Ask branch_get() for the new branch explicitly instead of
letting it return a potentially stale information.
Tighten the logic to find the tracking branch to deal better
with misconfigured repositories (i.e. branch.*.merge can exist
but it may not have a refspec that fetches to .it)
Also fixes grammar in a message, as pointed out by Jeff King.
The function is about reporting and not automatically
fast-forwarding to the upstream, so stop calling it
"adjust-to".
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Acked-by: Daniel Barkalow <barkalow@iabervon.org>
This path doesn't actually care where in the tree you started out,
since it must change the whole thing anyway. With the gratuitous bug
removed, the argument is unused.
Signed-off-by: Daniel Barkalow <barkalow@iabervon.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When switching branches from a subdirectory, checkout rewritten
in C extracted the toplevel of the tree in there.
This should fix it.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When checking out a branch that is behind or forked from a
branch you are building on top of, we used to show full
left-right log but if you already _know_ you have long history
since you forked, it is a bit too much.
This tones down the message quite a bit, by only showing the
number of commits each side has since they diverged. Also the
message is not shown at all under --quiet.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This also changes it such that:
$ git checkout
will give the same information without changing branches. This is good
for finding out if the fetch you did recently had anything to say
about the branch you've been on, whose name you don't remember at the
moment.
Signed-off-by: Daniel Barkalow <barkalow@iabervon.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Not terminating the options[] array with OPT_END can cause
usage ("git checkout -h") output to segfault.
Signed-off-by: Jay Soffian <jaysoffian@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When you are switching to a branch that is marked to merge from
somewhere else, e.g. when you have:
[branch "next"]
remote = upstream
merge = refs/heads/next
[remote "upstream"]
url = ...
fetch = refs/heads/*:refs/remotes/linus/*
and you say "git checkout next", the branch you checked out
may be behind, and you may want to update from the upstream
before continuing to work.
This patch makes the command to check the upstream (in this
example, "refs/remotes/linus/next") and our branch "next", and:
(1) if they match, nothing happens;
(2) if you are ahead (i.e. the upstream is a strict ancestor
of you), one line message tells you so;
(3) otherwise, you are either behind or you and the upstream
have forked. One line message will tell you which and
then you will see a "log --pretty=oneline --left-right".
We could enhance this with an option that tells the command to
check if there is no local change, and automatically fast
forward when you are truly behind. But I ripped out that change
because I was unsure what the right way should be to allow users
to control it (issues include that checkout should not become
automatically interactive).
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The only differences in behavior should be:
- git checkout -m with non-trivial merging won't print out
merge-recursive messages (see the change in t7201-co.sh)
- git checkout -- paths... will give a sensible error message if
HEAD is invalid as a commit.
- some intermediate states which were written to disk in the shell
version (in particular, index states) are only kept in memory in
this version, and therefore these can no longer be revealed by
later write operations becoming impossible.
- when we change branches, we discard MERGE_MSG, SQUASH_MSG, and
rr-cache/MERGE_RR, like reset always has.
I'm not 100% sure I got the merge recursive setup exactly right; the
base for a non-trivial merge in the shell code doesn't seem
theoretically justified to me, but I tried to match it anyway, and the
tests all pass this way.
Other than these items, the results should be identical to the shell
version, so far as I can tell.
[jc: squashed lock-file fix from Dscho in]
Signed-off-by: Daniel Barkalow <barkalow@iabervon.org>
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
If the test failed, it was giving really unclear ed script
output. Instead, give a diff that sort of suggests the problem. Also
replaces the use of "git diff" for this purpose with "diff -u".
Signed-off-by: Daniel Barkalow <barkalow@iabervon.org>
You can also create branches, in exactly the same way, with checkout -b.
This introduces branch.{c,h} library files for doing porcelain-level
operations on branches (such as creating them with their appropriate
default configuration).
Signed-off-by: Daniel Barkalow <barkalow@iabervon.org>
This makes write_tree_from_memory(), which writes the active cache as
a tree and returns the struct tree for it, available to other code. It
also makes available merge_trees(), which does the internal merge of
two trees with a known base, and merge_recursive(), which does the
recursive internal merge of two commits with a list of common
ancestors.
The first two of these will be used by checkout -m, and the third is
presumably useful in general, although the implementation of checkout
-m which entirely matches the behavior of the shell version does not
use it (since it ignores the difference of ancestry between the old
branch and the new branch).
Signed-off-by: Daniel Barkalow <barkalow@iabervon.org>
This option allows the caller to reset everything that isn't unmerged,
leaving the unmerged things to be resolved. If, after a merge of
"working" and "HEAD", this is used with "HEAD" (reset, !update), the
result will be that all of the changes from "local" are in the working
tree but not added to the index (either with the index clean but
unchanged, or with the index unmerged, depending on whether there are
conflicts).
This will be used in checkout -m.
Signed-off-by: Daniel Barkalow <barkalow@iabervon.org>
Way back in read-tree.c, we used a mode 0 cache entry to indicate that
an entry had been deleted, so that the update code would remove the
working tree file, and we would just skip it when writing out the
index file afterward.
These days, unpack_trees is a library function, and it is still
leaving these entries in the active cache. Furthermore, unpack_trees
doesn't correctly ignore those entries, and who knows what other code
wouldn't expect them to be there, but just isn't yet called after a
call to unpack_trees. To avoid having other code trip over these
entries, have check_updates() remove them after it removes the working
tree files.
While we're at it, simplify the loop in check_updates(), and avoid
passing global variables as parameters to check_updates(): there is
only one call site anyway.
Signed-off-by: Daniel Barkalow <barkalow@iabervon.org>
(This applies only to errors where a plausible operation is impossible due
to the particular data, not to errors resulting from misuse of the merge
functions.)
This will allow builtin-checkout to suppress merge errors if it's
going to try more merging methods.
Additionally, if unpack_trees() returns with an error, but without
printing anything, it will roll back any changes to the index (by
rereading the index, currently). This obviously could be done by the
caller, but chances are that the caller would forget and debugging
this is difficult. Also, future implementations may give unpack_trees() a
more efficient way of undoing its changes than the caller could.
Signed-off-by: Daniel Barkalow <barkalow@iabervon.org>
Return an error from unpack_trees() instead of calling die(), and exit
with an error in read-tree, builtin-commit, and diff-lib. merge-recursive
already expected an error return from unpack_trees, so it doesn't need to
be changed. The merge function can return negative to abort.
This will be used in builtin-checkout -m.
Signed-off-by: Daniel Barkalow <barkalow@iabervon.org>
This creates a hash index of every single file added to the index.
Right now that hash index isn't actually used for much: I implemented a
"cache_name_exists()" function that uses it to efficiently look up a
filename in the index without having to do the O(logn) binary search,
but quite frankly, that's not why this patch is interesting.
No, the whole and only reason to create the hash of the filenames in the
index is that by modifying the hash function, you can fairly easily do
things like making it always hash equivalent names into the same bucket.
That, in turn, means that suddenly questions like "does this name exist
in the index under an _equivalent_ name?" becomes much much cheaper.
Guiding principles behind this patch:
- it shouldn't be too costly. In fact, my primary goal here was to
actually speed up "git commit" with a fully populated kernel tree, by
being faster at checking whether a file already existed in the index. I
did succeed, but only barely:
Best before:
[torvalds@woody linux]$ time git commit > /dev/null
real 0m0.255s
user 0m0.168s
sys 0m0.088s
Best after:
[torvalds@woody linux]$ time ~/git/git commit > /dev/null
real 0m0.233s
user 0m0.144s
sys 0m0.088s
so some things are actually faster (~8%).
Caveat: that's really the best case. Other things are invariably going
to be slightly slower, since we populate that index cache, and quite
frankly, few things really use it to look things up.
That said, the cost is really quite small. The worst case is probably
doing a "git ls-files", which will do very little except puopulate the
index, and never actually looks anything up in it, just lists it.
Before:
[torvalds@woody linux]$ time git ls-files > /dev/null
real 0m0.016s
user 0m0.016s
sys 0m0.000s
After:
[torvalds@woody linux]$ time ~/git/git ls-files > /dev/null
real 0m0.021s
user 0m0.012s
sys 0m0.008s
and while the thing has really gotten relatively much slower, we're
still talking about something almost unmeasurable (eg 5ms). And that
really should be pretty much the worst case.
So we lose 5ms on one "benchmark", but win 22ms on another. Pick your
poison - this patch has the advantage that it will _likely_ speed up
the cases that are complex and expensive more than it slows down the
cases that are already so fast that nobody cares. But if you look at
relative speedups/slowdowns, it doesn't look so good.
- It should be simple and clean
The code may be a bit subtle (the reasons I do hash removal the way I
do etc), but it re-uses the existing hash.c files, so it really is
fairly small and straightforward apart from a few odd details.
Now, this patch on its own doesn't really do much, but I think it's worth
looking at, if only because if done correctly, the name hashing really can
make an improvement to the whole issue of "do we have a filename that
looks like this in the index already". And at least it gets real testing
by being used even by default (ie there is a real use-case for it even
without any insane filesystems).
NOTE NOTE NOTE! The current hash is a joke. I'm ashamed of it, I'm just
not ashamed of it enough to really care. I took all the numbers out of my
nether regions - I'm sure it's good enough that it works in practice, but
the whole point was that you can make a really much fancier hash that
hashes characters not directly, but by their upper-case value or something
like that, and thus you get a case-insensitive hash, while still keeping
the name and the index itself totally case sensitive.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This moves a common boolean expression into a helper function,
and makes the comparison between filesystem timestamp and index
timestamp done in the function in line with the other places.
st.st_mtime should be casted to (unsigned int) when compared to
an index timestamp ce_mtime.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
It is a D/F conflict if you want to add "foo/bar" to the index
when "foo" already exists. Also it is a conflict if you want to
add a file "foo" when "foo/bar" exists.
An exception is when the existing entry is there only to mark "I
used to be here but I am being removed". This is needed for
operations such as "git read-tree -m -u" that update the index
and then reflect the result to the work tree --- we need to
remember what to remove somewhere, and we use the index for
that. In such a case, an existing file "foo" is being removed
and we can create "foo/" directory and hang "bar" underneath it
without any conflict.
We used to use (ce->ce_mode == 0) to mark an entry that is being
removed, but (CE_REMOVE & ce->ce_flags) is used for that purpose
these days. An earlier commit forgot to convert the logic in
the code that checks D/F conflict condition.
The old code knew that "to be removed" entries cannot be at
higher stage and actively checked that condition, but it was an
unnecessary check. This patch removes the extra check as well.
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
As in run_diff_index(), we call unpack_trees() with the oneway_diff()
function in do_diff_cache() now. This makes the function diff_cache()
obsolete.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
A plain "git commit" would still run lstat() a lot more than necessary,
because wt_status_print() would cause the index to be repeatedly flushed
and re-read by wt_read_cache(), and that would cause the CE_UPTODATE bit
to be lost, resulting in the files in the index being lstat'ed three
times each.
The reason why wt-status.c ended up invalidating and re-reading the
cache multiple times was that it uses "run_diff_index()", which in turn
uses "read_tree()" to populate the index with *both* the old index and
the tree we want to compare against.
So this patch re-writes run_diff_index() to not use read_tree(), but
instead use "unpack_trees()" to diff the index to a tree. That, in
turn, means that we don't need to modify the index itself, which then
means that we don't need to invalidate it and re-read it!
This, together with the lstat() optimizations, means that "git commit"
on the kernel tree really only needs to lstat() the index entries once.
That noticeably cuts down on the cached timings.
Best time before:
[torvalds@woody linux]$ time git commit > /dev/null
real 0m0.399s
user 0m0.232s
sys 0m0.164s
Best time after:
[torvalds@woody linux]$ time git commit > /dev/null
real 0m0.254s
user 0m0.140s
sys 0m0.112s
so it's a noticeable improvement in addition to being a nice conceptual
cleanup (it's really not that pretty that "run_diff_index()" dirties the
index!)
Doing an "strace -c" on it also shows that as it cuts the number of
lstat() calls by two thirds, it goes from being lstat()-limited to being
limited by getdents() (which is the readdir system call):
Before:
% time seconds usecs/call calls errors syscall
------ ----------- ----------- --------- --------- ----------------
60.69 0.000704 0 69230 31 lstat
23.62 0.000274 0 5522 getdents
8.36 0.000097 0 5508 2638 open
2.59 0.000030 0 2869 close
2.50 0.000029 0 274 write
1.47 0.000017 0 2844 fstat
After:
% time seconds usecs/call calls errors syscall
------ ----------- ----------- --------- --------- ----------------
45.17 0.000276 0 5522 getdents
26.51 0.000162 0 23112 31 lstat
19.80 0.000121 0 5503 2638 open
4.91 0.000030 0 2864 close
1.48 0.000020 0 274 write
1.34 0.000018 0 2844 fstat
...
It passes the test-suite for me, but this is another of one of those
really core functions, and certainly pretty subtle, so..
NOTE! The Linux lstat() system call is really quite cheap when everything
is cached, so the fact that this is quite noticeable on Linux is likely to
mean that it is *much* more noticeable on other operating systems. I bet
you'll see a much bigger performance improvement from this on Windows in
particular.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Aside from the lstat(2) done for work tree files, there are
quite many lstat(2) calls in refname dwimming codepath. This
patch is not about reducing them.
* It adds a new ce_flag, CE_UPTODATE, that is meant to mark the
cache entries that record a regular file blob that is up to
date in the work tree. If somebody later walks the index and
wants to see if the work tree has changes, they do not have
to be checked with lstat(2) again.
* fill_stat_cache_info() marks the cache entry it just added
with CE_UPTODATE. This has the effect of marking the paths
we write out of the index and lstat(2) immediately as "no
need to lstat -- we know it is up-to-date", from quite a lot
fo callers:
- git-apply --index
- git-update-index
- git-checkout-index
- git-add (uses add_file_to_index())
- git-commit (ditto)
- git-mv (ditto)
* refresh_cache_ent() also marks the cache entry that are clean
with CE_UPTODATE.
* write_index is changed not to write CE_UPTODATE out to the
index file, because CE_UPTODATE is meant to be transient only
in core. For the same reason, CE_UPDATE is not written to
prevent an accident from happening.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We currently use lower 12-bit (masked with CE_NAMEMASK) in the
ce_flags field to store the length of the name in cache_entry,
without checking the length parameter given to
create_ce_flags(). This can make us store incorrect length.
Currently we are mostly protected by the fact that many
codepaths first copy the path in a variable of size PATH_MAX,
which typically is 4096 that happens to match the limit, but
that feels like a bug waiting to happen. Besides, that would
not allow us to shorten the width of CE_NAMEMASK to use the bits
for new flags.
This redefines the meaning of the name length stored in the
cache_entry. A name that does not fit is represented by storing
CE_NAMEMASK in the field, and the actual length needs to be
computed by actually counting the bytes in the name[] field.
This way, only the unusually long paths need to suffer.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This converts the index explicitly on read and write to its on-disk
format, allowing the in-core format to contain more flags, and be
simpler.
In particular, the in-core format is now host-endian (as opposed to the
on-disk one that is network endian in order to be able to be shared
across machines) and as a result we can dispense with all the
htonl/ntohl on accesses to the cache_entry fields.
This will make it easier to make use of various temporary flags that do
not exist in the on-disk format.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
* git://repo.or.cz/git-gui:
git-gui: Makefile - Handle $DESTDIR on Cygwin
git-gui: add french glossary: glossary/fr.po
git-gui: Refresh file status description after hunk application
git-gui: Allow 'Create New Repository' on existing directories
git-gui: Initial french translation
git-gui: Improve German translation.
git-gui: Updated Swedish translation after mailing list review.
git-gui: Fix broken revert confirmation.
git-gui: Update German translation
git-gui: Update glossary: add term "hunk"
The URL to a repository http-push and http-fetch takes should
have a trailing slash. Instead of failing the request, add it
ourselves before attempting such a request.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The first thing http-push does is a PROPFIND to see if the other
end supports locking. The failure message we give is always
reported as "no DAV locking support at the remote repository",
regardless of the reason why we ended up not finding the locking
support on the other end.
This moves the code to report "no DAV locking support" down the
codepath so that the message is issued only when we successfully
get a response to PROPFIND and the other end say it does not
support locking. Other failures, such as connectivity glitches
and credential mismatches, have their own error message issued
and we will not issue "no DAV locking" error (we do not even
know if the remote end supports it).
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Failing instead of silently not updating remote refs makes the things
clearer for the user when trying to push on a repository while another
person do (or while a dandling locks are waiting for a 10 minutes
timeout).
When silently not updating remote refs, the user does not even know
that git has pushed the objects but leaved the refs as they were
before (e.g. a new bunch of commits on branch "master" is uploaded,
however the branch by itsel still points on the previous head commit).
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Releasing webdav lock even if push fails because of bad (or no)
reference on command line.
To reproduce the issue that this patch fixes, prepare a test repository
availlable over http+webdav, say at http://myhost/myrepo.git/
Then:
$ git clone http://myhost/myrepo.git/
$ cd myrepo
$ git push http
Fetching remote heads...
refs/
refs/heads/
refs/tags/
No refs in common and none specified; doing nothing.
$ git push http
Fetching remote heads...
refs/
refs/heads/
refs/tags/
No refs in common and none specified; doing nothing.
$
Finally, you look at the web server logs, and will find one LOCK query
and no UNLOCK query, of course the second one will be in 423 return
code instead of 200:
1.2.3.4 - gb [19/Jan/2008:14:24:56 +0100] "LOCK /myrepo.git/info/refs HTTP/1.1" 200 465
(...)
1.2.3.4 - gb [19/Jan/2008:14:25:10 +0100] "LOCK /myrepo.git/info/refs HTTP/1.1" 423 363
With this patch, there would have be two UNLOCKs in addition of the LOCKs
From the user's point of view:
- If you realize that you should have typed e.g. "git push http
master" instead of "git push http", you will have to wait for 10
minutes for the lock to expire by its own.
- Furthermore, if somebody else is dumb enough to type "git push http"
while you need to push "master" branch, then you'll need too to wait
for 10 minutes too.
Signed-off-by: Gr.ANigoire Barbier <gb@gbarbier.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This tightens the parsing of a commit object in a couple of ways.
- The "tree " header must end with a LF (earlier we did not
check this condition).
- Make sure parsing of timestamp on the "committer " header
does not go beyond the buffer, even when (1) the "author "
header does not end with a LF (this means that the commit
object is malformed and lacks the committer information) or
(2) the "committer " header does not have ">" that is the end
of the e-mail address, or (3) the "committer " header does
not end with a LF.
We however still keep the existing behaviour to return a parsed
commit object even when non-structural headers such as committer
and author are malformed, so that tools that need to look at
commits to clean up a history with such broken commits can still
get at the structural data (i.e. the parents chain and the tree
object).
Signed-off-by: Martin Koegler <mkoegler@auto.tuwien.ac.at>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Replace the "This manual page describes only the most frequently used options."
text with the list of rev-list options in git-log manpage. (The git-diff-tree
options are already included.)
Move these options to a separate file and include it from both
git-rev-list.txt and git-log.txt.
Signed-off-by: Miklos Vajna <vmiklos@frugalware.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Make http-push always fail when not compiled with USE_CURL_MULTI, since
otherwise it corrupts the remote repository (and then fails anyway).
Signed-off-by: Grégoire Barbier <gb@gbarbier.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Since we are now sanity-checking the contents of patches and
refusing to send ones with long lines, this knob provides a
way for the user to override the new behavior (if, e.g., he
knows his SMTP path will handle it).
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We try to catch errors early so that we don't end up sending
half of a broken patch series. Right now the only validation
is checking that line-lengths are under the SMTP-mandated
limit of 998.
The validation parsing is very crude (it just checks each
line length without understanding the mailbox format) but
should work fine for this simple check.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We never even look at the command line arguments until after
we have prompted the user for some information. So running
"git send-email" without arguments would prompt for "from"
and "to" headers, only to then die with "No patch files
specified." Instead, let's try to do as much error checking
as possible before getting user input.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Without this change, each diagnostic could use an errno value
clobbered by the close or unlink in rollback_lock_file.
Signed-off-by: Jim Meyering <meyering@redhat.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When the specfile (export-subst) attribute was introduced, it added a
dependency from archive-{tar|zip}.c to builtin-archive.c. This broke the
support for archive-operations in libgit.a since builtin-archive.o doesn't
belong in libgit.a.
This patch moves the functions required by libgit.a from builtin-archive.c
to the new file archive.c (which becomes part of libgit.a).
Signed-off-by: Lars Hjemli <hjemli@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Update configure.ac (and config.mak.in) by adding test for unsetenv
(NO_UNSETENV). Add comment about NO_UNSETENV to Makefile header, as
original commit 731043fd adding compat/unsetenv.c didn't do that.
Signed-off-by: Jakub Narebski <jnareb@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
fast-import was relying on the fact that on most systems mmap() and
write() are synchronized by the filesystem's buffer cache. We were
relying on the ability to mmap() 20 bytes beyond the current end
of the file, then later fill in those bytes with a future write()
call, then read them through the previously obtained mmap() address.
This isn't always true with some implementations of NFS, but it is
especially not true with our NO_MMAP=YesPlease build time option used
on some platforms. If fast-import was built with NO_MMAP=YesPlease
we used the malloc()+pread() emulation and the subsequent write()
call does not update the trailing 20 bytes of a previously obtained
"mmap()" (aka malloc'd) address.
Under NO_MMAP that behavior causes unpack_entry() in sha1_file.c to
be unable to read an object header (or data) that has been unlucky
enough to be written to the packfile at a location such that it
is in the trailing 20 bytes of a window previously opened on that
same packfile.
This bug has gone unnoticed for a very long time as it is highly data
dependent. Not only does the object have to be placed at the right
position, but it also needs to be positioned behind some other object
that has been accessed due to a branch cache invalidation. In other
words the stars had to align just right, and if you did run into
this bug you probably should also have purchased a lottery ticket.
Fortunately the workaround is a lot easier than the bug explanation.
Before we allow unpack_entry() to read data from a pack window
that has also (possibly) been modified through write() we force
all existing windows on that packfile to be closed. By closing
the windows we ensure that any new access via the emulated mmap()
will reread the packfile, updating to the current file content.
This comes at a slight performance degredation as we cannot reuse
previously cached windows when we update the packfile. But it
is a fairly minor difference as the window closes happen at only
two points:
- When the packfile is finalized and its .idx is generated:
At this stage we are getting ready to update the refs and any
data access into the packfile is going to be random, and is
going after only the branch tips (to ensure they are valid).
Our existing windows (if any) are not likely to be positioned
at useful locations to access those final tip commits so we
probably were closing them before anyway.
- When the branch cache missed and we need to reload:
At this point fast-import is getting change commands for the next
commit and it needs to go re-read a tree object it previously
had written out to the packfile. What windows we had (if any)
are not likely to cover the tree in question so we probably were
closing them before anyway.
We do try to avoid unnecessarily closing windows in the second case
by checking to see if the packfile size has increased since the
last time we called unpack_entry() on that packfile. If the size
has not changed then we have not written additional data, and any
existing window is still vaild. This nicely handles the cases where
fast-import is going through a branch cache reload and needs to read
many trees at once. During such an event we are not likely to be
updating the packfile so we do not cycle the windows between reads.
With this change in place t9301-fast-export.sh (which was broken
by c3b0dec509) finally works again.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We also move the assignment of -1 to the lock file descriptor
up, so that rollback_lock_file() can be called safely after a
possible attempt to fclose(). This matches the contents of
the 'if' statement just above testing success of fdopen().
Signed-off-by: Brandon Casey <casey@nrlssc.navy.mil>
Acked-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
