shortlog: take the first populated line of the description
Way back the perl version of shortlog would take the first populated line
of the commit body. The builtin version mearly takes the first line.
This leads to empty shortlog entries when there is some viable text in
the commit.
Reinstate this behaviour igoring all lines with nothing but whitespace.
This is often useful when dealing with commits imported from foreign SCMs
that do not tidy up the log message of useless blank lines at the beginning.
Signed-off-by: Andy Whitcroft <apw@shadowen.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Fix 'git remote show' regression on empty repository in 1.5.4
Back in 18f7c51c we switched git-ls-remote/git-peek-remote to
use the transport backend, rather than do everything itself.
As part of that switch we started to produce a non-zero exit
status if no refs were received from the remote peer, which
happens when the remote peer has no commits pushed to it yet.
(E.g. "git --git-dir=foo.git init; git ls-remote foo.git")
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
A merge is not necessarily with a remote branch, it can be with any
commit.
Thanks to Paolo Ciarrocchi for pointing out the problem, and to
Nicolas Pitre for pointing out the fact that a merge is not
necessarily with a branch head.
Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr> Signed-off-by: Junio C Hamano <gitster@pobox.com>
git-merge.sh: better handling of combined --squash,--no-ff,--no-commit options
git-merge used to use either the --squash,--no-squash, --no-ff,--ff,
--no-commit,--commit option, whichever came last in the command line.
This lead to some un-intuitive behavior, having
git merge --no-commit --no-ff <branch>
actually commit the merge. Now git-merge respects --no-commit together
with --no-ff, as well as other combinations of the options. However,
this broke a selftest in t/t7600-merge.sh which expected to have --no-ff
completely override the --squash option, so that
git merge --squash --no-ff <branch>
fast-forwards, and makes a merge commit; combining --squash with --no-ff
doesn't really make sense though, and is now refused by git-merge. The
test is adapted to test --no-ff without the preceding --squash, and
another test is added to make sure the --squash --no-ff combination is
refused.
The unexpected behavior was reported by John Goerzen through
http://bing.sdebian.org/468568
Signed-off-by: Gerrit Pape <pape@smarden.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
For some reason, http_cleanup was running all active slots, which could
lead in situations where a freed slot would be accessed in
fill_active_slots. OTOH, we are cleaning up, which means the caller
doesn't care about pending requests. Just forget about them instead
or running them.
Signed-off-by: Mike Hommey <mh@glandium.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The previous code mistakenly used wt_status_prepare to check whether the
index had anything commitable in it; however, that function is just an
init function, and will never report a dirty index.
The correct way with wt_status_* would be to call wt_status_print with the
output pointing to /dev/null or similar. However, that does extra work by
both examining the working tree and spewing status information to nowhere.
Instead, let's just implement the useful subset of wt_status_print as an
"is_index_dirty" function.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
When we expect a git command to notice and signal errors, we
carelessly wrote in our tests:
test_expect_success 'reject bogus request' '
do something &&
do something else &&
! git command
'
but a non-zero exit could come from the "git command" segfaulting.
A new helper function "tset_must_fail" is introduced and it is
meant to be used to make sure the command gracefully fails (iow,
dying and exiting with non zero status is counted as a failure
to "gracefully fail"). The above example should be written as:
test_expect_success 'reject bogus request' '
do something &&
do something else &&
test_must_fail git command
'
receive-pack: Initialize PATH to include exec-dir.
511707d (use only the $PATH for exec'ing git commands) made it a
requirement to call setup_path() to include the git exec-dir in PATH
before spawning any other git commands. git-receive-pack was not yet
adapted to do this and therefore fails to spawn git-unpack-objects if that
is not in the standard PATH.
Signed-off-by: Björn Steinbrink <B.Steinbrink@gmx.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
git.el: find the git-status buffer whatever its name is
git-status used the buffer name to find git-status buffers, and that
can fail if the buffer has another name, for example when multiple
working directories is tracked.
Signed-off-by: Rémi Vanicat <vanicat@debian.org> Acked-by: Alexandre Julliard <julliard@winehq.org> Tested-by: Xavier Maillard <xma@gnu.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
templates/Makefile: don't depend on local umask setting
Don't take the local umask setting into account when installing the
templates/* files and directories, running 'make install' with umask set
to 077 resulted in template/* installed with permissions 700 and 600.
The problem was discovered by Florian Zumbiehl, reported through
http://bugs.debian.org/467518
Signed-off-by: Gerrit Pape <pape@smarden.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
git-gui: Paper bag fix info dialog when no files are staged at commit
If the user tries to commit their changes without actually staging
anything we used to display an informational dialog suggesting they
first stage those changes, then retry the commit feature.
Unfortunately I broke this in aba15f7 ("Ensure error dialogs always
appear over all other windows") and failed to fix it in the paper
bag fix that came one day after it.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
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>
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>
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>
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>
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>
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>
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
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>
git-gui: Focus insertion point at end of strings in repository chooser
When selecting a local working directory for a new repository or a
location to clone an existing repository into we now set the insert
point at the end of the selected path, allowing the user to type in
any additional parts of the path if they so desire.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
git-gui: Avoid hardcoded Windows paths in Cygwin package files
When we are being built by the Cygwin package maintainers we need to
embed the POSIX path to our library files and not the Windows path.
Embedding the Windows path means all end-users who install our Cygwin
package would be required to install Cygwin at the same Windows path
as the package maintainer had Cygwin installed to. This requirement
is simply not user-friendly and may be infeasible for a large number
of our users.
We now try to auto-detect if the Tcl/Tk binary we will use at runtime
is capable of translating POSIX paths into Windows paths the same way
that cygpath does the translations. If the Tcl/Tk binary gives us the
same results then it understands the Cygwin path translation process
and should be able to read our library files from a POSIX path name.
If it does not give us the same answer as cygpath then the Tcl/Tk
binary might actually be a native Win32 build (one that is not
linked against Cygwin) and thus requires the native Windows path
to our library files. We can assume this is not a Cygwin package
as the Cygwin maintainers do not currently ship a pure Win32 build
of Tcl/Tk.
Reported on the git mailing list by Jurko Gospodnetić.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
git-gui: Default TCL_PATH to same location as TCLTK_PATH
Most users set TCLTK_PATH to tell git-gui where to find wish, but they
fail to set TCL_PATH to the same Tcl installation. We use the non-GUI
tclsh during builds so headless systems are still able to create an
index file and create message files without GNU msgfmt. So it matters
to us that we find a working TCL_PATH at build time.
If TCL_PATH hasn't been set yet we can take a better guess about what
tclsh executable to use by replacing 'wish' in the executable path with
'tclsh'. We only do this replacement on the filename part of the path,
just in case the string "wish" appears in the directory paths. Most of
the time the tclsh will be installed alongside wish so this replacement
is a sensible and safe default.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
git-gui: Paper bag fix error dialogs opening over the main window
If the main window is the only toplevel we have open then we
don't have a valid grab right now, so we need to assume the
best toplevel to use for the parent is ".".
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
git-gui: Ensure error dialogs always appear over all other windows
If we are opening an error dialog we want it to appear above all of
the other windows, even those that we may have opened with a grab
to make the window modal. Failure to do so may allow an error
dialog to open up (and grab focus!) under an existing toplevel,
making the user think git-gui has frozen up and is unresponsive,
as they cannot get to the dialog.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Clarified the meaning of git-add -u in the documentation
The git-add documentation did not state clearly that the -u switch
updates only the tracked files that are in the current directory and
its subdirectories.
Signed-off-by: Pekka Kaitaniemi <kaitanie@cc.helsinki.fi> Signed-off-by: Junio C Hamano <gitster@pobox.com>
git-clone.sh: properly configure remote even if remote's head is dangling
When cloning a remote repository which's HEAD refers to a nonexistent
ref, git-clone cloned all existing refs, but failed to write the
configuration for 'remote'. Now it detects the dangling remote HEAD,
refuses to checkout any local branch since HEAD refers to nowhere, but
properly writes the configuration for 'remote', so that subsequent
'git fetch's don't fail.
The problem was reported by Daniel Jacobowitz through
http://bugs.debian.org/466581
Signed-off-by: Gerrit Pape <pape@smarden.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
send-email: squelch warning due to comparing undefined $_ to ""
The check to see if initial_reply_to is defined was also comparing $_ to
"" for a reason I cannot ascertain (looking at the commit which made the
change didn't provide enlightenment), but if $_ is undefined, perl
generates a warning.
Signed-off-by: Jay Soffian <jaysoffian@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Rename git-core rpm to just git and rename the meta-pacakge to git-all.
This fixes my favorite annoyance with the git rpm packaging: don't pull
in tla when I say yum install git! You wouldn't expect yum install gcc
to pull in gcc-gfortran, right?
With this change, and blanket 'yum update' will automatically pull in the
new 'git' package and push out the old 'git-core', and if the old 'git'
package was installed 'git-all' will be pulled in instead. A couple of
things do break though: 'yum update git-core', because yum behaves
differently when given a specific package name - it doesn't follow obsoletes.
Instead, 'yum install git' will pull in the new git rpm, which will then
push out the old 'git-core'. Similarly, to get the newest version of
the meta package, 'yum install git-all' will install git-all, which then
pushes out the old 'git' meta package.
Signed-off-by: Kristian Høgsberg <krh@redhat.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The output was meant to be a balance of self-explanatory and
terse. In case we have erred too far on the terse side, it
doesn't hurt to explain in more detail what each line means.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The previous text was correct, but it was easy to miss the
fact that we are talking about "matching" refs. That is, the
text can be parsed as "we push the union of the sets
of remote and local heads" and not "we push the intersection
of the sets of remote and local heads". (The former actually
doesn't make sense if you think about it, since we don't
even _have_ some of those heads). A careful reading would
reveal the correct meaning, but it makes sense to be as
explicit as possible in documentation.
We also explicitly use and introduce the term "matching";
this is a term discussed on the list, and it seems useful
to for users to be able to refer to this behavior by name.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The existing message indicates that an error occured during
push, but it is unclear whether _any_ refs were actually
pushed (even though the status table above shows which were
pushed successfully and which were not, the message "failed
to push" implies a total failure). By indicating that "some
refs" failed, we hopefully indicate to the user that the
table above contains the details.
We could also put in an explicit "see above for details"
message, but it seemed to clutter the output quite a bit
(both on a line of its own, or at the end of the error line,
which inevitably wraps).
This could also be made more fancy if the transport
mechanism passed back more details on how many refs
succeeded and failed:
error: failed to push %d out of %d refs to '%s'
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
"git gui" would complain at launch if the local version of Git was
"1.5.4.2.dirty". Loosen the regular expression to look for either
"-dirty" or ".dirty", thus eliminating spurious warnings.
Signed-off-by: Wincent Colaiuta <win@wincent.com> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Documentation/git-reset: don't mention --mixed for selected-paths reset
The option is accepted, but that is the only form selected-paths
variant of the reset command takes, so there is no point mentioning it.
And while we're at it, use the dashless git call.
Signed-off-by: Pieter de Bie <pdebie@ai.rug.nl> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Since 3368d11 (Remove unnecessary git-rm --cached reference from
status output), the status output marks the "Added but not yet
committed" section as "Changes to be committed".
commit: discard index after setting up partial commit
There may still be some entries from the original index that
should be discarded before we show the status. In
particular, if a file was added in the index but not
included in the partial commit, it would still show up in
the status listing as staged for commit.
Ultimately the correct fix is to keep the two states in
separate index_state variables. Then we can avoid having
to reload the cache from the temporary file altogether, and
just point wt_status_print at the correct index.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The command used a very old fashioned construct to extract
filenames out of diff-index and ended up corrupting the output.
We can simply use --name-only and pipe into --stdin mode of
update-index. It's been like that for the past 2 years or so
since a94d994 (update-index: work with c-quoted name).
Before objects are sent, the respective ref is locked. However,
without this patch, the lock is lifted before the last object for
that ref was sent. As a consequence, the lock data was accessed
after the lock structure was free()d.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
c1795bb (Unify whitespace checking) incorrectly made the
checking function return without incrementing the line numbers
when there is no whitespace problem is found on a '+' line.
This resurrects the earlier behaviour.
Noticed and reported by Jay Soffian. The test script was stolen
from Jay's independent fix.
Fix a bug in the hg-to-git convertor introduced by commit 1bc7c13af9f936aa80893100120b542338a10bf4: when searching the changeset
parents, 'hg log' returns an extra space at the end of the line, which
confuses the .split(' ') based tokenizer:
Traceback (most recent call last):
File "hg-to-git.py", line 123, in <module>
hgchildren[mparent] += ( str(cset), )
KeyError: ''
Signed-off-by: Stelian Pop <stelian@popies.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
mailinfo: feed only one line to handle_filter() for QP input
The function is intended to be fed one logical line at a time to
inspect, but a QP encoded raw input line can have more than one
lines, just like BASE64 encoded one.
Quoting LF as =0A may be unusual but RFC2045 allows it.
The issue was noticed and fixed by Jay Soffian. JC added a test
to protect the fix from regressing later.
Signed-off-by: Jay Soffian <jaysoffian@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
config: add 'git_config_string' to refactor string config variables.
In many places we just check if a value from the config file is not
NULL, then we duplicate it and return 0. This patch introduces the new
'git_config_string' function to do that.
This function is also used to refactor some code in 'config.c'.
Refactoring other files is left for other patches.
Also not all the code in "config.c" is refactored, because the function
takes a "const char **" as its first parameter, but in many places a
"char *" is used instead of a "const char *". (And C does not allow
using a "char **" instead of a "const char **" without a warning.)
Signed-off-by: Christian Couder <chriscool@tuxfamily.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
It is not necessary to check if value != NULL before calling
'parse_lldiff_command' as there is already a check inside this
function.
By the way this patch also improves the existing check inside
'parse_lldiff_command' by using:
return config_error_nonbool(var);
instead of:
return error("%s: lacks value", var);
Signed-off-by: Christian Couder <chriscool@tuxfamily.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
fast-import: check return value from unpack_entry()
If the tree object we have asked for is deltafied in the packfile and
the delta did not apply correctly or was not able to be decompressed
from the packfile then we can get back NULL instead of the tree data.
This is (part of) the reason why read_sha1_file() can return NULL, so
we need to also handle it the same way.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Validate nicknames of remote branches to prohibit confusing ones
The original problem was that the parsers for configuration files were
getting confused by seeing as nicknames remotes that involved
directory-changing characters. In particular, the branches config file
for ".." was particularly mystifying on platforms that can open
directories and read odd data from them.
The validation function was written by Junio Hamano (with a typo
corrected).
Signed-off-by: Daniel Barkalow <barkalow@iabervon.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
diff.c: fixup garding of config parser from value=NULL
Christian Couder noticed that there still were a handcrafted error()
call that we should have converted to config_error_nonbool() where
parse_lldiff_command() parses the configuration file.
cvsimport: have default merge regex also match beginning of commit message
The default value of @mergerx uses \W, which matches a non-word
character; this means that commit messages like "Merging FOO" are not
matched by default; using \b, which matches a word boundary, instead of
\W fixes that.
This change was suggested by Frédéric Brière through
http://bugs.debian.org/463468
Signed-off-by: Gerrit Pape <pape@smarden.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
status: suggest "git rm --cached" to unstage for initial commit
It makes no sense to suggest "git reset HEAD" since we have
no HEAD commit. This actually used to work but regressed in f26a0012.
wt_status_print_cached_header was updated to take the whole
wt_status struct rather than just the reference field.
Previously the various code paths were sometimes sending in
s->reference and sometimes sending in NULL, making the
decision on whether this was an initial commit before we
even got to this function. Now we must check the initial
flag here.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Protect get_author_ident_from_commit() from filenames in work tree
We used to use "cat-file commit $commit" to extract the original
author information from existing commit, but an earlier commit 5ac2715 (Consistent message encoding while reusing log from an
existing commit) changed it to use "git show -s $commit". If
you have a file in your work tree that can be interpreted as a
valid object name (e.g. "HEAD"), this conversion will not work.
Disambiguate by marking the end of revision parameter on the
comand line with an explicit "--" to fix this.
This breakage is most visible with rebase when a file called
"HEAD" exists in the worktree.
Since git-upload-pack has to spawn git-pack-objects, it has to make sure
that the latter can be found in the PATH. Without this patch an attempt
to clone or pull via ssh from a server fails if the git tools are not in
the standard PATH on the server even though git clone or git pull were
invoked with --upload-pack=/path/to/git-upload-pack.
Signed-off-by: Johannes Sixt <johannes.sixt@telecom.at> Signed-off-by: Junio C Hamano <gitster@pobox.com>
bisect: use verbatim commit subject in the bisect log
Due to a typo, the commit subject was shell expanded in the bisect log.
That is, if you had some shell pattern in the commit subject, bisect
would happily put all matching file names into the log.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Tested-by: Frans Pop <elendil@planet.nl> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Nico analyzed and found out that this does not really help, and
I agree with it.
By the time this gets into action and data is actively thrown
away, performance simply goes down the drain due to the data
constantly being reloaded over and over and over and over and
over and over again, to the point of virtually making no
relative progress at all. The previous behavior of enforcing
the memory limit by dynamically shrinking the window size at
least had the effect of allowing some kind of progress, even if
the end result wouldn't be optimal.
And that's the whole point behind this memory limiting feature:
allowing some progress to be made when resources are too limited
to let the repack go unbounded.