all touched files
and parses cvs's output to collect all available status information.
Signed-off-by: Steffen Prohaska <prohaska@zib.de> Signed-off-by: Junio C Hamano <junkio@cox.net>
Use $Id$ as the ident attribute keyword rather than $ident$ to be consistent with other VCSs
$Id$ is present already in SVN and CVS; it would mean that people
converting their existing repositories won't have to make any changes to
the source files should they want to make use of the ident attribute.
Given that it's a feature that's meant to calm those very people, it
seems obtuse to make them edit every file just to make use of it.
I think that bzr uses $Id$; Mercurial has examples hooks for $Id$;
monotone has $Id$ on its wishlist. I can't think of a good reason not
to stick with the de-facto standard and call ours $Id$ instead of
$ident$.
Signed-off-by: Andy Parkins <andyparkins@gmail.com> Signed-off-by: Junio C Hamano <junkio@cox.net>
* maint:
Prepare for 1.5.1.5 Release Notes
gitweb: Add a few comments about %feature hash
git-am: Clean up the asciidoc documentation
Documentation: format-patch has no --mbox option
builtin-log.c: Fix typo in comment
Fix git-clone buglet for remote case.
git-applymbox and git-mailinfo refer to a --mbox option of
git-format-patch when talking about their -k options. But there
is no such option. What -k does to the former two commands is
to keep the Subject: lines unmunged, meant to be used on output
generated with format-patch -k.
Signed-off-by: Frank Lichtenheld <frank@lichtenheld.de> Signed-off-by: Junio C Hamano <junkio@cox.net>
c2f599e09fd0496413d1744b5b89b9b5c223555d introduced a buglet while
cloning from a remote URL; we forgot to squelch the unnecessary
error message when we try to cd to the given "remote" name,
in order to see if it is a local directory.
After we send I HATE YOU we should probably exit and not happily
continue with I LOVE YOU and further communication.
Most clients will probably just exit and ignore everything we
send after the I HATE YOU and it is not a security problem
either because we don't really care about the user name anyway.
But it is still the right thing to do.
[jc: with a minor fixup to its exit code...]
Signed-off-by: Frank Lichtenheld <frank@lichtenheld.de> Acked-by: "Martin Langhoff" <martin.langhoff@gmail.com> Signed-off-by: Junio C Hamano <junkio@cox.net>
Documentation: Split description of pretty formats of commit log
Split description of pretty formats into list of pretty options
(--pretty and --encoding) in new file Documentation/pretty-options.txt
and description of formats itself as a separate "PRETTY FORMATS"
section in pretty-formats.txt
While at it correct formatting a bit, to be better laid out in the
resulting manpages: git-rev-list(1), git-show(1), git-log(1) and
git-diff-tree(1). Those manpages now include pretty options in the
same place as it was before, and description of formats just after
all options.
Inspired by the split into two filesdocumentation for merge strategies:
Documentation/merge-options.txt and Documentation/merge-strategies.txt
Signed-off-by: Jakub Narebski <jnareb@gmail.com> Signed-off-by: Junio C Hamano <junkio@cox.net>
* maint:
git-svn: don't attempt to minimize URLs by default
git-svn: fix segfaults due to initial SVN pool being cleared
git-svn: clean up caching of SVN::Ra functions
git-svn: don't drop the username from URLs when dcommit is run
RPM spec: include files in technical/ to package.
Remove stale non-static-inline prototype for tree_entry_extract()
git-config: test for 'do not forget "a.b.var" ends "a.var" section'.
git-config: do not forget seeing "a.b.var" means we are out of "a.var" section.
Change the configuration parser so that it ignores
everything except for ^gitcvs.((ext|pserver).)?
This greatly reduces the risk of failing while
parsing some unknown and irrelevant config option.
The bug that triggered this change was that the
parsing doesn't handle sections that have a
subsection and a variable with the same name.
While this bug still remains, all remaining
causes can be attributed to user error, since
there are no defined variables gitcvs.ext and
gitcvs.pserver.
Signed-off-by: Frank Lichtenheld <frank@lichtenheld.de> Signed-off-by: Junio C Hamano <junkio@cox.net>
Try to avoid "Use of uninitialized value ..." errors caused by bad
revision, incorrect filename, wrong object id, bad file etc. (wrong
value of 'h', 'hb', 'f', etc. parameters). This avoids polluting web
server errors log.
Correct git_get_hash_by_path and parse_commit_text (and, in turn,
parse_commit) to return undef if object does not exist. Check in
git_tag if requested tag exists.
Signed-off-by: Jakub Narebski <jnareb@gmail.com> Signed-off-by: Junio C Hamano <junkio@cox.net>
gives four hits that refer to .txt version of the documentation
set, but at least "hooks" and "cvs-migration" have HTML variants
installed, so refer to them instead.
git-svn: don't attempt to minimize URLs by default
For tracking branches and tags, git-svn prefers to connect
to the root of the repository or at least the level that
houses branches and tags as well as trunk. However, users
that are accustomed to tracking a single directory have
no use for this feature.
As pointed out by Junio, users may not have permissions to
connect to connect to a higher-level path in the repository.
While the current minimize_url() function detects lack of
permissions to certain paths _after_ successful logins, it
cannot effectively determine if it is trying to access a
login-only portion of a repo when the user expects to
connect to a part where anonymous access is allowed.
For people used to the git-svnimport switches of
--trunk, --tags, --branches, they'll already pass the
repository root (or root+subdirectory), so minimize URL
isn't of too much use to them, either.
For people *not* used to git-svnimport, git-svn also
supports:
git-svn: fix segfaults due to initial SVN pool being cleared
Some parts of SVN always seem to use it, even if the SVN::Ra
object we're using is no longer used and we've created a new one
in its place. It's also true that only one SVN::Ra connection
can exist at once... Using SVN::Pool->new_default when the
SVN::Ra object is created doesn't seem to help very much,
either...
Hopefully this fixes all segfault problems users have been
experiencing over the past few months.
This patch was originally intended to make the Perl GC more
sensitive to the SVN::Pool objects and not accidentally clean
them up when they shouldn't be (causing segfaults). That didn't
work, but this patch makes the code a bit cleaner regardless
Put our caches for get_dir and check_path calls directly into
the SVN::Ra object so they auto-expire when it is destroyed.
dirents returned by get_dir() no longer needs the pool object
stored persistently along with the cache data, as they'll be
converted to native Perl hash references.
Since calling rev_proplist repeatedly per-revision is no longer
needed in git-svn, we do not cache calls to it.
git-svn: don't drop the username from URLs when dcommit is run
We no longer store usernames in URLs stored in git-svn-id lines
for dcommit, so we shouldn't rely on those URLs when connecting
to the remote repository to commit.
git-config: do not forget seeing "a.b.var" means we are out of "a.var" section.
Earlier code tried to be half-careful and knew the logic that
seeing "a.var" after seeing "a.b.var" is a sign of the previous
"a.b." section has ended, but forgot it has to handle the other
way. Seeing "a.b.var" after seeing "a.var" is a sign that "a."
section has ended, so a new "a.var2" variable should be added
before the location "a.b.var" appears.
Minor fixup to documentation of hooks in git-receive-pack.
Small additional changes to the cbb84e5d174cf33fd4dcf3136de50a886ff9a2e2
commit, which introduced documentation to pre-receive and post-receive:
- Mention that stdout and stderr are equivalent.
- Add one cross-section link and fix one other.
- Fix information on advantages of post-receive over post-update.
Signed-off-by: Jan Hudec <bulb@ucw.cz> Signed-off-by: Junio C Hamano <junkio@cox.net>
* maint:
checkout: allow detaching to HEAD even when switching to the tip of a branch
Updated documentation of hooks in git-receive-pack.
Allow fetching references from any namespace
tiny fix in documentation of git-clone
Fix an unmatched comment end in arm/sha1_arm.S
t9400: Use the repository config and nothing else.
git-cvsserver has a bug in its configuration file output parser
that makes it choke if the configuration has these:
[diff]
color = auto
[diff.color]
whitespace = blue reverse
This needs to be fixed, but thanks to that bug, a separate bug
in t9400 test script was discovered. The test discarded
GIT_CONFIG instead of pointing at the proper one to be used in
the exoprted repository. This allowed user's .gitconfig and (if
exists) systemwide /etc/gitconfig to affect the outcome of the
test, which is a big no-no.
The patch fixes the problem in the test. Fixing the
git-cvsserver's configuration parser is left as an exercise to
motivated volunteers ;-)
not only from the three defined: heads, tags and remotes.
Noticed when I tried to fetch the references created by git-p4-import.bat:
they are placed into separate namespace (refs/p4import/, to avoid showing
them in git-branch output). As canon_refs_list_for_fetch always prepended
refs/heads/ it was impossible, and annoying: it worked before. Normally,
the p4import references are useless anywhere but in the directory managed
by perforce, but in this special case the cloned directory was supposed
to be a backup, including the p4import branch: it keeps information about
where the imported perforce state came from.
Signed-off-by: Alex Riesen <raa.lkml@gmail.com> Signed-off-by: Junio C Hamano <junkio@cox.net>
git-archive: don't die when repository uses subprojects
Both archive-tar and archive-zip needed to be taught about subprojects.
The tar function died when trying to read the subproject commit object,
while the zip function reported "unsupported file mode".
This fixes both by representing the subproject as an empty directory.
Signed-off-by: Lars Hjemli <hjemli@gmail.com> Signed-off-by: Junio C Hamano <junkio@cox.net>
gitweb: Test if $from_id and $to_id are defined before comparison
Get rid of "Use of uninitialized value in string eq at
gitweb/gitweb.perl line 2320" warning caused by the fact that "empty"
patches, consisting only of extended git diff header and with patch
body empty, such as patch for pure rename, does not have "index" line
in extended diff header. For such patches $from_id and $to_id, filled
from parsing extended diff header, are undefined. But such patches
cannot be continuation patches.
Test if $from_id and $to_id are defined before comparing them with
$diffinfo.
Signed-off-by: Jakub Narebski <jnareb@gmail.com> Signed-off-by: Junio C Hamano <junkio@cox.net>
read-tree -m -u: avoid getting confused by intermediate symlinks.
When switching from a branch with both x86_64/boot/Makefile and
i386/boot/Makefile to another branch that has x86_64/boot as a
symlink pointing at ../i386/boot, the code incorrectly removed
i386/boot/Makefile.
This was because we first removed everything under x86_64/boot
to make room to create a symbolic link x86_64/boot, then removed
x86_64/boot/Makefile which no longer exists but now is pointing
at i386/boot/Makefile, thanks to the symlink we just created.
This fixes it by using the has_symlink_leading_path() function
introduced previously for git-apply in the checkout codepath.
Earlier, "git checkout" was broken in t4122 test due to this
bug, and the test had an extra "git reset --hard" as a
workaround, which is removed because it is not needed anymore.
apply: do not get confused by symlinks in the middle
HPA noticed that git-rebase fails when changes involve symlinks
in the middle of the hierarchy. Consider:
* The tree state before the patch is applied has arch/x86_64/boot
as a symlink pointing at ../i386/boot/
* The patch tries to remove arch/x86_64/boot symlink, and
create bunch of files there: .gitignore, Makefile, etc.
git-apply tries to be careful while applying patches; it never
touches the working tree until it is convinced that the patch
would apply cleanly. One of the check it does is that when it
knows a path is going to be created by the patch, it runs
lstat() on the path to make sure it does not exist.
This leads to a false alarm. Because we do not touch the
working tree before all the check passes, when we try to make
sure that arch/x86_64/boot/.gitignore does not exist yet, we
haven't removed the arch/x86_64/boot symlink. The lstat() check
ends up seeing arch/i386/boot/.gitignore through the
yet-to-be-removed symlink, and says "Hey, you already have a
file there, but what you fed me is a patch to create a new
file. I am not going to clobber what you have in the working
tree."
We have similar checks to see a file we are going to modify does
exist and match the preimage of the diff, which is done by
directly opening and reading the file.
For a file we are going to delete, we make sure that it does
exist and matches what is going to be removed (a removal patch
records the full preimage, so we check what you have in your
working tree matches it in full -- otherwise we would risk
losing your local changes), which again is done by directly
opening and reading the file.
These checks need to be adjusted so that they are not fooled by
symlinks in the middle.
- To make sure something does not exist, first lstat(). If it
does not exist, it does not, so be happy. If it _does_, we
might be getting fooled by a symlink in the middle, so break
leading paths and see if there are symlinks involved. When
we are checking for a path a/b/c/d, if any of a, a/b, a/b/c
is a symlink, then a/b/c/d does _NOT_ exist, for the purpose
of our test.
This would fix this particular case you saw, and would not
add extra overhead in the usual case.
- To make sure something already exists, first lstat(). If it
does not exist, barf (up to this, we already do). Even if it
does seem to exist, we might be getting fooled by a symlink
in the middle, so make sure leading paths are not symlinks.
This would make the normal codepath much more expensive for
deep trees, which is a bit worrisome.
This patch implements the first side of the check "making sure
it does not exist". The latter "making sure it exists" check is
not done yet, so applying the patch in reverse would still
fail, but we have to start from somewhere.
When we are applying a patch that creates a blob at a path, or
when we are switching from a branch that does not have a blob at
the path to another branch that has one, we need to make sure
that there is nothing at the path in the working tree, as such a
file is a local modification made by the user that would be lost
by the operation.
Normally, lstat() on the path and making sure ENOENT is returned
is good enough for that purpose. However there is a twist. We
may be creating a regular file arch/x86_64/boot/Makefile, while
removing an existing symbolic link at arch/x86_64/boot that
points at existing ../i386/boot directory that has Makefile in
it. We always first check without touching filesystem and then
perform the actual operation, so when we verify the new file,
arch/x86_64/boot/Makefile, does not exist, we haven't removed
the symbolic link arc/x86_64/boot symbolic link yet. lstat() on
the file sees through the symbolic link and reports the file is
there, which is not what we want.
The function has_symlink_leading_path() function takes a path,
and sees if any of the leading directory component is a symbolic
link.
When files in a new directory are created, we tend to process
them together because both index and tree are sorted. The
function takes advantage of this and allows the caller to cache
and reuse which symbolic link on the filesystem caused the
function to return true.
The calling sequence would be:
char last_symlink[PATH_MAX];
*last_symlink = '\0';
for each index entry {
if (!lose)
continue;
if (lstat(it))
if (errno == ENOENT)
; /* happy */
else
error;
else if (has_symlink_leading_path(it, last_symlink))
; /* happy */
else
error; /* would lose local changes */
unlink_entry(it, last_symlink);
}