Change calls to die(..., strerror(errno)) to use the new die_errno().

In the process, also make slight style adjustments: at least state
_something_ about the function that failed (instead of just printing
the pathname), and put paths in single quotes.
Signed-off-by: NThomas Rast <trast@student.ethz.ch>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

One of the ways that locking might fail is that there is a
DF conflict between two refs (e.g., you want to lock
"foo/bar" but "foo" already exists). In this case, we return
an error, but there is no way for the caller to know the
specific problem.

This patch sets errno to ENOTDIR, which is the most sensible
code. It's what we would see if the refs were stored purely
in the filesystem (but these days we must check the
namespace manually due to packed refs).
Signed-off-by: NJeff King <peff@peff.net>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

In preparation to be used when the ref object is not available
Signed-off-by: NFelipe Contreras <felipe.contreras@gmail.com>
Acked-by: NJeff King <peff@peff.net>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

This is asking for trouble since '\' is a directory separator in
Windows and thus may produce unpredictable results.
Signed-off-by: NRobin Rosenberg <robin.rosenberg@dewire.com>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

This helps to notice when something's going wrong, especially on
systems which lock open files.

I used the following criteria when selecting the code for replacement:
- it was already printing a warning for the unlink failures
- it is in a function which already printing something or is
  called from such a function
- it is in a static function, returning void and the function is only
  called from a builtin main function (cmd_)
- it is in a function which handles emergency exit (signal handlers)
- it is in a function which is obvously cleaning up the lockfiles
Signed-off-by: NAlex Riesen <raa.lkml@gmail.com>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

Add the strict mode of abbreviation to shorten_unambiguous_ref(), i.e. the
resulting ref won't trigger the ambiguous ref warning.

All users of shorten_unambiguous_ref() still use the loose mode.
Signed-off-by: NBert Wesarg <bert.wesarg@googlemail.com>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

Often we want to shorten a full ref name to something "prettier"
to show a user. For example, "refs/heads/master" is often shown
simply as "master", or "refs/remotes/origin/master" is shown as
"origin/master".

Many places in the code use a very simple formula: skip common
prefixes like refs/heads, refs/remotes, etc. This is codified in
the prettify_ref function.

for-each-ref has a more correct (but more expensive) approach:
consider the ref lookup rules, and try shortening as much as
possible while remaining unambiguous.

This patch makes the latter strategy globally available as
shorten_unambiguous_ref.
Signed-off-by: NJeff King <peff@peff.net>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

The "for_each_{tag,branch,remote,replace,}_ref" functions are
redefined in terms of "for_each_ref_in" so that we can lose the
hardcoded length of prefix strings from the code.
Signed-off-by: NChristian Couder <chriscool@tuxfamily.org>

We already skip over loose refs under $GIT_DIR/refs if the name
ends with ".lock", so creating a branch named "foo.lock" will not
appear in the output of "git branch", "git for-each-ref", nor will
its commit be considered reachable by "git rev-list --all".

In the latter case this is especially evil, as it may cause
repository corruption when objects reachable only through such a
ref are deleted by "git prune".

It should be reasonably safe to deny use of ".lock" as a ref suffix.
In prior versions of Git such branches would be "phantom branches";
you can create it, but you can't see it in "git branch" output.
Signed-off-by: NShawn O. Pearce <spearce@spearce.org>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

This changes the rules for refnames to forbid:

 (1) a refname that contains "@{" in it.

     Some people and foreign SCM converter may have named their branches
     as frotz@24 and we still want to keep supporting it.

     However, "git branch frotz@{24}" is a disaster.  It cannot even
     checked out because "git checkout frotz@{24}" will interpret it as
     "detach the HEAD at twenty-fourth reflog entry of the frotz branch".

 (2) a refname that ends with a dot.

     We already reject a path component that begins with a dot, primarily
     to avoid ambiguous range interpretation.  If we allowed ".B" as a
     valid ref, it is unclear if "A...B" means "in dot-B but not in A" or
     "either in A or B but not in both".

     But for this to be complete, we need also to forbid "A." to avoid "in
     B but not in A-dot".  This was not a problem in the original range
     notation, but we should have added this restriction when three-dot
     notation was introduced.

     Unlike "no dot at the beginning of any path component" rule, this
     rule does not have to be "no dot at the end of any path component",
     because you cannot abbreviate the tail end away, similar to you can
     say "dot-B" to mean "refs/heads/dot-B".

For these reasons, it is not likely people created branches with these
names on purpose, but we have allowed such names to be used for quite some
time, and it is possible that people created such branches by mistake or
by accident.

To help people with branches with such unfortunate names to recover,
we still allow "branch -d 'bad.'" to delete such branches, and also allow
"branch -m bad. good" to rename them.
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

Signed-off-by: NMiklos Vajna <vmiklos@frugalware.org>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

The result should be consistent between fetch and push, so we ought to
use the same code in both cases, even though it's short.
Signed-off-by: NDaniel Barkalow <barkalow@iabervon.org>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

In order to keep the requirements strict, each * has to be a full path
component, and there may only be one * per side. This requirement is
enforced entirely by check_ref_format(); the matching implementation
will substitute the whatever matches the * in the lhs for the * in the
rhs.
Signed-off-by: NDaniel Barkalow <barkalow@iabervon.org>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

In order to do anything more capable with refspecs, the first step is
to keep the entire input. Additionally, validate patterns by checking
for the ref matching the rules for a pattern as given by
check_ref_format(). This requires a slight change to
check_ref_format() to make it enforce the requirement that the '*'
immediately follow a '/'.
Signed-off-by: NDaniel Barkalow <barkalow@iabervon.org>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

Since it doesn't actually touch its argument, this makes
sense.

However, we still want to return a non-const version (which
requires a cast) so that this:

  struct ref *a, *b;
  a = find_ref_by_name(b);

works. Unfortunately, you can also silently strip the const
from a variable:

  struct ref *a;
  const struct ref *b;
  a = find_ref_by_name(b);

This is a classic C const problem because there is no way to
say "return the type with the same constness that was passed
to us"; we provide the same semantics as standard library
functions like strchr.
Signed-off-by: NJeff King <peff@peff.net>
Signed-off-by: NJay Soffian <jaysoffian@gmail.com>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

If you prune from the remote "frotz" that deleted the ref your tracking
branch remotes/frotz/HEAD points at, the symbolic ref will become
dangling.  We used to detect this as an error condition and issued a
message every time refs are enumerated.

This stops the error message, but moves the warning to "remote prune".
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

This can be used to scan only the last few kilobytes of a reflog, as a
cheap optimization when the data you are looking for is likely to be
found near the end of it.  The caller is expected to fall back to the
full scan if that is not the case.
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

These were found using gcc 4.3.2-1ubuntu11 with the warning:

    warning: format not a string literal and no format arguments

Incorporated suggestions from Brandon Casey <casey@nrlssc.navy.mil>.
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

We force writing a ref if it does not exist. Originally, we only had to look
for the ref file to check if it existed. Now we have to look for a packed ref
as well. Luckily, resolve_ref already does all the work for us.
Signed-off-by: NClemens Buchacher <drizzd@aon.at>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

In this case we did nothing in the past, but we should delete the
reference in fact.

The problem was that when the symref is not packed but the referenced
ref is packed, then we assumed that the symref is packed as well, but
symrefs are never packed.
Signed-off-by: NMiklos Vajna <vmiklos@frugalware.org>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

Signed-off-by: NJunio C Hamano <gitster@pobox.com>

Signed-off-by: NJunio C Hamano <gitster@pobox.com>

There may be cases where one would really want to rename the symbolic
ref without changing its value, but "git branch -m" is not such a
use-case.
Signed-off-by: NMiklos Vajna <vmiklos@frugalware.org>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

We tried to check if a reflog of a ref is a symlink without first
checking if it exists, which is a bug.
Signed-off-by: NMiklos Vajna <vmiklos@frugalware.org>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

This had two problems with symrefs. First, it copied the actual sha1
instead of the "pointer", second it failed to remove the old ref after a
successful rename.

Given that till now delete_ref() always dereferenced symrefs, a new
parameters has been introduced to delete_ref() to allow deleting refs
without a dereference.
Signed-off-by: NMiklos Vajna <vmiklos@frugalware.org>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

This changes the "die_on_error" boolean parameter to a mere "flags", and
changes the existing callers of hold_lock_file_for_update/append()
functions to pass LOCK_DIE_ON_ERROR.
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

The existing in-code comment was misleading.  An access that is not
"reading" may often be "writing", but it does not have to be.
Signed-off-by: NChristian Couder <chriscool@tuxfamily.org>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

When the user specifies a ref by a reflog entry older than
one we have (e.g., "HEAD@{20 years ago"}), we issue a
warning and give them the "from" value of the oldest reflog
entry. That is, we say "we don't know what happened before
this entry, but before this we know we had some particular
SHA1".

However, the oldest reflog entry is often a creation event
such as clone or branch creation. In this case, the entry
claims that the ref went from "00000..." (the null sha1) to
the new value, and the reflog lookup returns the null sha1.

While this is technically correct (the entry tells us that
the ref didn't exist at the specified time) it is not
terribly useful to the end user. What they probably want
instead is "the oldest useful sha1 that this ref ever had".
This patch changes the behavior such that if the oldest ref
would return the null sha1, it instead returns the first
value the ref ever had.

We never discovered this problem in the test scripts because
we created "fake" reflogs that had only a specified segment
of history. This patch updates the tests with a creation
event at the beginning of history.
Signed-off-by: NJeff King <peff@peff.net>
Acked-by: NShawn O. Pearce <spearce@spearce.org>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

If git attempts to delete a ref, but the unlink of the ref
file fails, we print a message to stderr. This is usually a
good thing, but if the error is ENOENT, then it indicates
that the ref has _already_ been deleted. And since that's
our goal, it doesn't make sense to complain to the user.

This harmonizes the error reporting behavior for the
unpacked and packed cases; the packed case already printed
nothing on ENOENT, but the unpacked printed unconditionally.

Additionally, send-pack would, when deleting the tracking
ref corresponding to a remote delete, print "Failed to
delete" on any failure. This can be a misleading
message, since we actually _did_ delete at the remote side,
but we failed to delete locally. Rather than make the
message more precise, let's just eliminate it entirely; the
delete_ref routine already takes care of printing out a much
more specific message about what went wrong.
Signed-off-by: NJeff King <peff@peff.net>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

These refs can be anything, but they are most likely useful as
pointing to objects that you know are in the object database but don't
have any regular refs for. For example, when cloning with --reference,
the refs in this repository should be listed as objects that we have,
even though we don't have refs in our newly-created repository for
them yet.
Signed-off-by: NDaniel Barkalow <barkalow@iabervon.org>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

xread() and xwrite() return ssize_t values as their native POSIX
counterparts read(2) and write(2).

To be consistent, read_in_full() and write_in_full() should also return
ssize_t values.
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

When .git in a submodule is a file, resolve_gitlink_ref() needs to pick up
the real GIT_DIR of the submodule from that file.
Signed-off-by: NLars Hjemli <hjemli@gmail.com>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

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: NShawn O. Pearce <spearce@spearce.org>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

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: NShawn O. Pearce <spearce@spearce.org>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

This is in preparation to the reflog-expire changes which will
allow updating the ref after expiring the reflog.
Signed-off-by: NBrandon Casey <casey@nrlssc.navy.mil>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

Instead of calling close_lock_file() and commit_lock_file() directly,
which take a struct lock_file argument, add two new functions:
close_ref() and commit_ref(), which handle calling the previous
lock_file functions and modifying the ref_lock structure.
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

Remove remaining double close(2)'s.  i.e. close() before
commit_locked_index() or commit_lock_file().
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

This makes write_ref_sha1() more careful: it actually checks the SHA1 of
the ref it is updating, and refuses to update a ref with an object that it
cannot find.

Perhaps more importantly, it also refuses to update a branch head with a
non-commit object. I don't quite know *how* the stable series maintainers
were able to corrupt their repository to have a HEAD that pointed to a tag
rather than a commit object, but they did. Which results in a totally
broken repository that cannot be cloned or committed on.

So make it harder for people to shoot themselves in the foot like that.

The test t1400-update-ref.sh is fixed at the same time, as it
assumed that the commands involved in the particular test would
not care about corrupted repositories whose refs point at
nonexistant bogus objects.  That assumption does not hold true
anymore.
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

This updates send-pack and fast-import to use symbolic constants
for checking the return values from check_ref_format(), and also
futureproof the logic in lock_any_ref_for_update() to explicitly
name the case that is usually considered an error but is Ok for
this particular use.
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

Recent check_ref_format() returns -3 as well as -1 (general
error) and -2 (less than two levels).  The caller was explicitly
checking for -1, to allow "HEAD" but still needed to disallow
bogus refs.

This introduces symbolic constants for the return values from
check_ref_format() to make them read better and more
meaningful.  Normal ref creation codepath can still treat
non-zero return values as errors.
Signed-off-by: NJunio C Hamano <gitster@pobox.com>