pathspec: don't error out on all-exclusionary pathspec patterns
Instead of erroring out and telling the user that they should add a
positive pattern that covers everything else, just _do_ that.
For commands where we honor the current cwd by default (ie grep, ls-files
etc), we make that default positive pathspec be the current working
directory. And for commands that default to the whole project (ie diff,
log, etc), the default positive pathspec is the whole project.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The choice of '!' for a negative pathspec ends up not only not matching
what we do for revisions, it's also a horrible character for shell
expansion since it needs quoting.
So add '^' as an alternative alias for an excluding pathspec entry.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The "core.logAllRefUpdates" that used to be boolean has been
enhanced to take 'always' as well, to record ref updates to refs
other than the ones that are expected to be updated (i.e. branches,
remote-tracking branches and notes).
* cw/log-updates-for-all-refs-really:
doc: add note about ignoring '--no-create-reflog'
update-ref: add test cases for bare repository
refs: add option core.logAllRefUpdates = always
config: add markup to core.logAllRefUpdates doc
"uchar [40]" to "struct object_id" conversion continues.
* rs/object-id:
checkout: convert post_checkout_hook() to struct object_id
use oidcpy() for copying hashes between instances of struct object_id
use oid_to_hex_r() for converting struct object_id hashes to hex strings
"make -C t failed" will now run only the tests that failed in the
previous run. This is usable only when prove is not use, and gives
a useless error message when run after "make clean", but otherwise
is serviceable.
* js/re-running-failed-tests:
t/Makefile: add a rule to re-run previously-failed tests
The user can specify a custom update method that is run when
"submodule update" updates an already checked out submodule. This
was ignored when checking the submodule out for the first time and
we instead always just checked out the commit that is bound to the
path in the superproject's index.
* sb/submodule-update-initial-runs-custom-script:
submodule update: run custom update script for initial populating as well
When a submodule "A", which has another submodule "B" nested within
it, is "absorbed" into the top-level superproject, the inner
submodule "B" used to be left in a strange state. The logic to
adjust the .git pointers in these submodules has been corrected.
* sb/submodule-recursive-absorb:
submodule absorbing: fix worktree/gitdir pointers recursively for non-moves
cache.h: expose the dying procedure for reading gitlinks
setup: add gentle version of resolve_git_dir
Some people feel the default set of colors used by "git log --graph"
rather limiting. A mechanism to customize the set of colors has
been introduced.
* nd/log-graph-configurable-colors:
document behavior of empty color name
color_parse_mem: allow empty color spec
log --graph: customize the graph lines with config log.graphColors
color.c: trim leading spaces in color_parse_mem()
color.c: fix color_parse_mem() with value_len == 0
* ep/commit-static-buf-cleanup:
builtin/commit.c: switch to strbuf, instead of snprintf()
builtin/commit.c: remove the PATH_MAX limitation via dynamic allocation
Asciidoctor, an alternative reimplementation of AsciiDoc, still
needs some changes to work with documents meant to be formatted
with AsciiDoc. "make USE_ASCIIDOCTOR=YesPlease" to use it out of
the box to document our pages is getting closer to reality.
* bc/use-asciidoctor-opt:
Documentation: implement linkgit macro for Asciidoctor
Makefile: add a knob to enable the use of Asciidoctor
Documentation: move dblatex arguments into variable
Documentation: add XSLT to fix DocBook for Texinfo
Documentation: sort sources for gitman.texi
Documentation: remove unneeded argument in cat-texi.perl
Documentation: modernize cat-texi.perl
Documentation: fix warning in cat-texi.perl
Names of the various hook scripts must be spelled exactly, but on
Windows, an .exe binary must be named with .exe suffix; notice
$GIT_DIR/hooks/<hookname>.exe as a valid <hookname> hook.
* js/mingw-hooks-with-exe-suffix:
mingw: allow hooks to be .exe files
After starting "git rebase -i", which first opens the user's editor
to edit the series of patches to apply, but before saving the
contents of that file, "git status" failed to show the current
state (i.e. you are in an interactive rebase session, but you have
applied no steps yet) correctly.
* js/status-pre-rebase-i:
status: be prepared for not-yet-started interactive rebase
Merge branch 'jk/execv-dashed-external' into maint
Typing ^C to pager, which usually does not kill it, killed Git and
took the pager down as a collateral damage in certain process-tree
structure. This has been fixed.
* jk/execv-dashed-external:
execv_dashed_external: wait for child on signal death
execv_dashed_external: stop exiting with negative code
execv_dashed_external: use child_process struct
Commit 55cccf4bb (color_parse_mem: allow empty color spec,
2017-02-01) clearly defined the behavior of an empty color
config variable. Let's document that, and give a hint about
why it might be useful.
It's important not to say that it makes the item uncolored,
because it doesn't. It just sets no attributes, which means
that any previous attributes continue to take effect.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The commands git-branch and git-tag accept the '--create-reflog'
option, and create reflog even when core.logallrefupdates
configuration is explicitly set not to.
On the other hand, the negated form '--no-create-reflog' is accepted
as a valid option but has no effect (other than overriding an
earlier '--create-reflog' on the command line). This silent noop may
puzzle users. To communicate that this is a known limitation, add a
short note in the manuals for git-branch and git-tag.
Signed-off-by: Cornelius Weig <cornelius.weig@tngtech.com> Helped-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Prior to c2f41bf52 (color.c: fix color_parse_mem() with
value_len == 0, 2017-01-19), the empty string was
interpreted as a color "reset". This was an accidental
outcome, and that commit turned it into an error.
However, scripts may pass the empty string as a default
value to "git config --get-color" to disable color when the
value is not defined. The git-add--interactive script does
this. As a result, the script is unusable since c2f41bf52
unless you have color.diff.plain defined (if it is defined,
then we don't parse the empty default at all).
Our test scripts didn't notice the recent breakage because
they run without a terminal, and thus without color. They
never hit this code path at all. And nobody noticed the
original buggy "reset" behavior, because it was effectively
a noop.
Let's fix the code to have an empty color name produce an
empty sequence of color codes. The tests need a few fixups:
- we'll add a new test in t4026 to cover this case. But
note that we need to tweak the color() helper. While
we're there, let's factor out the literal ANSI ESC
character. Otherwise it makes the diff quite hard to
read.
- we'll add a basic sanity-check in t4026 that "git add
-p" works at all when color is enabled. That would have
caught this bug, as well as any others that are specific
to the color code paths.
- 73c727d69 (log --graph: customize the graph lines with
config log.graphColors, 2017-01-19) added a test to
t4202 that checks some "invalid" graph color config.
Since ",, blue" before yielded only "blue" as valid, and
now yields "empty, empty, blue", we don't match the
expected output.
One way to fix this would be to change the expectation
to the empty color strings. But that makes the test much
less interesting, since we show only two graph lines,
both of which would be colorless.
Since the empty-string case is now covered by t4026,
let's remove them entirely here. They're just in the way
of the primary thing the test is supposed to be
checking.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Merge branch 'jk/archive-zip-userdiff-config' into maint
"git archive" did not read the standard configuration files, and
failed to notice a file that is marked as binary via the userdiff
driver configuration.
Merge branch 'dt/disable-bitmap-in-auto-gc' into maint
It is natural that "git gc --auto" may not attempt to pack
everything into a single pack, and there is no point in warning
when the user has configured the system to use the pack bitmap,
leading to disabling further "gc".
* dt/disable-bitmap-in-auto-gc:
repack: die on incremental + write-bitmap-index
auto gc: don't write bitmaps for incremental repacks
Leakage of lockfiles in the config subsystem has been fixed.
* nd/config-misc-fixes:
config.c: handle lock file in error case in git_config_rename_...
config.c: rename label unlock_and_out
config.c: handle error case for fstat() calls
Compression setting for producing packfiles were spread across
three codepaths, one of which did not honor any configuration.
Unify these so that all of them honor core.compression and
pack.compression variables the same way.
Merge branch 'jk/make-tags-find-sources-tweak' into maint
Update the procedure to generate "tags" for developer support.
* jk/make-tags-find-sources-tweak:
Makefile: exclude contrib from FIND_SOURCE_FILES
Makefile: match shell scripts in FIND_SOURCE_FILES
Makefile: exclude test cruft from FIND_SOURCE_FILES
Makefile: reformat FIND_SOURCE_FILES
Some platforms no longer understand "latin-1" that is still seen in
the wild in e-mail headers; replace them with "iso-8859-1" that is
more widely known when conversion fails from/to it.
* jc/latin-1:
utf8: accept "latin-1" as ISO-8859-1
utf8: refactor code to decide fallback encoding
"git fsck --connectivity-check" was not working at all.
* jk/fsck-connectivity-check-fix:
fsck: lazily load types under --connectivity-only
fsck: move typename() printing to its own function
t1450: use "mv -f" within loose object directory
fsck: check HAS_OBJ more consistently
fsck: do not fallback "git fsck <bogus>" to "git fsck"
fsck: tighten error-checks of "git fsck <head>"
fsck: prepare dummy objects for --connectivity-check
fsck: report trees as dangling
t1450: clean up sub-objects in duplicate-entry test
* js/difftool-builtin:
difftool: hack around -Wzero-length-format warning
difftool: retire the scripted version
difftool: implement the functionality in the builtin
difftool: add a skeleton for the upcoming builtin
A few codepaths had to rely on a global variable when sorting
elements of an array because sort(3) API does not allow extra data
to be passed to the comparison function. Use qsort_s() when
natively available, and a fallback implementation of it when not,
to eliminate the need, which is a prerequisite for making the
codepath reentrant.
* rs/qsort-s:
ref-filter: use QSORT_S in ref_array_sort()
string-list: use QSORT_S in string_list_sort()
perf: add basic sort performance test
add QSORT_S
compat: add qsort_s()
"git show-ref HEAD" used with "--verify" because the user is not
interested in seeing refs/remotes/origin/HEAD, and used with
"--head" because the user does not want HEAD to be filtered out,
i.e. "git show-ref --head --verify HEAD", did not work as expected.
* vp/show-ref-verify-head:
show-ref: remove a stale comment
show-ref: remove dead `if (verify)' check
show-ref: detect dangling refs under --verify as well
show-ref: move --quiet handling into show_one()
show-ref: allow -d to work with --verify
show-ref: accept HEAD with --verify
With anticipatory tweaking for remotes defined in ~/.gitconfig
(e.g. "remote.origin.prune" set to true, even though there may or
may not actually be "origin" remote defined in a particular Git
repository), "git remote rename" and other commands misinterpreted
and behaved as if such a non-existing remote actually existed.
* js/remote-rename-with-half-configured-remote:
remote rename: more carefully determine whether a remote is configured
remote rename: demonstrate a bogus "remote exists" bug
The sequencer machinery has been further enhanced so that a later
set of patches can start using it to reimplement "rebase -i".
* js/sequencer-i-countdown-3: (38 commits)
sequencer (rebase -i): write out the final message
sequencer (rebase -i): write the progress into files
sequencer (rebase -i): show the progress
sequencer (rebase -i): suggest --edit-todo upon unknown command
sequencer (rebase -i): show only failed cherry-picks' output
sequencer (rebase -i): show only failed `git commit`'s output
sequencer: use run_command() directly
sequencer: update reading author-script
sequencer (rebase -i): differentiate between comments and 'noop'
sequencer (rebase -i): implement the 'drop' command
sequencer (rebase -i): allow rescheduling commands
sequencer (rebase -i): respect strategy/strategy_opts settings
sequencer (rebase -i): respect the rebase.autostash setting
sequencer (rebase -i): run the post-rewrite hook, if needed
sequencer (rebase -i): record interrupted commits in rewritten, too
sequencer (rebase -i): copy commit notes at end
sequencer (rebase -i): set the reflog message consistently
sequencer (rebase -i): refactor setting the reflog message
sequencer (rebase -i): allow fast-forwarding for edit/reword
sequencer (rebase -i): implement the 'reword' command
...
"git submodule push" learned "--recurse-submodules=only option to
push submodules out without pushing the top-level superproject.
* bw/push-submodule-only:
push: add option to push only submodules
submodules: add RECURSE_SUBMODULES_ONLY value
transport: reformat flag #defines to be more readable
An error message with an ASCII control character like '\r' in it
can alter the message to hide its early part, which is problematic
when a remote side gives such an error message that the local side
will relay with a "remote: " prefix.
Documentation: implement linkgit macro for Asciidoctor
AsciiDoc uses a configuration file to implement macros like linkgit,
while Asciidoctor uses Ruby extensions. Implement a Ruby extension that
implements the linkgit macro for Asciidoctor in the same way that
asciidoc.conf does for AsciiDoc. Adjust the Makefile to use it by
default.
Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
builtin/commit.c: switch to strbuf, instead of snprintf()
Switch to dynamic allocation with strbuf, so we can avoid dealing
with magic numbers in the code and reduce the cognitive burden from
the programmers. The original code is correct, but programmers no
longer have to count bytes needed for static allocation to know that.
As a side effect of this change, we also reduce the snprintf()
calls, that may silently truncate results if the programmer is not
careful.
Helped-by: René Scharfe <l.s.r@web.de> Helped-by: Junio C Hamano <gitster@pobox.com> Helped-by: Jeff King <peff@peff.net> Signed-off-by: Elia Pinto <gitter.spiros@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The default behavior of update-ref to create reflogs differs in
repositories with worktree and bare ones. The existing tests cover only
the behavior of repositories with worktree.
This commit adds tests that assert the correct behavior in bare
repositories for update-ref. Two cases are covered:
- If core.logAllRefUpdates is not set, no reflogs should be created
- If core.logAllRefUpdates is true, reflogs should be created
Signed-off-by: Cornelius Weig <cornelius.weig@tngtech.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
When core.logallrefupdates is true, we only create a new reflog for refs
that are under certain well-known hierarchies. The reason is that we
know that some hierarchies (like refs/tags) are not meant to change, and
that unknown hierarchies might not want reflogs at all (e.g., a
hypothetical refs/foo might be meant to change often and drop old
history immediately).
However, sometimes it is useful to override this decision and simply log
for all refs, because the safety and audit trail is more important than
the performance implications of keeping the log around.
This patch introduces a new "always" mode for the core.logallrefupdates
option which will log updates to everything under refs/, regardless
where in the hierarchy it is (we still will not log things like
ORIG_HEAD and FETCH_HEAD, which are known to be transient).
Based-on-patch-by: Jeff King <peff@peff.net> Signed-off-by: Cornelius Weig <cornelius.weig@tngtech.com> Reviewed-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
t0001: don't let a default ACL interfere with the umask test
The "init creates a new deep directory (umask vs. shared)" test expects
the permissions of newly created files to be based on the umask, which
fails if a default ACL is inherited from the working tree for git. So
attempt to remove a default ACL if there is one. Same idea as 8ed0a740dd42bd0724aebed6e3b07c4ea2a2d5e8. (I guess I'm the only one who
ever runs the test suite with a default ACL set.)
Signed-off-by: Matt McCutchen <matt@mattmccutchen.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
When running git-p4 on Windows, with multiple git-p4.mapUser entries in
git config - no user mappings are applied to the generated repository.
Reproduction Steps:
1. Add multiple git-p4.mapUser entries to git config on a Windows
machine
2. Attempt to clone a p4 repository
None of the user mappings will be applied.
This issue is actually caused by gitConfigList, using split(os.linesep)
to convert the output of git config --get-all into a list. On Windows,
os.linesep is equal to '\r\n' - however git.exe returns configuration
with a line seperator of '\n'.
This leads to the list returned by gitConfigList containing only one
element - which contains the full output of git config --get-all in
string form, which causes problems for the code introduced to
getUserMapFromPerforceServer in 10d08a149d ("git-p4: map a P4 user to
Git author name and email address", 2016-03-01)
This issue should be caught by the test introduced in 10d08a1, however
would require running on Windows to reproduce.
Using splitlines solves this issue, by splitting config on all
typical delimiters ('\n', '\r\n' etc.)
Signed-off-by: George Vanburgh <gvanburgh@bloomberg.net> Reviewed-by: Luke Diamand <luke@diamand.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
On Windows, executables need to have the file extension `.exe`, or they
are not executables. Hence, to support scripts, Git for Windows also
looks for a she-bang line by opening the file in question, and executing
it via the specified script interpreter.
To figure out whether files in the `PATH` are executable, `git help` has
code that imitates this behavior. With one exception: it *always* opens
the files and looks for a she-bang line *or* an `MZ` tell-tale
(nevermind that files with the magic `MZ` but without file extension
`.exe` would still not be executable).
Opening this many files leads to performance problems that are even more
serious when a virus scanner is running. Therefore, let's change the
code to look for the file extension `.exe` early, and avoid opening the
file altogether if we already know that it is executable.
See the following measurements (in seconds) as an example, where we
execute a simple program that simply lists the directory contents and
calls open() on every listed file:
With virus scanner running (coldcache):
$ ./a.exe /libexec/git-core/
before open (git-add.exe): 0.000000
after open (git-add.exe): 0.412873
before open (git-annotate.exe): 0.000175
after open (git-annotate.exe): 0.397925
before open (git-apply.exe): 0.000243
after open (git-apply.exe): 0.399996
before open (git-archive.exe): 0.000147
after open (git-archive.exe): 0.397783
before open (git-bisect--helper.exe): 0.000160
after open (git-bisect--helper.exe): 0.397700
before open (git-blame.exe): 0.000160
after open (git-blame.exe): 0.399136
...
With virus scanner running (hotcache):
$ ./a.exe /libexec/git-core/
before open (git-add.exe): 0.000000
after open (git-add.exe): 0.000325
before open (git-annotate.exe): 0.000229
after open (git-annotate.exe): 0.000177
before open (git-apply.exe): 0.000167
after open (git-apply.exe): 0.000150
before open (git-archive.exe): 0.000154
after open (git-archive.exe): 0.000156
before open (git-bisect--helper.exe): 0.000132
after open (git-bisect--helper.exe): 0.000180
before open (git-blame.exe): 0.000718
after open (git-blame.exe): 0.000724
...
With this patch I get:
$ time git help git
Launching default browser to display HTML ...
real 0m8.723s
user 0m0.000s
sys 0m0.000s
and without
$ time git help git
Launching default browser to display HTML ...
real 1m37.734s
user 0m0.000s
sys 0m0.031s
both tests with cold cache and giving the machine some time to settle
down after restart.
[jes: adjusted the commit message]
Signed-off-by: Heiko Voigt <heiko.voigt@mahr.de> Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Executable files in Windows need to have the extension '.exe', otherwise
they do not work. Extend the hooks to not just look at the hard coded
names, but also at the names extended by the custom STRIP_EXTENSION,
which is defined as '.exe' in Windows.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
doc: clarify distinction between sign-off and pgp-signing
The documentation for submission discourages pgp-signing, but demands
a proper sign-off by contributors. However, when skimming the headings,
the wording of the section for sign-off could mistakenly be understood
as concerning pgp-signing. Thus, new contributors could oversee the
necessary sign-off.
This commit improves the wording such that the section about sign-off
cannot be misunderstood as pgp-signing. In addition, the paragraph about
pgp-signing is changed such that it avoids the impression that
pgp-signing could be relevant at later stages of the submission.
Signed-off-by: Cornelius Weig <cornelius.weig@tngtech.com> Helped-by: Philip Oakley <philipoakley@iee.org> Helped-by: Stefan Beller <sbeller@google.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
test-lib: on FreeBSD, look for unzip(1) in /usr/local/bin/
Eric Wong reported that while FreeBSD has a /usr/bin/unzip, it uses
different semantics from those that are needed by Git's tests: When
passing the -a option to Info-Zip, it heeds the text attribute of the
.zip file's central directory, while FreeBSD's unzip ignores that
attribute.
The common work-around is to install Info-Zip on FreeBSD, into
/usr/local/bin/.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Tested-by: Eric Wong <e@80x24.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
t/Makefile: add a rule to re-run previously-failed tests
This patch automates the process of determining which tests failed
previously and re-running them.
While developing patch series, it is a good practice to run the test
suite from time to time, just to make sure that obvious bugs are caught
early. With complex patch series, it is common to run `make -j15 -k
test`, i.e. run the tests in parallel and *not* stop at the first
failing test but continue. This has the advantage of identifying
possibly multiple problems in one big test run.
It is particularly important to reduce the turn-around time thusly on
Windows, where the test suite spends 45 minutes on the computer on which
this patch was developed.
It is the most convenient way to determine which tests failed after
running the entire test suite, in parallel, to look for left-over "trash
directory.t*" subdirectories in the t/ subdirectory. However, those
directories might live outside t/ when overridden using the
--root=<directory> option, to which the Makefile has no access. The next
best method is to grep explicitly for failed tests in the test-results/
directory, which the Makefile *can* access.
Please note that the often-recommended `prove` tool requires Perl, and
that opens a whole new can of worms on Windows. As no native Windows Perl
comes with Subversion bindings, we have to use a Perl in Git for Windows
that uses the POSIX emulation layer named MSYS2 (which is a portable
version of Cygwin). When using this emulation layer under stress, e.g.
when running massively-parallel tests, unexplicable crashes occur quite
frequently, and instead of having a solution to the original problem, the
developer now has an additional, quite huge problem. For that reason, this
developer rejected `prove` as a solution and went with this patch instead.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Reviewed-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Apply the semantic patch for converting callers that duplicate the
result of absolute_path() to call absolute_pathdup() instead, which
avoids an extra string copy to a static buffer.
Signed-off-by: Rene Scharfe <l.s.r@web.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add a function that returns a buffer containing the absolute path of its
argument and a semantic patch for its intended use. It avoids an extra
string copy to a static buffer.
Signed-off-by: Rene Scharfe <l.s.r@web.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
status: be prepared for not-yet-started interactive rebase
Some developers might want to call `git status` in a working
directory where they just started an interactive rebase, but the
edit script is still opened in the editor.
Let's show a meaningful message in such cases.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
submodule update: run custom update script for initial populating as well
In 1b4735d9f3 (submodule: no [--merge|--rebase] when newly cloned,
2011-02-17), all actions were defaulted to checkout for populating
a submodule initially, because merging or rebasing makes no sense
in that situation.
Other commands however do make sense, such as the custom command
that was added later (6cb5728c43, submodule update: allow custom
command to update submodule working tree, 2013-07-03).
I am unsure about the "none" command, as I can see an initial
checkout there as a useful thing. On the other hand going strictly
by our own documentation, we should do nothing in case of "none"
as well, because the user asked for it.
Reported-by: Han-Wen Nienhuys <hanwen@google.com> Signed-off-by: Stefan Beller <sbeller@google.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
submodule absorbing: fix worktree/gitdir pointers recursively for non-moves
Consider having a submodule 'sub' and a nested submodule at 'sub/nested'.
When nested is already absorbed into sub, but sub is not absorbed into
its superproject, then we need to fixup the gitfile and core.worktree
setting for 'nested' when absorbing 'sub', but we do not need to move
its git dir around.
Previously 'nested's gitfile contained "gitdir: ../.git/modules/nested";
it has to be corrected to "gitdir: ../../.git/modules/sub1/modules/nested".
An alternative I considered to do this work lazily, i.e. when resolving
"../.git/modules/nested", we would notice the ".git" being a gitfile
linking to another path. That seemed to be robuster by design, but harder
to get the implementation right. Maybe we have to do that anyway once we
try to have submodules and worktrees working nicely together, but for now
just produce 'correct' (i.e. direct) pointers.
Signed-off-by: Stefan Beller <sbeller@google.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
cache.h: expose the dying procedure for reading gitlinks
In a later patch we want to react to only a subset of errors, defaulting
the rest to die as usual. Separate the block that takes care of dying
into its own function so we have easy access to it.
Signed-off-by: Stefan Beller <sbeller@google.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
This follows a93bedada (setup: add gentle version of read_gitfile,
2015-06-09), and assumes the same reasoning. resolve_git_dir is unsuited
for speculative calls, so we want to use the gentle version to find out
about potential errors.
Signed-off-by: Stefan Beller <sbeller@google.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The recent fixes to "fsck --connectivity-only" load all of
the objects with their correct types. This keeps the
connectivity-only code path close to the regular one, but it
also introduces some unnecessary inefficiency. While getting
the type of an object is cheap compared to actually opening
and parsing the object (as the non-connectivity-only case
would do), it's still not free.
For reachable non-blob objects, we end up having to parse
them later anyway (to see what they point to), making our
type lookup here redundant.
For unreachable objects, we might never hit them at all in
the reachability traversal, making the lookup completely
wasted. And in some cases, we might have quite a few
unreachable objects (e.g., when alternates are used for
shared object storage between repositories, it's normal for
there to be objects reachable from other repositories but
not the one running fsck).
The comment in mark_object_for_connectivity() claims two
benefits to getting the type up front:
1. We need to know the types during fsck_walk(). (And not
explicitly mentioned, but we also need them when
printing the types of broken or dangling commits).
We can address this by lazy-loading the types as
necessary. Most objects never need this lazy-load at
all, because they fall into one of these categories:
a. Reachable from our tips, and are coerced into the
correct type as we traverse (e.g., a parent link
will call lookup_commit(), which converts OBJ_NONE
to OBJ_COMMIT).
b. Unreachable, but not at the tip of a chunk of
unreachable history. We only mention the tips as
"dangling", so an unreachable commit which links
to hundreds of other objects needs only report the
type of the tip commit.
2. It serves as a cross-check that the coercion in (1a) is
correct (i.e., we'll complain about a parent link that
points to a blob). But we get most of this for free
already, because right after coercing, we'll parse any
non-blob objects. So we'd notice then if we expected a
commit and got a blob.
The one exception is when we expect a blob, in which
case we never actually read the object contents.
So this is a slight weakening, but given that the whole
point of --connectivity-only is to sacrifice some data
integrity checks for speed, this seems like an
acceptable tradeoff.
Here are before and after timings for an extreme case with
~5M reachable objects and another ~12M unreachable (it's the
torvalds/linux repository on GitHub, connected to shared
storage for all of the other kernel forks):
[before]
$ time git fsck --no-dangling --connectivity-only
real 3m4.323s
user 1m25.121s
sys 1m38.710s
[after]
$ time git fsck --no-dangling --connectivity-only
real 0m51.497s
user 0m49.575s
sys 0m1.776s
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
fsck: move typename() printing to its own function
When an object has a problem, we mention its type. But we do
so by feeding the result of typename() directly to
fprintf(). This is potentially dangerous because typename()
can return NULL for some type values (like OBJ_NONE).
It's doubtful that this can be triggered in practice with
the current code, so this is probably not fixing a bug. But
it future-proofs us against modifications that make things
like OBJ_NONE more likely (and gives future patches a
central point to handle them).
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The files in contrib/examples are meant to illustrate "you could
combine plumbing commands to implement something like these"; this
is an opposite and is an example of what not to do, e.g. accessing
the object store directly bypassing Git.
Remove it.
Suggested-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>