builtin-reflog.c: don't install new reflog on write failure
When expiring reflog entries, a new temporary log is written which contains
only the entries to retain. After it is written, it is renamed to replace
the existing reflog. Currently, we check that writing of the new log is
successful and print a message on failure, but the original reflog is still
replaced with the new reflog even on failure. This patch causes the
original reflog to be retained if we fail when writing the new reflog.
Signed-off-by: Brandon Casey <casey@nrlssc.navy.mil> Signed-off-by: Junio C Hamano <gitster@pobox.com>
1ca3d6e (send-email: squelch warning due to comparing undefined $_ to "")
where if the user was prompted for an initial In-Reply-To and didn't
provide one, messages would be sent out with an invalid In-Reply-To of
"<>"
Also add test cases for the regression and the fix. A small modification
was needed to allow send-email to take its replies from stdin if the
environment variable GIT_SEND_EMAIL_NOTTY is set.
Signed-off-by: Jay Soffian <jaysoffian@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
gitweb: Clearly distinguish regexp / exact match searches
This patch does a couple of things:
* Makes commit/author/committer search case insensitive
To be consistent with the grep search; I see no convincing
reason for the search to be case sensitive, and you might
get in trouble especially with contributors e.g. from Japan
or France where they sometimes like to uppercase their last
name.
* Makes grep search by default search for fixed strings.
* Introduces 're' checkbox that enables POSIX extended regexp searches
This works for all the search types. The idea comes from Jakub.
It does not make much sense (and is not easy at all) to untangle most
of these changes from each other, thus they all go in a single patch.
[jn: Cherry-picked from Pasky's http://repo.or.cz/git/gitweb.git]
Signed-off-by: Petr Baudis <pasky@suse.cz> Signed-off-by: Jakub Narebski <jnareb@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Use '--fixed-strings' option to git-rev-list to simplify and improve
searching commit messages (commit search). It allows to search for
example for "don't" successfully from gitweb.
Signed-off-by: Jakub Narebski <jnareb@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
gitweb: Change parse_commits signature to allow for multiple options
Change order of parameters in parse_commits() to have $filename
before @args (extra options), to allow for multiple extra options,
for example both '--grep=<pattern>' and '--fixed-strings'.
Change all callers to follow new calling convention.
Originally by Petr Baudis, in http://repo.or.cz/git/gitweb.git:
b98f0a7c gitweb: Clearly distinguish regexp / exact match searches
Signed-off-by: Petr Baudis <pasky@suse.cz> Signed-off-by: Jakub Narebski <jnareb@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Signed-off-by: Alexandre Julliard <julliard@winehq.org> Tested-by: Karl Hasselström <kha@treskal.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
* db/checkout: (21 commits)
checkout: error out when index is unmerged even with -m
checkout: show progress when checkout takes long time while switching branches
Add merge-subtree back
checkout: updates to tracking report
builtin-checkout.c: Remove unused prefix arguments in switch_branches path
checkout: work from a subdirectory
checkout: tone down the "forked status" diagnostic messages
Clean up reporting differences on branch switch
builtin-checkout.c: fix possible usage segfault
checkout: notice when the switched branch is behind or forked
Build in checkout
Move code to clean up after a branch change to branch.c
Library function to check for unmerged index entries
Use diff -u instead of diff in t7201
Move create_branch into a library file
Build-in merge-recursive
Add "skip_unmerged" option to unpack_trees.
Discard "deleted" cache entries after using them to update the working tree
Send unpack-trees debugging output to stderr
Add flag to make unpack_trees() not print errors.
...
* db/cover-letter:
Improve collection of information for format-patch --cover-letter
Add API access to shortlog
t4014: Replace sed's non-standard 'Q' by standard 'q'
Support a --cc=<email> option in format-patch
Combine To: and Cc: headers
Fix format.headers not ending with a newline
Add tests for extra headers in format-patch
Add a --cover-letter option to format-patch
Export some email and pretty-printing functions
Improve message-id generation flow control for format-patch
Add more tests for format-patch
* js/run-command:
start_command(), if .in/.out > 0, closes file descriptors, not the callers
start_command(), .in/.out/.err = -1: Callers must close the file descriptor
* ae/pack-autothread:
Revert "pack-objects: Print a message describing the number of threads for packing"
pack-objects: Print a message describing the number of threads for packing
pack-objects: Add runtime detection of online CPU's
* sp/describe:
Use git-describe --exact-match in bash prompt on detached HEAD
Teach git-describe --exact-match to avoid expensive tag searches
Avoid accessing non-tag refs in git-describe unless --all is requested
Teach git-describe to use peeled ref information when scanning tags
Optimize peel_ref for the current ref of a for_each_ref callback
Fix 'git cvsexportcommit -w $cvsdir ...' when used with relative $GIT_DIR
When using the '-w $cvsdir' option to cvsexportcommit, it will chdir into
$cvsdir before executing several other git commands. If $GIT_DIR is set to
a relative path (e.g. '.'), the git commands executed by cvsexportcommit
will naturally fail.
Therefore, ensure that $GIT_DIR is absolute before the chdir to $cvsdir.
Signed-off-by: Johan Herland <johan@herland.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
When syncing, git-p4 will only download files that are included in the active
perforce client spec. This does not change the default behaviour - it requires
that the user either supplies the command line argument --use-client-spec, or
sets the git config option p4.useclientspec to "true".
Signed-off-by: Tor Arvid Lund <torarvid@gmail.com> Signed-off-by: Simon Hausmann <simon@lst.de>
This feature was originally meant to allow for quicker direct submits into perforce, but
it turns out that it is not actually quicker than doing a git commit and then running
git-p4 submit.
git-p4: Clean up git-p4 submit's log message handling.
Instead of trying to substitute fields in the p4 submit template we now simply
replace the description of the submit with the log message of the git commit.
Teach git-p4 about the -/ option which adds depot paths to the exclude
list, used when cloning. The option is chosen such that the natural
Perforce syntax works, eg:
For some reason, t5505-remote was setting GIT_CONFIG to .git/config
and exporting it. This should have been no-op, as test framework did
the same for a long time anyway.
Signed-off-by: Daniel Barkalow <barkalow@iabervon.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add '--fixed-strings' option to "git log --grep" and friends
Add support for -F | --fixed-strings option to "git log --grep"
and friends: "git log --author", "git log --committer=<pattern>".
Code is based on implementation of this option in "git grep".
Signed-off-by: Jakub Narebski <jnareb@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Revert "pack-objects: Print a message describing the number of threads for packing"
This reverts commit 6c723f5e6bc579e06a904874f1ceeb8ff2b5a17c.
The additional message may be interesting for git developers,
but not useful for the end users, and clutters the output.
When a patch adds a whitespace followed by end-of-line, the
trailing whitespace error was detected correctly but was not
fixed, due to misconversion in 42ab241 (builtin-apply.c: do not
feed copy_wsfix() leading '+').
* maint:
Documentation/git-am.txt: Pass -r in the example invocation of rm -f .dotest
timezone_names[]: fixed the tz offset for New Zealand.
filter-branch documentation: non-zero exit status in command abort the filter
rev-parse: fix potential bus error with --parseopt option spec handling
Use a single implementation and API for copy_file()
Documentation/git-filter-branch: add a new msg-filter example
Correct fast-export file mode strings to match fast-import standard
parse_commit ignores parent commits with certain errors
(eg. a non commit object is already loaded under the sha1 of
the parent). To make fsck reports such errors, it has to compare
the nummer of parent commits returned by parse commit with the
number of parent commits in the object or in the graft/shallow file.
Signed-off-by: Martin Koegler <mkoegler@auto.tuwien.ac.at> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The requirements are:
* it may not crash on NULL pointers
* a callback function is needed, as index-pack/unpack-objects
need to do different things
* the type information is needed to check the expected <-> real type
and print better error messages
Signed-off-by: Martin Koegler <mkoegler@auto.tuwien.ac.at> Signed-off-by: Junio C Hamano <gitster@pobox.com>
gitweb: Better cutting matched string and its context
Improve look of commit search output ('search' view) by better cutting
of matched string and its context in match info, as suggested by Junio.
For example, if you are looking for "very long search string" in the
following line:
Could somebody test this with very long search string, and see how
you would now see:
...this with <<very long ... string>>, and see...
instead of:
Could som... <<very long search...>>, and see...
(where <<something>> denotes emphasized / colored fragment; matched
fragment to be more exact).
For this feature, support for fourth [optional] parameter to chop_str
subroutine was added. This fourth parameter is used to denote where
to cut string to make it shorter. chop_str can now cut at the
beginning (from the _left_ side of the string), in the middle
(_center_ of the string), or at the end (from the _right_ side of
the string); cutting from right is the default:
If you want to use default slop (default additional length), use undef
as value for third parameter to chop_str.
While at it, return from chop_str early if given string is so short
that chop_str couldn't shorten it. Simplify also regexp used by
chop_str. Make ellipsis (dots) stick to shortened fragment for
cutting at ends, to better see which part got shortened.
Simplify passing all arguments to chop_str in chop_and_escape_str
subroutine. This was needed to pass additional options to chop_str.
Signed-off-by: Jakub Narebski <jnareb@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
filter-branch documentation: non-zero exit status in command abort the filter
Since commit 8c1ce0f46b85d40f215084eed7313896300082df filter-branch fails
when a <command> has a non-zero exit status. This commit makes it clear
in the documentation and also fixes the parent-filter example, that was
incorrectly returning non-zero when the commit being tested wasn't the
one to be rewritten.
Signed-off-by: Caio Marcelo de Oliveira Filho <cmarcelo@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
rev-parse: fix potential bus error with --parseopt option spec handling
A non-empty line containing no spaces should be treated by --parseopt as
an option group header, but was causing a bus error. Also added a test
script for rev-parse --parseopt.
Signed-off-by: Jay Soffian <jaysoffian@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Improve collection of information for format-patch --cover-letter
Use the "boundary" feature to find the origin (or find that there are
multiple origins), and use the actual list of commits to pass to
shortlog.
This makes all cover letter include shortlogs, and all cover letters
for series with a single boundary commit include diffstats (if there
are multiple boundary commits it's unclear what would be meaningful as
a diffstat). Note that the single boundary test is empirical, not
theoretical; even a -2 limiting condition will give a diffstat if there's
only one boundary commit in this particular case.
Signed-off-by: Daniel Barkalow <barkalow@iabervon.org>
Shortlog is gives a pretty simple API for cases where you're already
identifying all of the individual commits. Make this available to
other code instead of requiring them to use the revision API and
command line.
Signed-off-by: Daniel Barkalow <barkalow@iabervon.org>
git-describe: --long shows the object name even for a tagged commit
This is useful when you want to see parts of the commit object name
in "describe" output, even when the commit in question happens to be
a tagged version. Instead of just emitting the tag name, it will
describe such a commit as v1.2-0-deadbeef (0th commit since tag v1.2
that points at object deadbeef....).
Signed-off-by: Santi Béjar <sbejar@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add tests for filesystem challenges (case and unicode normalization)
Git has difficulties on file systems that do not properly
distinguish case or modify filenames in unexpected ways. The two
major examples are Windows and Mac OS X. Both systems preserve
case of file names but do not distinguish between filenames that
differ only by case. Simple operations such as "git mv" or
"git merge" can fail unexpectedly. In addition, Mac OS X normalizes
unicode, which make git's life even harder.
This commit adds tests that currently fail but should pass if
file system as decribed above are fully supported. The test need
to be run on Windows and Mac X as they already pass on Linux.
Mitch Tishmack is the original author of the tests for unicode
normalization.
[jc: fixed-up so that it will use test_expect_success to test
on sanely behaving filesystems.]
Signed-off-by: Steffen Prohaska <prohaska@zib.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Earlier we had a cop-out in the documentation to make the
behaviour "undefined" if configuration had more than one
insteadOf that would match the target URL, like this:
This allows users with different preferences for access methods to the
same remote repositories to rewrite each other's URLs by pattern
matching across a large set of similiarly set up repositories to each
get the desired access.
For example, if you don't have a kernel.org account, you might want
settings like:
If we have an alias "foo" defined, then the help text for
"foo" (via "git help foo" or "git foo --help") now shows the
definition of the alias.
Before showing an alias definition, we make sure that there
is no git command which would override the alias (so that
even though you may have a "log" alias, even though it will
not work, we don't want to it supersede "git help log").
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
This converts git_config_alias to the public alias_lookup
function. Because of the nature of our config parser, we
still have to rely on setting static data. However, that
interface is wrapped so that you can just say
value = alias_lookup(key);
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Previously, the fake.sendmail test harness would write its
output to a hardcoded file, allowing only a single message
to be tested. Instead, let's have it save the messages for
all of its invocations so that we can see which messages
were sent, and in which order.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Eliminate confusing "won't bisect on seeked tree" failure
This error message is very confusing---it doesn't tell the user
anything about how to fix the situation. And the actual fix
for the situation ("git bisect reset") does a checkout of a
potentially random branch, (compared to what the user wants to
be on for the bisect she is starting).
The simplest way to eliminate the confusion is to just make
"git bisect start" do the cleanup itself. There's no significant
loss of safety here since we already have a general safety in
the form of the reflog.
Note: We preserve the warning for any cogito users. We do this
by switching from .git/head-name to .git/BISECT_START for the
extra state, (which is a more descriptive name anyway).
Signed-off-by: Carl Worth <cworth@cworth.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
* jc/apply-whitespace:
ws_fix_copy(): move the whitespace fixing function to ws.c
apply: do not barf on patch with too large an offset
core.whitespace: cr-at-eol
git-apply --whitespace=fix: fix whitespace fuzz introduced by previous run
builtin-apply.c: pass ws_rule down to match_fragment()
builtin-apply.c: move copy_wsfix() function a bit higher.
builtin-apply.c: do not feed copy_wsfix() leading '+'
builtin-apply.c: simplify calling site to apply_line()
builtin-apply.c: clean-up apply_one_fragment()
builtin-apply.c: mark common context lines in lineinfo structure.
builtin-apply.c: optimize match_beginning/end processing a bit.
builtin-apply.c: make it more line oriented
builtin-apply.c: push match-beginning/end logic down
builtin-apply.c: restructure "offset" matching
builtin-apply.c: refactor small part that matches context
Use git-describe --exact-match in bash prompt on detached HEAD
Most of the time when I am on a detached HEAD and I am not doing
a rebase or bisect operation the working directory is sitting on a
tagged release of the repository. Showing the tag name instead of
the commit SHA-1 is much more descriptive and a much better reminder
of the state of this working directory.
Now that git-describe --exact-match is available as a cheap means
of obtaining the exact annotated tag or nothing at all, we can
favor the annotated tag name over the abbreviated commit SHA-1.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Teach git-describe --exact-match to avoid expensive tag searches
Sometimes scripts want (or need) the annotated tag name that exactly
matches a specific commit, or no tag at all. In such cases it can be
difficult to determine if the output of `git describe $commit` is a
real tag name or a tag+abbreviated commit. A common idiom is to run
git-describe twice:
if test $(git describe $commit) = $(git describe --abbrev=0 $commit)
...
but this is a huge waste of time if the caller is just going to pick a
different method to describe $commit or abort because it is not exactly
an annotated tag.
Setting the maximum number of candidates to 0 allows the caller to ask
for only a tag that directly points at the supplied commit, or to have
git-describe abort if no such item exists.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Avoid accessing non-tag refs in git-describe unless --all is requested
If we aren't going to use a ref there is no reason for us to open
its object from the object database. This avoids opening any of
the head commits reachable from refs/heads/ unless they are also
reachable through the commit we have been asked to describe and
we need to walk through it to find a tag.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Teach git-describe to use peeled ref information when scanning tags
By using the peeled ref information inside of the packed-refs file we
can avoid opening tag objects to obtain the commits they reference.
This speeds up git-describe when there are a large number of tags
in the repository as we have less objects to parse before we can
start commit matching.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Optimize peel_ref for the current ref of a for_each_ref callback
Currently the only caller of peel_ref is show-ref, which is using
this function to show the peeled tag information if it is available
from an existing packed-refs file. The call happens during the
for_each_ref callback function, so we have the proper struct ref_list
already on the call stack but it is not easily available to return
the peeled information to the caller.
We now save the current struct ref_list item before calling back
into the callback function so that future calls to peel_ref from
within the callback function can quickly access the current ref.
Doing so will save us an lstat() per ref processed as we no longer
have to check the filesystem to see if the ref exists as a loose
file or is packed. This current ref caching also saves a linear
scan of the cached packed refs list.
As a micro-optimization we test the address of the passed ref name
against the current_ref->name before we go into the much more costly
strcmp(). Nearly any caller of peel_ref will be passing us the same
string do_for_each_ref passed them, which is current_ref->name.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
So I find it irritating when git thinks for a long time without telling me
what's taking so long. And by "long time" I definitely mean less than two
seconds, which is already way too long for me.
This hits me when doing a large pull and the checkout takes a long time,
or when just switching to another branch that is old and again checkout
takes a while.
Now, git read-tree already had support for the "-v" flag that does nice
updates about what's going on, but it was delayed by two seconds, and if
the thing had already done more than half by then it would be quiet even
after that, so in practice it meant that we migth be quiet for up to four
seconds. Much too long.
So this patch changes the timeout to just one second, which makes it much
more palatable to me.
The other thing this patch does is that "git checkout" now doesn't disable
the "-v" flag when doing its thing, and only disables the output when
given the -q flag. When allowing "checkout -m" to fall back to a 3-way
merge, the users will see the error message from straight "checkout",
so we will tell them that we do fall back to make them look less scary.
Protect peel_ref fallback case from NULL parse_object result
If the SHA-1 we are requesting the object for does not exist in
the object database we get a NULL back. Accessing the type from
that is not likely to succeed on any system.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Ensure 'make dist' compiles git-archive.exe on Cygwin
On Cygwin we have to use git-archive.exe as the target, otherwise
running 'make dist' does not compile git-archive in the current
directory. That may cause 'make dist' to fail on a clean source
tree that has never been built before.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
t4014: Replace sed's non-standard 'Q' by standard 'q'
t4014 test used GNU extension 'Q' in its sed scripts, but the
uses can safely be replaced with 'q'. Among other platforms,
sed on Mac OS X 10.4 does not accept the former.
Signed-off-by: Steffen Prohaska <prohaska@zib.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
pack-objects: Add runtime detection of online CPU's
Packing objects can be done in parallell nowadays, but it's
only done if the config option pack.threads is set to a value
above 1. Because of that, the code-path used is often not the
most optimal one.
This patch adds a routine to detect the number of online CPU's
at runtime (online_cpus()). When pack.threads (or --threads=) is
given a value of 0, the number of threads is set to the number of
online CPU's. This feature is also documented.
As per Nicolas Pitre's recommendations, the default is still to
run pack-objects single-threaded unless explicitly activated,
either by configuration or by command line parameter.
The routine online_cpus() is a rework of "numcpus.c", written by
one Philip Willoughby <pgw99@doc.ic.ac.uk>. numcpus.c is in the
public domain and can presently be downloaded from
http://csgsoft.doc.ic.ac.uk/numcpus/
Signed-off-by: Andreas Ericsson <ae@op5.se> Signed-off-by: Brandon Casey <casey@nrlssc.navy.mil> Signed-off-by: Junio C Hamano <gitster@pobox.com>
start_command(), if .in/.out > 0, closes file descriptors, not the callers
Callers of start_command() can set the members .in and .out of struct
child_process to a value > 0 to specify that this descriptor is used as
the stdin or stdout of the child process.
Previously, if start_command() was successful, this descriptor was closed
upon return. Here we now make sure that the descriptor is also closed in
case of failures. All callers are updated not to close the file descriptor
themselves after start_command() was called.
Note that earlier run_gpg_verify() of git-verify-tag set .out = 1, which
worked because start_command() treated this as a special case, but now
this is incorrect because it closes the descriptor. The intent here is to
inherit stdout to the child, which is achieved by .out = 0.
Signed-off-by: Johannes Sixt <johannes.sixt@telecom.at> Signed-off-by: Junio C Hamano <gitster@pobox.com>
start_command(), .in/.out/.err = -1: Callers must close the file descriptor
By setting .in, .out, or .err members of struct child_process to -1, the
callers of start_command() can request that a pipe is allocated that talks
to the child process and one end is returned by replacing -1 with the
file descriptor.
Previously, a flag was set (for .in and .out, but not .err) to signal
finish_command() to close the pipe end that start_command() had handed out,
so it was optional for callers to close the pipe, and many already do so.
Now we make it mandatory to close the pipe.
Signed-off-by: Johannes Sixt <johannes.sixt@telecom.at> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Merge branch 'maint' of git://repo.or.cz/git-gui into maint
* 'maint' of git://repo.or.cz/git-gui:
git-gui: Focus insertion point at end of strings in repository chooser
git-gui: Avoid hardcoded Windows paths in Cygwin package files
git-gui: Default TCL_PATH to same location as TCLTK_PATH
git-gui: Paper bag fix error dialogs opening over the main window
git-gui: Ensure error dialogs always appear over all other windows
git-gui: relax "dirty" version detection
git-gui: support Git Gui.app under OS X 10.5
builtin-reflog.c: don't install new reflog on write failure
When expiring reflog entries, a new temporary log is written which contains
only the entries to retain. After it is written, it is renamed to replace
the existing reflog. Currently, we check that writing of the new log is
successful and print a message on failure, but the original reflog is still
replaced with the new reflog even on failure. This patch causes the
original reflog to be retained if we fail when writing the new reflog.
Signed-off-by: Brandon Casey <casey@nrlssc.navy.mil> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Use helper function for copying index entry information
We used to just memcpy() the index entry when we copied the stat() and
SHA1 hash information, which worked well enough back when the index
entry was just an exact bit-for-bit representation of the information on
disk.
However, these days we actually have various management information in
the cache entry too, and we should be careful to not overwrite it when
we copy the stat information from another index entry.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Name hash fixups: export (and rename) remove_hash_entry
This makes the name hash removal function (which really just sets the
bit that disables lookups of it) available to external routines, and
makes read_cache_unmerged() use it when it drops an unmerged entry from
the index.
It's renamed to remove_index_entry(), and we drop the (unused) 'istate'
argument.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
We handled the case of removing and re-inserting cache entries badly,
which is something that merging commonly needs to do (removing the
different stages, and then re-inserting one of them as the merged
state).
We even had a rather ugly special case for this failure case, where
replace_index_entry() basically turned itself into a no-op if the new
and the old entries were the same, exactly because the hash routines
didn't handle it on their own.
So what this patch does is to not just have the UNHASHED bit, but a
HASHED bit too, and when you insert an entry into the name hash, that
involves:
- clear the UNHASHED bit, because now it's valid again for lookup
(which is really all that UNHASHED meant)
- if we're being lazy, we're done here (but we still want to clear the
UNHASHED bit regardless of lazy mode, since we can become unlazy
later, and so we need the UNHASHED bit to always be set correctly,
even if we never actually insert the entry into the hash list)
- if it was already hashed, we just leave it on the list
- otherwise mark it HASHED and insert it into the list
this all means that unhashing and rehashing a name all just works
automatically. Obviously, you cannot change the name of an entry (that
would be a serious bug), but nothing can validly do that anyway (you'd
have to allocate a new struct cache_entry anyway since the name length
could change), so that's not a new limitation.
The code actually gets simpler in many ways, although the lazy hashing
does mean that there are a few odd cases (ie something can be marked
unhashed even though it was never on the hash in the first place, and
isn't actually marked hashed!).
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
This change removes all obvious useless if-before-free tests.
E.g., it replaces code like this:
if (some_expression)
free (some_expression);
with the now-equivalent:
free (some_expression);
It is equivalent not just because POSIX has required free(NULL)
to work for a long time, but simply because it has worked for
so long that no reasonable porting target fails the test.
Here's some evidence from nearly 1.5 years ago:
Note however, that it doesn't handle brace-enclosed blocks like
"if (x) { free (x); }". But that's ok, since there were none like
that in git sources.
Beware: if you do use the above snippet, note that it can
produce syntactically invalid C code. That happens when the
affected "if"-statement has a matching "else".
E.g., it would transform this
if (x)
free (x);
else
foo ();
into this:
free (x);
else
foo ();
There were none of those here, either.
If you're interested in automating detection of the useless
tests, you might like the useless-if-before-free script in gnulib:
[it *does* detect brace-enclosed free statements, and has a --name=S
option to make it detect free-like functions with different names]
We were returning the _address of_ the stored item (or NULL)
instead of the item itself. While this sort of indirection
is useful for insertion (since you can lookup and then
modify), it is unnecessary for read-only lookup. Since the
hash code splits these functions between the internal
lookup_hash_entry function and the public lookup_hash
function, it makes sense for the latter to provide what
users of the library expect.
The result of this was that the index caching returned bogus
results on lookup. We unfortunately didn't catch this
because we were returning a "struct cache_entry **" as a
"void *", and accidentally assigning it to a "struct
cache_entry *".
As it happens, this actually _worked_ most of the time,
because the entries were defined as:
meaning that interpreting a "struct cache_entry **" as a
"struct cache_entry *" would yield an entry where all fields
were totally bogus _except_ for the next pointer, which
pointed to the actual cache entry. When walking the list, we
would look at the bogus "name" field, which was unlikely to
match our lookup, and then proceed to the "real" entry.
The reading of bogus data was silently ignored most of the
time, but could cause a segfault for some data (which seems
to be more common on OS X).
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>