Now that struct commit.util is not used until after we've checked that
the argument doesn't exactly match a tag, we can wait until then to
look up the commits for each tag.
This avoids a lot of I/O on --exact-match queries in repositories with
many tags. For example, 'git describe --exact-match HEAD' becomes
about 12 times faster on a cold cache (3.2s instead of 39s) in a
linux-2.6 repository with 2000 packed tags. That is a huge win for the
interactivity of the __git_ps1 shell prompt helper when on a detached
HEAD.
Signed-off-by: Anders Kaseorg <andersk@ksplice.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
describe is currently forced to look up the commit at each tag in
order to store the struct commit_name pointers in struct commit.util.
For --exact-match queries, those lookups are wasteful. In preparation
for removing them, put the commit_names into a hash table, indexed by
commit SHA1, that can be used to quickly check for exact matches.
Signed-off-by: Anders Kaseorg <andersk@ksplice.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Now add_to_known_names overwrites commit_names in place when multiple
tags point to the same commit. This will make it easier to store
commit_names in a hash table.
Signed-off-by: Anders Kaseorg <andersk@ksplice.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Don't waste time checking for dangling refs; they wouldn't affect the
output of 'git describe' anyway. Although this does not gain much
performance by itself, it does in conjunction with the next commits.
Signed-off-by: Anders Kaseorg <andersk@ksplice.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The msvc linker complains about external symbols referenced by
the new poll() emulation code. In particular, it complains about
the DispatchMessage(), PeekMessage(), TranslateMessage() and
MsgWaitForMultipleObjects() functions.
In order to satisfy the external references, we link against the
user32.lib library.
Signed-off-by: Ramsay Jones <ramsay@ramsay1.demon.co.uk>
Acked-by: Erik Faye-Lund <kusmabite@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The msvc winsock2.h header file conditionally defines or declares
poll() related symbols which cause many macro redefinition errors,
a struct type redefinition error and syntax errors. These symbols
are defined in support of the WSAPoll() API, new in Windows Vista,
when the symbol _WIN32_WINNT is defined and _WIN32_WINNT >= 0x0600.
In order to avoid the compilation errors, we set _WIN32_WINNT to
0x0502 (which would target Windows Server 2003) prior to including
the winsock2.h header file.
Signed-off-by: Ramsay Jones <ramsay@ramsay1.demon.co.uk>
Acked-by: Erik Faye-Lund <kusmabite@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The function was returning 0 for failure and 1 for success which was
breaking the logic in the main loop. It now also returns in all
cases, rather than exiting.
Signed-off-by: Alan Raison <alan@theraisons.me.uk>
Acked-by: Kevin P. Fleming <kpfleming@digium.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Running test t9010 without svn currently errors out for no good reason.
The test uses "svnadmin" without checking if svn is available. This was a
regression introduced by b0ad24b (t9010 (svn-fe): Eliminate dependency on
svn perl bindings, 2010-10-10) when it stopped including ./lib-git-svn.sh
that had the safety.
This should fix it.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
With 401857c (imap-send: link against libcrypto for HMAC and others,
2010-11-24) we started linking imap-send unconditionally with -lcrypto by
mistake; disable this when we are building under NO_OPENSSL.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
* gc/http-with-non-ascii-username-url:
Fix username and password extraction from HTTP URLs
t5550: test HTTP authentication and userinfo decoding
Conflicts:
t/lib-httpd/apache.conf
* gb/gitweb-remote-heads:
git instaweb: enable remote_heads
gitweb: group remote heads by remote
gitweb: provide a routine to display (sub)sections
gitweb: refactor repository URL printing
gitweb: remotes view for a single remote
gitweb: allow action specialization in page header
gitweb: nagivation menu for tags, heads and remotes
gitweb: separate heads and remotes lists
gitweb: git_get_heads_list accepts an optional list of refs
gitweb: introduce remote_heads feature
gitweb: use fullname as hash_base in heads link
* mg/maint-tag-rfc1991:
tag: recognize rfc1991 signatures
tag: factor out sig detection for tag display
tag: factor out sig detection for body edits
verify-tag: factor out signature detection
t/t7004-tag: test handling of rfc1991 signatures
* jh/notes-merge: (23 commits)
Provide 'git merge --abort' as a synonym to 'git reset --merge'
cmd_merge(): Parse options before checking MERGE_HEAD
Provide 'git notes get-ref' to easily retrieve current notes ref
git notes merge: Add testcases for merging notes trees at different fanouts
git notes merge: Add another auto-resolving strategy: "cat_sort_uniq"
git notes merge: --commit should fail if underlying notes ref has moved
git notes merge: List conflicting notes in notes merge commit message
git notes merge: Manual conflict resolution, part 2/2
git notes merge: Manual conflict resolution, part 1/2
Documentation: Preliminary docs on 'git notes merge'
git notes merge: Add automatic conflict resolvers (ours, theirs, union)
git notes merge: Handle real, non-conflicting notes merges
builtin/notes.c: Refactor creation of notes commits.
git notes merge: Initial implementation handling trivial merges only
builtin/notes.c: Split notes ref DWIMmery into a separate function
notes.c: Use two newlines (instead of one) when concatenating notes
(trivial) t3303: Indent with tabs instead of spaces for consistency
notes.h/c: Propagate combine_notes_fn return value to add_note() and beyond
notes.h/c: Allow combine_notes functions to remove notes
notes.c: Reorder functions in preparation for next commit
...
Conflicts:
builtin.h
Pointers to directory entries do not remain valid after a call to
dent_insert.
Noticed in the course of importing a small Subversion repository
(~1000 revs); after setting up a dirent for a certain path as a
placeholder, by luck dent_insert would trigger a realloc that
shifted around addresses, resulting in an import with that file
replaced by a directory.
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Suppose I try the following:
struct int_node *node = node_pointer(node_alloc(1));
node->n = 5;
treap_insert(&root, node);
printf("%d\n", node->n);
Usually the result will be 5. But since treap_insert draws memory
from the node pool, if the caller is unlucky then (1) the node pool
will be full and (2) realloc will be forced to move the node pool to
find room, so the node address becomes invalid and the result of
dereferencing it is undefined.
So we ought to use offsets in preference to pointers for references
that would remain valid after a call to treap_insert. Tweak the
signature to hint at a certain special case: since the inserted node
can change address (though not offset), as a convenience teach
treap_insert to return its new address.
So the motivational example could be fixed by adding "node =".
struct int_node *node = node_pointer(node_alloc(1));
node->n = 5;
node = treap_insert(&root, node);
printf("%d\n", node->n);
Based on a true story.
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Currently :path and ref:path can be used to refer to a specific object
in index or ref respectively. "path" component is absolute path. This
patch allows "path" to be written as "./path" or "../path", which is
relative to user's original cwd.
This does not work in commands for which startup_info is NULL
(i.e. non-builtin ones, it seems none of them needs this anyway).
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
prefix_path() allocates new buffer. There's no reason for it to keep
the buffer for itself and waste memory.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
--refresh and --really-refresh accept flags (like -q) and modify
an error indicator. It might make sense to make the error
indicator global, but just pass the flags and a pointer to the error
indicator in a struct instead.
--cacheinfo wants 3 arguments. Use the OPTION_LOWLEVEL_CALLBACK
extension to grab them and PARSE_OPT_NOARG to disallow the "sticked"
--cacheinfo=foo form. (The resulting message
$ git update-index --cacheinfo=foo
error: option `cacheinfo' takes no value
is unfortunately incorrect.)
--assume-unchanged and --no-assume-unchanged probably should use the
OPT_UYN feature; but use a callback for now so the existing MARK_FLAG
and UNMARK_FLAG values can be used.
--stdin and --index-info are still constrained to be the last argument
(implemented using the OPTION_LOWLEVEL_CALLBACK extension).
--unresolve and --again consume all arguments that come after them
(also using OPTION_LOWLEVEL_CALLBACK).
The order of options matters. Each path on the command line is
affected only by the options that come before it. A custom
argument-parsing loop with parse_options_step() brings that about.
In exchange for all the fuss, we get the usual perks: support for
un-sticked options, better usage error messages, more useful -h
output, and argument parsing code that should be easier to tweak
in the future.
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Save the path from the original cwd to the cwd at the end of the
setup procedure in the startup_info struct introduced in e37c1329
(2010-08-05). The value cannot vary from thread to thread anyway,
since the cwd is global.
So now in your builtin command, instead of passing prefix around,
when you want to convert a user-supplied path to a cwd-relative
path, you can use startup_info->prefix directly.
Caveat: As with the return value from setup_git_directory_gently(),
startup_info->prefix would be NULL when the original cwd is not a
subdir of the toplevel.
Longer term, this would allow the prefix to be reused when several
noncooperating functions require access to the same repository (for
example, when accessing configuration before running a builtin).
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Introduce a PARSE_OPT_NON_OPTION state, so parse_option_step()
callers can easily distinguish between non-options and other
reasons for option parsing termination (like "--").
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
parse-options provides a variety of option behaviors, including
OPTION_CALLBACK, which should take care of just about any sane
behavior. All supported behaviors obey the following constraint:
A --foo option can only accept (and base its behavior on)
one argument, which would be the following command-line
argument in the "unsticked" form.
Alas, some existing git commands have options that do not obey that
constraint. For example, update-index --cacheinfo takes three
arguments, and update-index --resolve takes all later parameters as
arguments.
Introduces an OPTION_LOWLEVEL_CALLBACK backdoor to parse-options so
such option types can be supported without tempting inventors of other
commands through mention in the public API. Commands can set the
callback field to a function accepting three arguments: the option
parsing context, the option itself, and a flag indicating whether the
the option was negated. When the option is encountered, that function
is called to take over from get_value(). The return value should be
zero for success, -1 for usage errors.
Thanks to Stephen Boyd for API guidance.
Improved-by: Stephen Boyd <bebarino@gmail.com>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The PARSE_OPT_LITERAL_ARGHELP flag allows a program to override the
standard "<argument> for mandatory, [argument] for optional" markup in
its help message. Extend it to override the usual "no text for
disallowed", too (for the PARSE_OPT_NOARG | PARSE_OPT_LITERAL_ARGHELP
case, which was previously meaningless), to be more intuitive.
The motivation is to allow update-index to correctly advertise
--cacheinfo <mode> <object> <path>
add the specified entry to the index
while abusing PARSE_OPT_NOARG to disallow the "sticked form"
--cacheinfo=<mode> <object> <path>
Noticed-by: Stephen Boyd <bebarino@gmail.com>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Simplify the "takes no value" error path by relying on PARSE_OPT_NOARG
being set correctly. That is:
- if the PARSE_OPT_NOARG flag is set, reject --opt=value
regardless of the option type;
- if the PARSE_OPT_NOARG flag is unset, accept --opt=value
regardless of the option type.
This way, the accepted usage more closely matches the usage advertised
with --help-all.
No functional change intended, since the NOARG flag is only used
with "boolean-only" option types in existing parse_options callers.
Signed-off-by: Stephen Boyd <bebarino@gmail.com>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Some option types cannot use an argument --- boolean options that
would set a bit or flag or increment a counter, for example. If
configured in the flag word to accept an argument anyway, the result
is an argument that is advertised in "program -h" output only to be
rejected by parse-options::get_value.
Luckily all current users of these option types use PARSE_OPT_NOARG
and do not use PARSE_OPT_OPTARG. Add a check to ensure that that
remains true. The check is run once for each invocation of
parse_option_start().
Improved-by: Stephen Boyd <bebarino@gmail.com>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
A dashless switch (like '(' passed to 'git grep') cannot be negated,
cannot be attached to an argument, and cannot have a long form.
Currently parse-options runs the related sanity checks when the
dashless option is used; better to always check them at the start of
option parsing, so mistakes can be caught more quickly.
The error message at the new call site is less specific about the
nature of the error, for simplicity. On the other hand, it prints
which switch was problematic. Before:
fatal: BUG: dashless options can't be long
After:
error: BUG: switch '(' uses feature not supported for dashless options
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The PARSE_OPT_LASTARG_DEFAULT flag is meant for options like
--contains that (1) traditionally had a mandatory argument and
(2) have some better behavior to use when appearing in the final
position. It makes no sense to combine this with OPTARG, so ever
since v1.6.4-rc0~71 (parse-options: add parse_options_check to
validate option specs, 2009-07-09) this mistake is flagged with
error: `--option` uses incompatible flags LASTARG_DEFAULT and OPTARG
and an exit status representing an error in commandline usage.
Unfortunately that which might confuse scripters calling such an
erroneous program into thinking the _script_ contains an error.
Clarify that it is an internal error by dying with a message beginning
"error: BUG: ..." and status 128.
While at it, clean up parse_options_check to prepare for more checks.
Long term, it would be nicer to make such checks happen at compile
time.
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Literal " produces typographically incorrect quotations, but "works" in
most circumstances. In the subheadings of git-rm.txt, it "works" for the
html backend but not for the docbook conversion to nroff: double "" and
spurious double spaces appear in the output.
Replace "incorrect" quotations by ``correct'' ones, and fix other
"quotations" which are really `code fragments`.
This should make git-rm.txt "-clean.
Reported-by: Jeff King <peff@peff.net>
Signed-off-by: Michael J Gruber <git@drmicha.warpmail.net>
Helped-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The vcs-svn tests (test-treap.o et al) depend on the vcs-svn
headers for declarations and inline functions. Declare the
dependency. While at it, declare a dependency of the vcs-svn
objects (vcs-svn/string_pool.o et al) on $(LIB_H) to reflect use
within the vcs-svn library of git-compat-util.h and cache.h.
Without this change, tweaks to inline functions in those headers
do not provoke rebuilds of the corresponding tests[*], making
such changes unnecessarily difficult to test.
Before:
$ touch vcs-svn/*.h && make test-treap
$
After:
$ touch vcs-svn/*.h && make test-treap
CC test-treap.o
LINK test-treap
$
[*] unless COMPUTE_HEADER_DEPENDENCIES is enabled
Detected with "make CHECK_HEADER_DEPENDENCIES=1".
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
get_remote_merge_branch with zero or one arguments returns the
upstream branch. But a detached HEAD does no have an upstream branch,
as it is not tracking anything. Handle this case testing the exit code
of "git symbolic-ref -q HEAD".
Reported-by: Sverre Rabbelier <srabbelier@gmail.com>
Signed-off-by: Santi Béjar <santi@agolina.net>
Tested-by: Sverre Rabbelier <srabbelier@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
parse_options_check() is being called for each invocation of
parse_options_step which can be quite a bit for some commands. The
commit introducing this function cb9d398 (parse-options: add
parse_options_check to validate option specs., 2009-06-09) had the
correct motivation and explicitly states that parse_options_check()
should be called from parse_options_start(). However, the implementation
differs from the motivation. Fix it.
Signed-off-by: Stephen Boyd <bebarino@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Since 55892d23 "git clone" itself checks that the destination path is not
a file but an empty directory if it exists, so there is no need anymore
for module_clone() to check that too.
Two tests have been added to test the behavior of "git submodule add" when
path is a file or a directory (A subshell had to be added to the former
last test to stay in the right directory).
Signed-off-by: Jens Lehmann <Jens.Lehmann@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
