Merge branch 'nh/http'
authorJunio C Hamano <junkio@cox.net>
Sat, 8 Apr 2006 00:59:36 +0000 (17:59 -0700)
committerJunio C Hamano <junkio@cox.net>
Sat, 8 Apr 2006 00:59:36 +0000 (17:59 -0700)
* nh/http:
  Fix compile with expat, but an old curl version
  http-fetch: add optional DAV-based pack list

26 files changed:
.gitignore
Documentation/diff-options.txt
Documentation/git-clean.txt [new file with mode: 0644]
Documentation/git-commit.txt
Documentation/git-imap-send.txt [new file with mode: 0644]
Documentation/technical/pack-format.txt [new file with mode: 0644]
Documentation/technical/pack-heuristics.txt [new file with mode: 0644]
Makefile
blame.c
combine-diff.c
commit.c
commit.h
date.c
delta.h
diff.c
diff.h
diffcore-pickaxe.c
git-clean.sh [new file with mode: 0755]
git-clone.sh
gitk
imap-send.c
patch-delta.c
rev-list.c
sha1_file.c
xdiff-interface.c [new file with mode: 0644]
xdiff-interface.h [new file with mode: 0644]

index 75891c3..b5959d6 100644 (file)
@@ -15,6 +15,7 @@ git-checkout
 git-checkout-index
 git-cherry
 git-cherry-pick
+git-clean
 git-clone
 git-clone-pack
 git-commit
index 2a0275e..ec6811c 100644 (file)
        changeset, not just the files that contain the change
        in <string>.
 
+--pickaxe-regex::
+       Make the <string> not a plain string but an extended POSIX
+       regex to match.
+
 -O<orderfile>::
        Output the patch in the order specified in the
        <orderfile>, which has one shell glob pattern per line.
diff --git a/Documentation/git-clean.txt b/Documentation/git-clean.txt
new file mode 100644 (file)
index 0000000..36890c5
--- /dev/null
@@ -0,0 +1,50 @@
+git-clean(1)
+============
+
+NAME
+----
+git-clean - Remove untracked files from the working tree
+
+SYNOPSIS
+--------
+[verse]
+'git-clean' [-d] [-n] [-q] [-x | -X]
+
+DESCRIPTION
+-----------
+Removes files unknown to git.  This allows to clean the working tree
+from files that are not under version control.  If the '-x' option is
+specified, ignored files are also removed, allowing to remove all
+build products.
+
+OPTIONS
+-------
+-d::
+       Remove untracked directories in addition to untracked files.
+
+-n::
+       Don't actually remove anything, just show what would be done.
+
+-q::
+       Be quiet, only report errors, but not the files that are
+       successfully removed.
+
+-x::
+       Don't use the ignore rules.  This allows removing all untracked
+       files, including build products.  This can be used (possibly in
+       conjunction with gitlink:git-reset[1]) to create a pristine
+       working directory to test a clean build.
+
+-X::
+       Remove only files ignored by git.  This may be useful to rebuild
+       everything from scratch, but keep manually created files.
+
+
+Author
+------
+Written by Pavel Roskin <proski@gnu.org>
+
+
+GIT
+---
+Part of the gitlink:git[7] suite
index d04b342..0a7365b 100644 (file)
@@ -9,7 +9,8 @@ SYNOPSIS
 --------
 [verse]
 'git-commit' [-a] [-s] [-v] [(-c | -C) <commit> | -F <file> | -m <msg>]
-          [-e] [--author <author>] [--] [[-i | -o ]<file>...]
+          [--no-verify] [--amend] [-e] [--author <author>]
+          [--] [[-i | -o ]<file>...]
 
 DESCRIPTION
 -----------
@@ -71,6 +72,28 @@ OPTIONS
        commit log message unmodified.  This option lets you
        further edit the message taken from these sources.
 
+--amend::
+
+       Used to amend the tip of the current branch. Prepare the tree
+       object you would want to replace the latest commit as usual
+       (this includes the usual -i/-o and explicit paths), and the
+       commit log editor is seeded with the commit message from the
+       tip of the current branch. The commit you create replaces the
+       current tip -- if it was a merge, it will have the parents of
+       the current tip as parents -- so the current top commit is
+       discarded.
++
+--
+It is a rough equivalent for:
+------
+       $ git reset --soft HEAD^
+       $ ... do something else to come up with the right tree ...
+       $ git commit -c ORIG_HEAD
+
+------
+but can be used to amend a merge commit.
+--
+
 -i|--include::
        Instead of committing only the files specified on the
        command line, update them in the index file and then
diff --git a/Documentation/git-imap-send.txt b/Documentation/git-imap-send.txt
new file mode 100644 (file)
index 0000000..cfc0d88
--- /dev/null
@@ -0,0 +1,60 @@
+git-imap-send(1)
+================
+
+NAME
+----
+git-imap-send - Dump a mailbox from stdin into an imap folder
+
+
+SYNOPSIS
+--------
+'git-imap-send'
+
+
+DESCRIPTION
+-----------
+This command uploads a mailbox generated with git-format-patch
+into an imap drafts folder.  This allows patches to be sent as
+other email is sent with mail clients that cannot read mailbox
+files directly.
+
+Typical usage is something like:
+
+git-format-patch --signoff --stdout --attach origin | git-imap-send
+
+
+CONFIGURATION
+-------------
+
+git-imap-send requires the following values in the repository
+configuration file (shown with examples):
+
+[imap]
+    Folder = "INBOX.Drafts"
+
+[imap]
+    Tunnel = "ssh -q user@server.com /usr/bin/imapd ./Maildir 2> /dev/null"
+
+[imap]
+    Host = imap.server.com
+    User = bob
+    Password = pwd
+    Port = 143
+
+
+BUGS
+----
+Doesn't handle lines starting with "From " in the message body.
+
+
+Author
+------
+Derived from isync 1.0.1 by Mike McCormack.
+
+Documentation
+--------------
+Documentation by Mike McCormack
+
+GIT
+---
+Part of the gitlink:git[7] suite
diff --git a/Documentation/technical/pack-format.txt b/Documentation/technical/pack-format.txt
new file mode 100644 (file)
index 0000000..ed2decc
--- /dev/null
@@ -0,0 +1,111 @@
+GIT pack format
+===============
+
+= pack-*.pack file has the following format:
+
+   - The header appears at the beginning and consists of the following:
+
+     4-byte signature
+     4-byte version number (network byte order)
+     4-byte number of objects contained in the pack (network byte order)
+
+     Observation: we cannot have more than 4G versions ;-) and
+     more than 4G objects in a pack.
+
+   - The header is followed by number of object entries, each of
+     which looks like this:
+
+     (undeltified representation)
+     n-byte type and length (4-bit type, (n-1)*7+4-bit length)
+     compressed data
+
+     (deltified representation)
+     n-byte type and length (4-bit type, (n-1)*7+4-bit length)
+     20-byte base object name
+     compressed delta data
+
+     Observation: length of each object is encoded in a variable
+     length format and is not constrained to 32-bit or anything.
+
+  - The trailer records 20-byte SHA1 checksum of all of the above.
+
+= pack-*.idx file has the following format:
+
+  - The header consists of 256 4-byte network byte order
+    integers.  N-th entry of this table records the number of
+    objects in the corresponding pack, the first byte of whose
+    object name are smaller than N.  This is called the
+    'first-level fan-out' table.
+
+    Observation: we would need to extend this to an array of
+    8-byte integers to go beyond 4G objects per pack, but it is
+    not strictly necessary.
+
+  - The header is followed by sorted 28-byte entries, one entry
+    per object in the pack.  Each entry is:
+
+    4-byte network byte order integer, recording where the
+    object is stored in the packfile as the offset from the
+    beginning.
+
+    20-byte object name.
+
+    Observation: we would definitely need to extend this to
+    8-byte integer plus 20-byte object name to handle a packfile
+    that is larger than 4GB.
+
+  - The file is concluded with a trailer:
+
+    A copy of the 20-byte SHA1 checksum at the end of
+    corresponding packfile.
+
+    20-byte SHA1-checksum of all of the above.
+
+Pack Idx file:
+
+       idx
+           +--------------------------------+
+           | fanout[0] = 2                  |-.
+           +--------------------------------+ |
+           | fanout[1]                      | |
+           +--------------------------------+ |
+           | fanout[2]                      | |
+           ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
+           | fanout[255]                    | |
+           +--------------------------------+ |
+main       | offset                         | |
+index      | object name 00XXXXXXXXXXXXXXXX | |
+table      +--------------------------------+ | 
+           | offset                         | |
+           | object name 00XXXXXXXXXXXXXXXX | |
+           +--------------------------------+ |
+         .-| offset                         |<+
+         | | object name 01XXXXXXXXXXXXXXXX |
+         | +--------------------------------+
+         | | offset                         |
+         | | object name 01XXXXXXXXXXXXXXXX |
+         | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+         | | offset                         |
+         | | object name FFXXXXXXXXXXXXXXXX |
+         | +--------------------------------+
+trailer          | | packfile checksum              |
+         | +--------------------------------+
+         | | idxfile checksum               |
+         | +--------------------------------+
+          .-------.      
+                  |
+Pack file entry: <+
+
+     packed object header:
+       1-byte type (upper 4-bit)
+              size0 (lower 4-bit) 
+        n-byte sizeN (as long as MSB is set, each 7-bit)
+               size0..sizeN form 4+7+7+..+7 bit integer, size0
+               is the most significant part.
+     packed object data:
+        If it is not DELTA, then deflated bytes (the size above
+               is the size before compression).
+       If it is DELTA, then
+         20-byte base object name SHA1 (the size above is the
+               size of the delta data that follows).
+          delta data, deflated.
diff --git a/Documentation/technical/pack-heuristics.txt b/Documentation/technical/pack-heuristics.txt
new file mode 100644 (file)
index 0000000..eaab3ee
--- /dev/null
@@ -0,0 +1,466 @@
+        Concerning Git's Packing Heuristics
+        ===================================
+
+        Oh, here's a really stupid question:
+
+                  Where do I go
+               to learn the details
+           of git's packing heuristics?
+
+Be careful what you ask!
+
+Followers of the git, please open the git IRC Log and turn to
+February 10, 2006.
+
+It's a rare occasion, and we are joined by the King Git Himself,
+Linus Torvalds (linus).  Nathaniel Smith, (njs`), has the floor
+and seeks enlightenment.  Others are present, but silent.
+
+Let's listen in!
+
+    <njs`> Oh, here's a really stupid question -- where do I go to
+        learn the details of git's packing heuristics?  google avails
+        me not, reading the source didn't help a lot, and wading
+        through the whole mailing list seems less efficient than any
+        of that.
+
+It is a bold start!  A plea for help combined with a simultaneous
+tri-part attack on some of the tried and true mainstays in the quest
+for enlightenment.  Brash accusations of google being useless. Hubris!
+Maligning the source.  Heresy!  Disdain for the mailing list archives.
+Woe.
+
+    <pasky> yes, the packing-related delta stuff is somewhat
+        mysterious even for me ;)
+
+Ah!  Modesty after all.
+
+    <linus> njs, I don't think the docs exist. That's something where
+        I don't think anybody else than me even really got involved.
+        Most of the rest of git others have been busy with (especially
+        Junio), but packing nobody touched after I did it.
+
+It's cryptic, yet vague.  Linus in style for sure.  Wise men
+interpret this as an apology.  A few argue it is merely a
+statement of fact.
+
+    <njs`> I guess the next step is "read the source again", but I
+        have to build up a certain level of gumption first :-)
+
+Indeed!  On both points.
+
+    <linus> The packing heuristic is actually really really simple.
+
+Bait...
+
+    <linus> But strange.
+
+And switch.  That ought to do it!
+
+    <linus> Remember: git really doesn't follow files. So what it does is
+        - generate a list of all objects
+        - sort the list according to magic heuristics
+        - walk the list, using a sliding window, seeing if an object
+          can be diffed against another object in the window
+        - write out the list in recency order
+
+The traditional understatement:
+
+    <njs`> I suspect that what I'm missing is the precise definition of
+        the word "magic"
+
+The traditional insight:
+
+    <pasky> yes
+
+And Bable-like confusion flowed.
+
+    <njs`> oh, hmm, and I'm not sure what this sliding window means either
+
+    <pasky> iirc, it appeared to me to be just the sha1 of the object
+        when reading the code casually ...
+
+        ... which simply doesn't sound as a very good heuristics, though ;)
+
+    <njs`> .....and recency order.  okay, I think it's clear I didn't
+       even realize how much I wasn't realizing :-)
+
+Ah, grasshopper!  And thus the enlightenment begins anew.
+
+    <linus> The "magic" is actually in theory totally arbitrary.
+        ANY order will give you a working pack, but no, it's not
+        ordered by SHA1.
+
+        Before talking about the ordering for the sliding delta
+        window, let's talk about the recency order. That's more
+        important in one way.
+
+    <njs`> Right, but if all you want is a working way to pack things
+        together, you could just use cat and save yourself some
+        trouble...
+
+Waaait for it....
+
+    <linus> The recency ordering (which is basically: put objects
+        _physically_ into the pack in the order that they are
+        "reachable" from the head) is important.
+
+    <njs`> okay
+
+    <linus> It's important because that's the thing that gives packs
+        good locality. It keeps the objects close to the head (whether
+        they are old or new, but they are _reachable_ from the head)
+        at the head of the pack. So packs actually have absolutely
+        _wonderful_ IO patterns.
+
+Read that again, because it is important.
+
+    <linus> But recency ordering is totally useless for deciding how
+        to actually generate the deltas, so the delta ordering is
+        something else.
+
+        The delta ordering is (wait for it):
+        - first sort by the "basename" of the object, as defined by
+          the name the object was _first_ reached through when
+          generating the object list
+        - within the same basename, sort by size of the object
+        - but always sort different types separately (commits first).
+
+        That's not exactly it, but it's very close.
+
+    <njs`> The "_first_ reached" thing is not too important, just you
+        need some way to break ties since the same objects may be
+        reachable many ways, yes?
+
+And as if to clarify:
+
+    <linus> The point is that it's all really just any random
+        heuristic, and the ordering is totally unimportant for
+        correctness, but it helps a lot if the heuristic gives
+        "clumping" for things that are likely to delta well against
+        each other.
+
+It is an important point, so secretly, I did my own research and have
+included my results below.  To be fair, it has changed some over time.
+And through the magic of Revisionistic History, I draw upon this entry
+from The Git IRC Logs on my father's birthday, March 1:
+
+    <gitster> The quote from the above linus should be rewritten a
+        bit (wait for it):
+        - first sort by type.  Different objects never delta with
+         each other.
+        - then sort by filename/dirname.  hash of the basename
+          occupies the top BITS_PER_INT-DIR_BITS bits, and bottom
+          DIR_BITS are for the hash of leading path elements.
+        - then if we are doing "thin" pack, the objects we are _not_
+          going to pack but we know about are sorted earlier than
+          other objects.
+        - and finally sort by size, larger to smaller.
+
+In one swell-foop, clarification and obscurification!  Nonetheless,
+authoritative.  Cryptic, yet concise.  It even solicits notions of
+quotes from The Source Code.  Clearly, more study is needed.
+
+    <gitster> That's the sort order.  What this means is:
+        - we do not delta different object types.
+       - we prefer to delta the objects with the same full path, but
+          allow files with the same name from different directories.
+       - we always prefer to delta against objects we are not going
+          to send, if there are some.
+       - we prefer to delta against larger objects, so that we have
+          lots of removals.
+
+        The penultimate rule is for "thin" packs.  It is used when
+        the other side is known to have such objects.
+
+There it is again. "Thin" packs.  I'm thinking to myself, "What
+is a 'thin' pack?"  So I ask:
+
+    <jdl> What is a "thin" pack?
+
+    <gitster> Use of --objects-edge to rev-list as the upstream of
+        pack-objects.  The pack transfer protocol negotiates that.
+
+Woo hoo!  Cleared that _right_ up!
+
+    <gitster> There are two directions - push and fetch.
+
+There!  Did you see it?  It is not '"push" and "pull"'!  How often the
+confusion has started here.  So casually mentioned, too!
+
+    <gitster> For push, git-send-pack invokes git-receive-pack on the
+        other end.  The receive-pack says "I have up to these commits".
+        send-pack looks at them, and computes what are missing from
+        the other end.  So "thin" could be the default there.
+
+        In the other direction, fetch, git-fetch-pack and
+        git-clone-pack invokes git-upload-pack on the other end
+       (via ssh or by talking to the daemon).
+
+       There are two cases: fetch-pack with -k and clone-pack is one,
+        fetch-pack without -k is the other.  clone-pack and fetch-pack
+        with -k will keep the downloaded packfile without expanded, so
+        we do not use thin pack transfer.  Otherwise, the generated
+        pack will have delta without base object in the same pack.
+
+        But fetch-pack without -k will explode the received pack into
+        individual objects, so we automatically ask upload-pack to
+        give us a thin pack if upload-pack supports it.
+
+OK then.
+
+Uh.
+
+Let's return to the previous conversation still in progress.
+
+    <njs`> and "basename" means something like "the tail of end of
+        path of file objects and dir objects, as per basename(3), and
+        we just declare all commit and tag objects to have the same
+        basename" or something?
+
+Luckily, that too is a point that gitster clarified for us!
+
+If I might add, the trick is to make files that _might_ be similar be
+located close to each other in the hash buckets based on their file
+names.  It used to be that "foo/Makefile", "bar/baz/quux/Makefile" and
+"Makefile" all landed in the same bucket due to their common basename,
+"Makefile". However, now they land in "close" buckets.
+
+The algorithm allows not just for the _same_ bucket, but for _close_
+buckets to be considered delta candidates.  The rationale is
+essentially that files, like Makefiles, often have very similar
+content no matter what directory they live in.
+
+    <linus> I played around with different delta algorithms, and with
+        making the "delta window" bigger, but having too big of a
+        sliding window makes it very expensive to generate the pack:
+        you need to compare every object with a _ton_ of other objects.
+
+        There are a number of other trivial heuristics too, which
+        basically boil down to "don't bother even trying to delta this
+        pair" if we can tell before-hand that the delta isn't worth it
+        (due to size differences, where we can take a previous delta
+        result into account to decide that "ok, no point in trying
+        that one, it will be worse").
+
+        End result: packing is actually very size efficient. It's
+        somewhat CPU-wasteful, but on the other hand, since you're
+        really only supposed to do it maybe once a month (and you can
+        do it during the night), nobody really seems to care.
+
+Nice Engineering Touch, there.  Find when it doesn't matter, and
+proclaim it a non-issue.  Good style too!
+
+    <njs`> So, just to repeat to see if I'm following, we start by
+        getting a list of the objects we want to pack, we sort it by
+        this heuristic (basically lexicographically on the tuple
+        (type, basename, size)).
+
+        Then we walk through this list, and calculate a delta of
+        each object against the last n (tunable paramater) objects,
+        and pick the smallest of these deltas.
+
+Vastly simplified, but the essence is there!
+
+    <linus> Correct.
+
+    <njs`> And then once we have picked a delta or fulltext to
+        represent each object, we re-sort by recency, and write them
+        out in that order.
+
+    <linus> Yup. Some other small details:
+
+And of course there is the "Other Shoe" Factor too.
+
+    <linus> - We limit the delta depth to another magic value (right
+        now both the window and delta depth magic values are just "10")
+
+    <njs`> Hrm, my intuition is that you'd end up with really _bad_ IO
+        patterns, because the things you want are near by, but to
+        actually reconstruct them you may have to jump all over in
+        random ways.
+
+    <linus> - When we write out a delta, and we haven't yet written
+        out the object it is a delta against, we write out the base
+        object first.  And no, when we reconstruct them, we actually
+        get nice IO patterns, because:
+        - larger objects tend to be "more recent" (Linus' law: files grow)
+        - we actively try to generate deltas from a larger object to a
+          smaller one
+        - this means that the top-of-tree very seldom has deltas
+          (ie deltas in _practice_ are "backwards deltas")
+
+Again, we should reread that whole paragraph.  Not just because
+Linus has slipped Linus's Law in there on us, but because it is
+important.  Let's make sure we clarify some of the points here:
+
+    <njs`> So the point is just that in practice, delta order and
+        recency order match each other quite well.
+
+    <linus> Yes. There's another nice side to this (and yes, it was
+       designed that way ;):
+        - the reason we generate deltas against the larger object is
+         actually a big space saver too!
+
+    <njs`> Hmm, but your last comment (if "we haven't yet written out
+        the object it is a delta against, we write out the base object
+        first"), seems like it would make these facts mostly
+        irrelevant because even if in practice you would not have to
+        wander around much, in fact you just brute-force say that in
+        the cases where you might have to wander, don't do that :-)
+
+    <linus> Yes and no. Notice the rule: we only write out the base
+        object first if the delta against it was more recent.  That
+        means that you can actually have deltas that refer to a base
+        object that is _not_ close to the delta object, but that only
+        happens when the delta is needed to generate an _old_ object.
+
+    <linus> See?
+
+Yeah, no.  I missed that on the first two or three readings myself.
+
+    <linus> This keeps the front of the pack dense. The front of the
+        pack never contains data that isn't relevant to a "recent"
+        object.  The size optimization comes from our use of xdelta
+        (but is true for many other delta algorithms): removing data
+        is cheaper (in size) than adding data.
+
+        When you remove data, you only need to say "copy bytes n--m".
+       In contrast, in a delta that _adds_ data, you have to say "add
+        these bytes: 'actual data goes here'"
+
+    *** njs` has quit: Read error: 104 (Connection reset by peer)
+
+    <linus> Uhhuh. I hope I didn't blow njs` mind.
+
+    *** njs` has joined channel #git
+
+    <pasky> :)
+
+The silent observers are amused.  Of course.
+
+And as if njs` was expected to be omniscient:
+
+    <linus> njs - did you miss anything?
+
+OK, I'll spell it out.  That's Geek Humor.  If njs` was not actually
+connected for a little bit there, how would he know if missed anything
+while he was disconnected?  He's a benevolent dictator with a sense of
+humor!  Well noted!
+
+    <njs`> Stupid router.  Or gremlins, or whatever.
+
+It's a cheap shot at Cisco.  Take 'em when you can.
+
+    <njs`> Yes and no. Notice the rule: we only write out the base
+        object first if the delta against it was more recent.
+
+        I'm getting lost in all these orders, let me re-read :-)
+       So the write-out order is from most recent to least recent?
+        (Conceivably it could be the opposite way too, I'm not sure if
+        we've said) though my connection back at home is logging, so I
+        can just read what you said there :-)
+
+And for those of you paying attention, the Omniscient Trick has just
+been detailed!
+
+    <linus> Yes, we always write out most recent first
+
+For the other record:
+
+    <pasky> njs`: http://pastebin.com/547965
+
+The 'net never forgets, so that should be good until the end of time.
+
+    <njs`> And, yeah, I got the part about deeper-in-history stuff
+        having worse IO characteristics, one sort of doesn't care.
+
+    <linus> With the caveat that if the "most recent" needs an older
+        object to delta against (hey, shrinking sometimes does
+        happen), we write out the old object with the delta.
+
+    <njs`> (if only it happened more...)
+
+    <linus> Anyway, the pack-file could easily be denser still, but
+        because it's used both for streaming (the git protocol) and
+        for on-disk, it has a few pessimizations.
+
+Actually, it is a made-up word. But it is a made-up word being
+used as setup for a later optimization, which is a real word:
+
+    <linus> In particular, while the pack-file is then compressed,
+        it's compressed just one object at a time, so the actual
+        compression factor is less than it could be in theory. But it
+        means that it's all nice random-access with a simple index to
+        do "object name->location in packfile" translation.
+
+    <njs`> I'm assuming the real win for delta-ing large->small is
+        more homogenous statistics for gzip to run over?
+
+        (You have to put the bytes in one place or another, but
+        putting them in a larger blob wins on compression)
+
+        Actually, what is the compression strategy -- each delta
+        individually gzipped, the whole file gzipped, somewhere in
+        between, no compression at all, ....?
+
+        Right.
+
+Reality IRC sets in.  For example:
+
+    <pasky> I'll read the rest in the morning, I really have to go
+        sleep or there's no hope whatsoever for me at the today's
+        exam... g'nite all.
+
+Heh.
+
+    <linus> pasky: g'nite
+
+    <njs`> pasky: 'luck
+
+    <linus> Right: large->small matters exactly because of compression
+        behaviour. If it was non-compressed, it probably wouldn't make
+        any difference.
+
+    <njs`> yeah
+
+    <linus> Anyway: I'm not even trying to claim that the pack-files
+        are perfect, but they do tend to have a nice balance of
+        density vs ease-of use.
+
+Gasp!  OK, saved.  That's a fair Engineering trade off.  Close call!
+In fact, Linus reflects on some Basic Engineering Fundamentals,
+design options, etc.
+
+    <linus> More importantly, they allow git to still _conceptually_
+        never deal with deltas at all, and be a "whole object" store.
+
+        Which has some problems (we discussed bad huge-file
+        behaviour on the git lists the other day), but it does mean
+        that the basic git concepts are really really simple and
+        straightforward.
+
+        It's all been quite stable.
+
+        Which I think is very much a result of having very simple
+        basic ideas, so that there's never any confusion about what's
+        going on.
+
+        Bugs happen, but they are "simple" bugs. And bugs that
+        actually get some object store detail wrong are almost always
+        so obious that they never go anywhere.
+
+    <njs`> Yeah.
+
+Nuff said.
+
+    <linus> Anyway.  I'm off for bed. It's not 6AM here, but I've got
+        three kids, and have to get up early in the morning to send
+        them off. I need my beauty sleep.
+
+    <njs`> :-)
+
+    <njs`> appreciate the infodump, I really was failing to find the
+        details on git packs :-)
+
+And now you know the rest of the story.
index 1641519..6b10eaa 100644 (file)
--- a/Makefile
+++ b/Makefile
@@ -114,7 +114,7 @@ SPARSE_FLAGS = -D__BIG_ENDIAN__ -D__powerpc__
 
 SCRIPT_SH = \
        git-add.sh git-bisect.sh git-branch.sh git-checkout.sh \
-       git-cherry.sh git-clone.sh git-commit.sh \
+       git-cherry.sh git-clean.sh git-clone.sh git-commit.sh \
        git-count-objects.sh git-diff.sh git-fetch.sh \
        git-format-patch.sh git-log.sh git-ls-remote.sh \
        git-merge-one-file.sh git-parse-remote.sh \
@@ -208,7 +208,7 @@ LIB_OBJS = \
        quote.o read-cache.o refs.o run-command.o \
        server-info.o setup.o sha1_file.o sha1_name.o strbuf.o \
        tag.o tree.o usage.o config.o environment.o ctype.o copy.o \
-       fetch-clone.o revision.o pager.o tree-walk.o \
+       fetch-clone.o revision.o pager.o tree-walk.o xdiff-interface.o \
        $(DIFF_OBJS)
 
 GITLIBS = $(LIB_FILE) $(XDIFF_LIB)
diff --git a/blame.c b/blame.c
index 98f9992..6730b10 100644 (file)
--- a/blame.c
+++ b/blame.c
@@ -16,6 +16,7 @@
 #include "diff.h"
 #include "diffcore.h"
 #include "revision.h"
+#include "xdiff-interface.h"
 
 #define DEBUG 0
 
@@ -57,116 +58,89 @@ static int num_get_patch = 0;
 static int num_commits = 0;
 static int patch_time = 0;
 
-#define TEMPFILE_PATH_LEN 60
-static struct patch *get_patch(struct commit *commit, struct commit *other)
-{
+struct blame_diff_state {
+       struct xdiff_emit_state xm;
        struct patch *ret;
-       struct util_info *info_c = (struct util_info *)commit->object.util;
-       struct util_info *info_o = (struct util_info *)other->object.util;
-       char tmp_path1[TEMPFILE_PATH_LEN], tmp_path2[TEMPFILE_PATH_LEN];
-       char diff_cmd[TEMPFILE_PATH_LEN*2 + 20];
-       struct timeval tv_start, tv_end;
-       int fd;
-       FILE *fin;
-       char buf[1024];
-
-       ret = xmalloc(sizeof(struct patch));
-       ret->chunks = NULL;
-       ret->num = 0;
-
-       get_blob(commit);
-       get_blob(other);
+};
 
-       gettimeofday(&tv_start, NULL);
+static void process_u0_diff(void *state_, char *line, unsigned long len)
+{
+       struct blame_diff_state *state = state_;
+       struct chunk *chunk;
 
-       fd = git_mkstemp(tmp_path1, TEMPFILE_PATH_LEN, "git-blame-XXXXXX");
-       if (fd < 0)
-               die("unable to create temp-file: %s", strerror(errno));
+       if (len < 4 || line[0] != '@' || line[1] != '@')
+               return;
 
-       if (xwrite(fd, info_c->buf, info_c->size) != info_c->size)
-               die("write failed: %s", strerror(errno));
-       close(fd);
+       if (DEBUG)
+               printf("chunk line: %.*s", (int)len, line);
+       state->ret->num++;
+       state->ret->chunks = xrealloc(state->ret->chunks,
+                                     sizeof(struct chunk) * state->ret->num);
+       chunk = &state->ret->chunks[state->ret->num - 1];
+
+       assert(!strncmp(line, "@@ -", 4));
+
+       if (parse_hunk_header(line, len,
+                             &chunk->off1, &chunk->len1,
+                             &chunk->off2, &chunk->len2)) {
+               state->ret->num--;
+               return;
+       }
 
-       fd = git_mkstemp(tmp_path2, TEMPFILE_PATH_LEN, "git-blame-XXXXXX");
-       if (fd < 0)
-               die("unable to create temp-file: %s", strerror(errno));
+       if (chunk->len1 == 0)
+               chunk->off1++;
+       if (chunk->len2 == 0)
+               chunk->off2++;
 
-       if (xwrite(fd, info_o->buf, info_o->size) != info_o->size)
-               die("write failed: %s", strerror(errno));
-       close(fd);
+       if (chunk->off1 > 0)
+               chunk->off1--;
+       if (chunk->off2 > 0)
+               chunk->off2--;
 
-       sprintf(diff_cmd, "diff -U 0 %s %s", tmp_path1, tmp_path2);
-       fin = popen(diff_cmd, "r");
-       if (!fin)
-               die("popen failed: %s", strerror(errno));
+       assert(chunk->off1 >= 0);
+       assert(chunk->off2 >= 0);
+}
 
-       while (fgets(buf, sizeof(buf), fin)) {
-               struct chunk *chunk;
-               char *start, *sp;
+static struct patch *get_patch(struct commit *commit, struct commit *other)
+{
+       struct blame_diff_state state;
+       xpparam_t xpp;
+       xdemitconf_t xecfg;
+       mmfile_t file_c, file_o;
+       xdemitcb_t ecb;
+       struct util_info *info_c = (struct util_info *)commit->object.util;
+       struct util_info *info_o = (struct util_info *)other->object.util;
+       struct timeval tv_start, tv_end;
 
-               if (buf[0] != '@' || buf[1] != '@')
-                       continue;
+       get_blob(commit);
+       file_c.ptr = info_c->buf;
+       file_c.size = info_c->size;
 
-               if (DEBUG)
-                       printf("chunk line: %s", buf);
-               ret->num++;
-               ret->chunks = xrealloc(ret->chunks,
-                                      sizeof(struct chunk) * ret->num);
-               chunk = &ret->chunks[ret->num - 1];
-
-               assert(!strncmp(buf, "@@ -", 4));
-
-               start = buf + 4;
-               sp = index(start, ' ');
-               *sp = '\0';
-               if (index(start, ',')) {
-                       int ret =
-                           sscanf(start, "%d,%d", &chunk->off1, &chunk->len1);
-                       assert(ret == 2);
-               } else {
-                       int ret = sscanf(start, "%d", &chunk->off1);
-                       assert(ret == 1);
-                       chunk->len1 = 1;
-               }
-               *sp = ' ';
-
-               start = sp + 1;
-               sp = index(start, ' ');
-               *sp = '\0';
-               if (index(start, ',')) {
-                       int ret =
-                           sscanf(start, "%d,%d", &chunk->off2, &chunk->len2);
-                       assert(ret == 2);
-               } else {
-                       int ret = sscanf(start, "%d", &chunk->off2);
-                       assert(ret == 1);
-                       chunk->len2 = 1;
-               }
-               *sp = ' ';
+       get_blob(other);
+       file_o.ptr = info_o->buf;
+       file_o.size = info_o->size;
 
-               if (chunk->len1 == 0)
-                       chunk->off1++;
-               if (chunk->len2 == 0)
-                       chunk->off2++;
+       gettimeofday(&tv_start, NULL);
 
-               if (chunk->off1 > 0)
-                       chunk->off1--;
-               if (chunk->off2 > 0)
-                       chunk->off2--;
+       xpp.flags = XDF_NEED_MINIMAL;
+       xecfg.ctxlen = 0;
+       xecfg.flags = 0;
+       ecb.outf = xdiff_outf;
+       ecb.priv = &state;
+       memset(&state, 0, sizeof(state));
+       state.xm.consume = process_u0_diff;
+       state.ret = xmalloc(sizeof(struct patch));
+       state.ret->chunks = NULL;
+       state.ret->num = 0;
 
-               assert(chunk->off1 >= 0);
-               assert(chunk->off2 >= 0);
-       }
-       pclose(fin);
-       unlink(tmp_path1);
-       unlink(tmp_path2);
+       xdl_diff(&file_c, &file_o, &xpp, &xecfg, &ecb);
 
        gettimeofday(&tv_end, NULL);
        patch_time += 1000000 * (tv_end.tv_sec - tv_start.tv_sec) +
                tv_end.tv_usec - tv_start.tv_usec;
 
        num_get_patch++;
-       return ret;
+       return state.ret;
 }
 
 static void free_patch(struct patch *p)
@@ -674,7 +648,7 @@ static void get_commit_info(struct commit* commit, struct commit_info* ret)
        static char author_buf[1024];
 
        tmp = strstr(commit->buffer, "\nauthor ") + 8;
-       len = index(tmp, '\n') - tmp;
+       len = strchr(tmp, '\n') - tmp;
        ret->author = author_buf;
        memcpy(ret->author, tmp, len);
 
@@ -729,11 +703,30 @@ static void* topo_getter(struct commit* c)
        return util->topo_data;
 }
 
+static int read_ancestry(const char *graft_file,
+                        unsigned char **start_sha1)
+{
+       FILE *fp = fopen(graft_file, "r");
+       char buf[1024];
+       if (!fp)
+               return -1;
+       while (fgets(buf, sizeof(buf), fp)) {
+               /* The format is just "Commit Parent1 Parent2 ...\n" */
+               int len = strlen(buf);
+               struct commit_graft *graft = read_graft_line(buf, len);
+               register_commit_graft(graft, 0);
+               if (!*start_sha1)
+                       *start_sha1 = graft->sha1;
+       }
+       fclose(fp);
+       return 0;
+}
+
 int main(int argc, const char **argv)
 {
        int i;
        struct commit *initial = NULL;
-       unsigned char sha1[20];
+       unsigned char sha1[20], *sha1_p = NULL;
 
        const char *filename = NULL, *commit = NULL;
        char filename_buf[256];
@@ -767,6 +760,14 @@ int main(int argc, const char **argv)
                                  !strcmp(argv[i], "--compability")) {
                                compability = 1;
                                continue;
+                       } else if(!strcmp(argv[i], "-S")) {
+                               if (i + 1 < argc &&
+                                   !read_ancestry(argv[i + 1], &sha1_p)) {
+                                       compability = 1;
+                                       i++;
+                                       continue;
+                               }
+                               usage(blame_usage);
                        } else if(!strcmp(argv[i], "--")) {
                                options = 0;
                                continue;
@@ -788,7 +789,9 @@ int main(int argc, const char **argv)
 
        if(!filename)
                usage(blame_usage);
-       if(!commit)
+       if (commit && sha1_p)
+               usage(blame_usage);
+       else if(!commit)
                commit = "HEAD";
 
        if(prefix)
@@ -797,9 +800,12 @@ int main(int argc, const char **argv)
                strcpy(filename_buf, filename);
        filename = filename_buf;
 
-       if (get_sha1(commit, sha1))
-               die("get_sha1 failed, commit '%s' not found", commit);
-       start_commit = lookup_commit_reference(sha1);
+       if (!sha1_p) {
+               if (get_sha1(commit, sha1))
+                       die("get_sha1 failed, commit '%s' not found", commit);
+               sha1_p = sha1;
+       }
+       start_commit = lookup_commit_reference(sha1_p);
        get_util(start_commit)->pathname = filename;
        if (fill_util_info(start_commit)) {
                printf("%s not found in %s\n", filename, commit);
@@ -813,6 +819,7 @@ int main(int argc, const char **argv)
        rev.prune_fn = simplify_commit;
        rev.topo_setter = topo_setter;
        rev.topo_getter = topo_getter;
+       rev.parents = 1;
        rev.limited = 1;
 
        commit_list_insert(start_commit, &rev.commits);
@@ -875,7 +882,7 @@ int main(int argc, const char **argv)
                        if(blame_contents[blame_len-1] != '\n')
                                putc('\n', stdout);
                } else {
-                       char* next_buf = index(buf, '\n') + 1;
+                       char* next_buf = strchr(buf, '\n') + 1;
                        fwrite(buf, next_buf - buf, 1, stdout);
                        buf = next_buf;
                }
index 7693884..eb0d757 100644 (file)
@@ -4,6 +4,7 @@
 #include "diff.h"
 #include "diffcore.h"
 #include "quote.h"
+#include "xdiff-interface.h"
 
 static int uninteresting(struct diff_filepair *p)
 {
@@ -110,78 +111,9 @@ static char *grab_blob(const unsigned char *sha1, unsigned long *size)
        return blob;
 }
 
-#define TMPPATHLEN 50
-#define MAXLINELEN 10240
-
-static void write_to_temp_file(char *tmpfile, void *blob, unsigned long size)
-{
-       int fd = git_mkstemp(tmpfile, TMPPATHLEN, ".diff_XXXXXX");
-       if (fd < 0)
-               die("unable to create temp-file");
-       if (write(fd, blob, size) != size)
-               die("unable to write temp-file");
-       close(fd);
-}
-
-static void write_temp_blob(char *tmpfile, const unsigned char *sha1)
-{
-       unsigned long size;
-       void *blob;
-       blob = grab_blob(sha1, &size);
-       write_to_temp_file(tmpfile, blob, size);
-       free(blob);
-}
-
-static int parse_num(char **cp_p, unsigned int *num_p)
-{
-       char *cp = *cp_p;
-       unsigned int num = 0;
-       int read_some;
-
-       while ('0' <= *cp && *cp <= '9')
-               num = num * 10 + *cp++ - '0';
-       if (!(read_some = cp - *cp_p))
-               return -1;
-       *cp_p = cp;
-       *num_p = num;
-       return 0;
-}
-
-static int parse_hunk_header(char *line, int len,
-                            unsigned int *ob, unsigned int *on,
-                            unsigned int *nb, unsigned int *nn)
-{
-       char *cp;
-       cp = line + 4;
-       if (parse_num(&cp, ob)) {
-       bad_line:
-               return error("malformed diff output: %s", line);
-       }
-       if (*cp == ',') {
-               cp++;
-               if (parse_num(&cp, on))
-                       goto bad_line;
-       }
-       else
-               *on = 1;
-       if (*cp++ != ' ' || *cp++ != '+')
-               goto bad_line;
-       if (parse_num(&cp, nb))
-               goto bad_line;
-       if (*cp == ',') {
-               cp++;
-               if (parse_num(&cp, nn))
-                       goto bad_line;
-       }
-       else
-               *nn = 1;
-       return -!!memcmp(cp, " @@", 3);
-}
-
-static void append_lost(struct sline *sline, int n, const char *line)
+static void append_lost(struct sline *sline, int n, const char *line, int len)
 {
        struct lline *lline;
-       int len = strlen(line);
        unsigned long this_mask = (1UL<<n);
        if (line[len-1] == '\n')
                len--;
@@ -216,70 +148,93 @@ static void append_lost(struct sline *sline, int n, const char *line)
        sline->lost_tail = &lline->next;
 }
 
-static void combine_diff(const unsigned char *parent, const char *ourtmp,
+struct combine_diff_state {
+       struct xdiff_emit_state xm;
+
+       unsigned int lno;
+       int ob, on, nb, nn;
+       unsigned long nmask;
+       int num_parent;
+       int n;
+       struct sline *sline;
+       struct sline *lost_bucket;
+};
+
+static void consume_line(void *state_, char *line, unsigned long len)
+{
+       struct combine_diff_state *state = state_;
+       if (5 < len && !memcmp("@@ -", line, 4)) {
+               if (parse_hunk_header(line, len,
+                                     &state->ob, &state->on,
+                                     &state->nb, &state->nn))
+                       return;
+               state->lno = state->nb;
+               if (!state->nb)
+                       /* @@ -1,2 +0,0 @@ to remove the
+                        * first two lines...
+                        */
+                       state->nb = 1;
+               if (state->nn == 0)
+                       /* @@ -X,Y +N,0 @@ removed Y lines
+                        * that would have come *after* line N
+                        * in the result.  Our lost buckets hang
+                        * to the line after the removed lines,
+                        */
+                       state->lost_bucket = &state->sline[state->nb];
+               else
+                       state->lost_bucket = &state->sline[state->nb-1];
+               if (!state->sline[state->nb-1].p_lno)
+                       state->sline[state->nb-1].p_lno =
+                               xcalloc(state->num_parent,
+                                       sizeof(unsigned long));
+               state->sline[state->nb-1].p_lno[state->n] = state->ob;
+               return;
+       }
+       if (!state->lost_bucket)
+               return; /* not in any hunk yet */
+       switch (line[0]) {
+       case '-':
+               append_lost(state->lost_bucket, state->n, line+1, len-1);
+               break;
+       case '+':
+               state->sline[state->lno-1].flag |= state->nmask;
+               state->lno++;
+               break;
+       }
+}
+
+static void combine_diff(const unsigned char *parent, mmfile_t *result_file,
                         struct sline *sline, int cnt, int n, int num_parent)
 {
-       FILE *in;
-       char parent_tmp[TMPPATHLEN];
-       char cmd[TMPPATHLEN * 2 + 1024];
-       char line[MAXLINELEN];
-       unsigned int lno, ob, on, nb, nn, p_lno;
+       unsigned int p_lno, lno;
        unsigned long nmask = (1UL << n);
-       struct sline *lost_bucket = NULL;
+       xpparam_t xpp;
+       xdemitconf_t xecfg;
+       mmfile_t parent_file;
+       xdemitcb_t ecb;
+       struct combine_diff_state state;
+       unsigned long sz;
 
        if (!cnt)
                return; /* result deleted */
 
-       write_temp_blob(parent_tmp, parent);
-       sprintf(cmd, "diff --unified=0 -La/x -Lb/x '%s' '%s'",
-               parent_tmp, ourtmp);
-       in = popen(cmd, "r");
-       if (!in)
-               die("cannot spawn %s", cmd);
-
-       lno = 1;
-       while (fgets(line, sizeof(line), in) != NULL) {
-               int len = strlen(line);
-               if (5 < len && !memcmp("@@ -", line, 4)) {
-                       if (parse_hunk_header(line, len,
-                                             &ob, &on, &nb, &nn))
-                               break;
-                       lno = nb;
-                       if (!nb)
-                               /* @@ -1,2 +0,0 @@ to remove the
-                                * first two lines...
-                                */
-                               nb = 1;
-                       if (nn == 0)
-                               /* @@ -X,Y +N,0 @@ removed Y lines
-                                * that would have come *after* line N
-                                * in the result.  Our lost buckets hang
-                                * to the line after the removed lines,
-                                */
-                               lost_bucket = &sline[nb];
-                       else
-                               lost_bucket = &sline[nb-1];
-                       if (!sline[nb-1].p_lno)
-                               sline[nb-1].p_lno =
-                                       xcalloc(num_parent,
-                                               sizeof(unsigned long));
-                       sline[nb-1].p_lno[n] = ob;
-                       continue;
-               }
-               if (!lost_bucket)
-                       continue; /* not in any hunk yet */
-               switch (line[0]) {
-               case '-':
-                       append_lost(lost_bucket, n, line+1);
-                       break;
-               case '+':
-                       sline[lno-1].flag |= nmask;
-                       lno++;
-                       break;
-               }
-       }
-       fclose(in);
-       unlink(parent_tmp);
+       parent_file.ptr = grab_blob(parent, &sz);
+       parent_file.size = sz;
+       xpp.flags = XDF_NEED_MINIMAL;
+       xecfg.ctxlen = 0;
+       xecfg.flags = 0;
+       ecb.outf = xdiff_outf;
+       ecb.priv = &state;
+       memset(&state, 0, sizeof(state));
+       state.xm.consume = consume_line;
+       state.nmask = nmask;
+       state.sline = sline;
+       state.lno = 1;
+       state.num_parent = num_parent;
+       state.n = n;
+
+       xdl_diff(&parent_file, result_file, &xpp, &xecfg, &ecb);
+       free(parent_file.ptr);
 
        /* Assign line numbers for this parent.
         *
@@ -625,61 +580,56 @@ static int show_patch_diff(struct combine_diff_path *elem, int num_parent,
                           int dense, const char *header,
                           struct diff_options *opt)
 {
-       unsigned long size, cnt, lno;
+       unsigned long result_size, cnt, lno;
        char *result, *cp, *ep;
        struct sline *sline; /* survived lines */
        int mode_differs = 0;
        int i, show_hunks, shown_header = 0;
-       char ourtmp_buf[TMPPATHLEN];
-       char *ourtmp = ourtmp_buf;
        int working_tree_file = !memcmp(elem->sha1, null_sha1, 20);
        int abbrev = opt->full_index ? 40 : DEFAULT_ABBREV;
+       mmfile_t result_file;
 
        /* Read the result of merge first */
-       if (!working_tree_file) {
-               result = grab_blob(elem->sha1, &size);
-               write_to_temp_file(ourtmp, result, size);
-       }
+       if (!working_tree_file)
+               result = grab_blob(elem->sha1, &result_size);
        else {
                /* Used by diff-tree to read from the working tree */
                struct stat st;
                int fd;
-               ourtmp = elem->path;
-               if (0 <= (fd = open(ourtmp, O_RDONLY)) &&
+               if (0 <= (fd = open(elem->path, O_RDONLY)) &&
                    !fstat(fd, &st)) {
                        int len = st.st_size;
                        int cnt = 0;
 
                        elem->mode = canon_mode(st.st_mode);
-                       size = len;
+                       result_size = len;
                        result = xmalloc(len + 1);
                        while (cnt < len) {
                                int done = xread(fd, result+cnt, len-cnt);
                                if (done == 0)
                                        break;
                                if (done < 0)
-                                       die("read error '%s'", ourtmp);
+                                       die("read error '%s'", elem->path);
                                cnt += done;
                        }
                        result[len] = 0;
                }
                else {
                        /* deleted file */
-                       size = 0;
+                       result_size = 0;
                        elem->mode = 0;
                        result = xmalloc(1);
                        result[0] = 0;
-                       ourtmp = "/dev/null";
                }
                if (0 <= fd)
                        close(fd);
        }
 
-       for (cnt = 0, cp = result; cp - result < size; cp++) {
+       for (cnt = 0, cp = result; cp - result < result_size; cp++) {
                if (*cp == '\n')
                        cnt++;
        }
-       if (size && result[size-1] != '\n')
+       if (result_size && result[result_size-1] != '\n')
                cnt++; /* incomplete line */
 
        sline = xcalloc(cnt+1, sizeof(*sline));
@@ -689,7 +639,7 @@ static int show_patch_diff(struct combine_diff_path *elem, int num_parent,
                sline[lno].lost_tail = &sline[lno].lost_head;
                sline[lno].flag = 0;
        }
-       for (lno = 0, cp = result; cp - result < size; cp++) {
+       for (lno = 0, cp = result; cp - result < result_size; cp++) {
                if (*cp == '\n') {
                        sline[lno].len = cp - sline[lno].bol;
                        lno++;
@@ -697,8 +647,11 @@ static int show_patch_diff(struct combine_diff_path *elem, int num_parent,
                                sline[lno].bol = cp + 1;
                }
        }
-       if (size && result[size-1] != '\n')
-               sline[cnt-1].len = size - (sline[cnt-1].bol - result);
+       if (result_size && result[result_size-1] != '\n')
+               sline[cnt-1].len = result_size - (sline[cnt-1].bol - result);
+
+       result_file.ptr = result;
+       result_file.size = result_size;
 
        sline[0].p_lno = xcalloc((cnt+1) * num_parent, sizeof(unsigned long));
        for (lno = 0; lno < cnt; lno++)
@@ -714,7 +667,7 @@ static int show_patch_diff(struct combine_diff_path *elem, int num_parent,
                        }
                }
                if (i <= j)
-                       combine_diff(elem->parent[i].sha1, ourtmp, sline,
+                       combine_diff(elem->parent[i].sha1, &result_file, sline,
                                     cnt, i, num_parent);
                if (elem->parent[i].mode != elem->mode)
                        mode_differs = 1;
@@ -767,8 +720,6 @@ static int show_patch_diff(struct combine_diff_path *elem, int num_parent,
                }
                dump_sline(sline, cnt, num_parent);
        }
-       if (ourtmp == ourtmp_buf)
-               unlink(ourtmp);
        free(result);
 
        for (i = 0; i < cnt; i++) {
index d4976fb..d534c9b 100644 (file)
--- a/commit.c
+++ b/commit.c
@@ -101,11 +101,7 @@ static unsigned long parse_commit_date(const char *buf)
        return date;
 }
 
-static struct commit_graft {
-       unsigned char sha1[20];
-       int nr_parent;
-       unsigned char parent[0][20]; /* more */
-} **commit_graft;
+static struct commit_graft **commit_graft;
 static int commit_graft_alloc, commit_graft_nr;
 
 static int commit_graft_pos(const unsigned char *sha1)
@@ -127,70 +123,98 @@ static int commit_graft_pos(const unsigned char *sha1)
        return -lo - 1;
 }
 
-static void prepare_commit_graft(void)
+int register_commit_graft(struct commit_graft *graft, int ignore_dups)
+{
+       int pos = commit_graft_pos(graft->sha1);
+       
+       if (0 <= pos) {
+               if (ignore_dups)
+                       free(graft);
+               else {
+                       free(commit_graft[pos]);
+                       commit_graft[pos] = graft;
+               }
+               return 1;
+       }
+       pos = -pos - 1;
+       if (commit_graft_alloc <= ++commit_graft_nr) {
+               commit_graft_alloc = alloc_nr(commit_graft_alloc);
+               commit_graft = xrealloc(commit_graft,
+                                       sizeof(*commit_graft) *
+                                       commit_graft_alloc);
+       }
+       if (pos < commit_graft_nr)
+               memmove(commit_graft + pos + 1,
+                       commit_graft + pos,
+                       (commit_graft_nr - pos - 1) *
+                       sizeof(*commit_graft));
+       commit_graft[pos] = graft;
+       return 0;
+}
+
+struct commit_graft *read_graft_line(char *buf, int len)
+{
+       /* The format is just "Commit Parent1 Parent2 ...\n" */
+       int i;
+       struct commit_graft *graft = NULL;
+
+       if (buf[len-1] == '\n')
+               buf[--len] = 0;
+       if (buf[0] == '#')
+               return 0;
+       if ((len + 1) % 41) {
+       bad_graft_data:
+               error("bad graft data: %s", buf);
+               free(graft);
+               return NULL;
+       }
+       i = (len + 1) / 41 - 1;
+       graft = xmalloc(sizeof(*graft) + 20 * i);
+       graft->nr_parent = i;
+       if (get_sha1_hex(buf, graft->sha1))
+               goto bad_graft_data;
+       for (i = 40; i < len; i += 41) {
+               if (buf[i] != ' ')
+                       goto bad_graft_data;
+               if (get_sha1_hex(buf + i + 1, graft->parent[i/41]))
+                       goto bad_graft_data;
+       }
+       return graft;
+}
+
+int read_graft_file(const char *graft_file)
 {
-       char *graft_file = get_graft_file();
        FILE *fp = fopen(graft_file, "r");
        char buf[1024];
-       if (!fp) {
-               commit_graft = (struct commit_graft **) "hack";
-               return;
-       }
+       if (!fp)
+               return -1;
        while (fgets(buf, sizeof(buf), fp)) {
                /* The format is just "Commit Parent1 Parent2 ...\n" */
                int len = strlen(buf);
-               int i;
-               struct commit_graft *graft = NULL;
-
-               if (buf[len-1] == '\n')
-                       buf[--len] = 0;
-               if (buf[0] == '#')
-                       continue;
-               if ((len + 1) % 41) {
-               bad_graft_data:
-                       error("bad graft data: %s", buf);
-                       free(graft);
-                       continue;
-               }
-               i = (len + 1) / 41 - 1;
-               graft = xmalloc(sizeof(*graft) + 20 * i);
-               graft->nr_parent = i;
-               if (get_sha1_hex(buf, graft->sha1))
-                       goto bad_graft_data;
-               for (i = 40; i < len; i += 41) {
-                       if (buf[i] != ' ')
-                               goto bad_graft_data;
-                       if (get_sha1_hex(buf + i + 1, graft->parent[i/41]))
-                               goto bad_graft_data;
-               }
-               i = commit_graft_pos(graft->sha1);
-               if (0 <= i) {
+               struct commit_graft *graft = read_graft_line(buf, len);
+               if (register_commit_graft(graft, 1))
                        error("duplicate graft data: %s", buf);
-                       free(graft);
-                       continue;
-               }
-               i = -i - 1;
-               if (commit_graft_alloc <= ++commit_graft_nr) {
-                       commit_graft_alloc = alloc_nr(commit_graft_alloc);
-                       commit_graft = xrealloc(commit_graft,
-                                               sizeof(*commit_graft) *
-                                               commit_graft_alloc);
-               }
-               if (i < commit_graft_nr)
-                       memmove(commit_graft + i + 1,
-                               commit_graft + i,
-                               (commit_graft_nr - i - 1) *
-                               sizeof(*commit_graft));
-               commit_graft[i] = graft;
        }
        fclose(fp);
+       return 0;
+}
+
+static void prepare_commit_graft(void)
+{
+       static int commit_graft_prepared;
+       char *graft_file;
+
+       if (commit_graft_prepared)
+               return;
+       graft_file = get_graft_file();
+       read_graft_file(graft_file);
+       commit_graft_prepared = 1;
 }
 
 static struct commit_graft *lookup_commit_graft(const unsigned char *sha1)
 {
        int pos;
-       if (!commit_graft)
-               prepare_commit_graft();
+       prepare_commit_graft();
        pos = commit_graft_pos(sha1);
        if (pos < 0)
                return NULL;
index 98682b2..918c9ab 100644 (file)
--- a/commit.h
+++ b/commit.h
@@ -90,4 +90,15 @@ void sort_in_topological_order(struct commit_list ** list, int lifo);
 void sort_in_topological_order_fn(struct commit_list ** list, int lifo,
                                  topo_sort_set_fn_t setter,
                                  topo_sort_get_fn_t getter);
+
+struct commit_graft {
+       unsigned char sha1[20];
+       int nr_parent;
+       unsigned char parent[FLEX_ARRAY][20]; /* more */
+};
+
+struct commit_graft *read_graft_line(char *buf, int len);
+int register_commit_graft(struct commit_graft *, int);
+int read_graft_file(const char *graft_file);
+
 #endif /* COMMIT_H */
diff --git a/date.c b/date.c
index 1c1917b..376d25d 100644 (file)
--- a/date.c
+++ b/date.c
@@ -255,10 +255,10 @@ static int match_multi_number(unsigned long num, char c, const char *date, char
                                break;
                }
                /* mm/dd/yy ? */
-               if (is_date(num3, num2, num, tm))
+               if (is_date(num3, num, num2, tm))
                        break;
                /* dd/mm/yy ? */
-               if (is_date(num3, num, num2, tm))
+               if (is_date(num3, num2, num, tm))
                        break;
                return 0;
        }
diff --git a/delta.h b/delta.h
index a15350d..9464f3e 100644 (file)
--- a/delta.h
+++ b/delta.h
@@ -16,7 +16,8 @@ extern void *patch_delta(void *src_buf, unsigned long src_size,
  * This must be called twice on the delta data buffer, first to get the
  * expected reference buffer size, and again to get the result buffer size.
  */
-static inline unsigned long get_delta_hdr_size(const unsigned char **datap)
+static inline unsigned long get_delta_hdr_size(const unsigned char **datap,
+                                              const unsigned char *top)
 {
        const unsigned char *data = *datap;
        unsigned char cmd;
@@ -26,7 +27,7 @@ static inline unsigned long get_delta_hdr_size(const unsigned char **datap)
                cmd = *data++;
                size |= (cmd & ~0x80) << i;
                i += 7;
-       } while (cmd & 0x80);
+       } while (cmd & 0x80 && data < top);
        *datap = data;
        return size;
 }
diff --git a/diff.c b/diff.c
index e496905..ce98a90 100644 (file)
--- a/diff.c
+++ b/diff.c
@@ -883,6 +883,8 @@ int diff_opt_parse(struct diff_options *options, const char **av, int ac)
                options->filter = arg + 14;
        else if (!strcmp(arg, "--pickaxe-all"))
                options->pickaxe_opts = DIFF_PICKAXE_ALL;
+       else if (!strcmp(arg, "--pickaxe-regex"))
+               options->pickaxe_opts = DIFF_PICKAXE_REGEX;
        else if (!strncmp(arg, "-B", 2)) {
                if ((options->break_opt =
                     diff_scoreopt_parse(arg)) == -1)
@@ -1280,28 +1282,34 @@ void diff_flush(struct diff_options *options)
 
        for (i = 0; i < q->nr; i++) {
                struct diff_filepair *p = q->queue[i];
-               if ((diff_output_format == DIFF_FORMAT_NO_OUTPUT) ||
-                   (p->status == DIFF_STATUS_UNKNOWN))
-                       continue;
-               if (p->status == 0)
-                       die("internal error in diff-resolve-rename-copy");
-               switch (diff_output_format) {
-               case DIFF_FORMAT_PATCH:
-                       diff_flush_patch(p, options);
-                       break;
-               case DIFF_FORMAT_RAW:
-               case DIFF_FORMAT_NAME_STATUS:
-                       diff_flush_raw(p, line_termination,
-                                      inter_name_termination,
-                                      options);
+
+               switch (p->status) {
+               case DIFF_STATUS_UNKNOWN:
                        break;
-               case DIFF_FORMAT_NAME:
-                       diff_flush_name(p,
-                                       inter_name_termination,
-                                       line_termination);
+               case 0:
+                       die("internal error in diff-resolve-rename-copy");
                        break;
+               default:
+                       switch (diff_output_format) {
+                       case DIFF_FORMAT_PATCH:
+                               diff_flush_patch(p, options);
+                               break;
+                       case DIFF_FORMAT_RAW:
+                       case DIFF_FORMAT_NAME_STATUS:
+                               diff_flush_raw(p, line_termination,
+                                              inter_name_termination,
+                                              options);
+                               break;
+                       case DIFF_FORMAT_NAME:
+                               diff_flush_name(p,
+                                               inter_name_termination,
+                                               line_termination);
+                               break;
+                       case DIFF_FORMAT_NO_OUTPUT:
+                               break;
+                       }
                }
-               diff_free_filepair(q->queue[i]);
+               diff_free_filepair(p);
        }
        free(q->queue);
        q->queue = NULL;
diff --git a/diff.h b/diff.h
index a268d16..0cebec1 100644 (file)
--- a/diff.h
+++ b/diff.h
@@ -102,6 +102,7 @@ extern int diff_setup_done(struct diff_options *);
 #define DIFF_DETECT_COPY       2
 
 #define DIFF_PICKAXE_ALL       1
+#define DIFF_PICKAXE_REGEX     2
 
 extern void diffcore_std(struct diff_options *);
 
index 50e46ab..cfcce31 100644 (file)
@@ -5,8 +5,11 @@
 #include "diff.h"
 #include "diffcore.h"
 
+#include <regex.h>
+
 static unsigned int contains(struct diff_filespec *one,
-                            const char *needle, unsigned long len)
+                            const char *needle, unsigned long len,
+                            regex_t *regexp)
 {
        unsigned int cnt;
        unsigned long offset, sz;
@@ -18,15 +21,28 @@ static unsigned int contains(struct diff_filespec *one,
        data = one->data;
        cnt = 0;
 
-       /* Yes, I've heard of strstr(), but the thing is *data may
-        * not be NUL terminated.  Sue me.
-        */
-       for (offset = 0; offset + len <= sz; offset++) {
-               /* we count non-overlapping occurrences of needle */
-               if (!memcmp(needle, data + offset, len)) {
-                       offset += len - 1;
+       if (regexp) {
+               regmatch_t regmatch;
+               int flags = 0;
+
+               while (*data && !regexec(regexp, data, 1, &regmatch, flags)) {
+                       flags |= REG_NOTBOL;
+                       data += regmatch.rm_so;
+                       if (*data) data++;
                        cnt++;
                }
+
+       } else { /* Classic exact string match */
+               /* Yes, I've heard of strstr(), but the thing is *data may
+                * not be NUL terminated.  Sue me.
+                */
+               for (offset = 0; offset + len <= sz; offset++) {
+                       /* we count non-overlapping occurrences of needle */
+                       if (!memcmp(needle, data + offset, len)) {
+                               offset += len - 1;
+                               cnt++;
+                       }
+               }
        }
        return cnt;
 }
@@ -36,10 +52,24 @@ void diffcore_pickaxe(const char *needle, int opts)
        struct diff_queue_struct *q = &diff_queued_diff;
        unsigned long len = strlen(needle);
        int i, has_changes;
+       regex_t regex, *regexp = NULL;
        struct diff_queue_struct outq;
        outq.queue = NULL;
        outq.nr = outq.alloc = 0;
 
+       if (opts & DIFF_PICKAXE_REGEX) {
+               int err;
+               err = regcomp(&regex, needle, REG_EXTENDED | REG_NEWLINE);
+               if (err) {
+                       /* The POSIX.2 people are surely sick */
+                       char errbuf[1024];
+                       regerror(err, &regex, errbuf, 1024);
+                       regfree(&regex);
+                       die("invalid pickaxe regex: %s", errbuf);
+               }
+               regexp = &regex;
+       }
+
        if (opts & DIFF_PICKAXE_ALL) {
                /* Showing the whole changeset if needle exists */
                for (i = has_changes = 0; !has_changes && i < q->nr; i++) {
@@ -48,16 +78,16 @@ void diffcore_pickaxe(const char *needle, int opts)
                                if (!DIFF_FILE_VALID(p->two))
                                        continue; /* ignore unmerged */
                                /* created */
-                               if (contains(p->two, needle, len))
+                               if (contains(p->two, needle, len, regexp))
                                        has_changes++;
                        }
                        else if (!DIFF_FILE_VALID(p->two)) {
-                               if (contains(p->one, needle, len))
+                               if (contains(p->one, needle, len, regexp))
                                        has_changes++;
                        }
                        else if (!diff_unmodified_pair(p) &&
-                                contains(p->one, needle, len) !=
-                                contains(p->two, needle, len))
+                                contains(p->one, needle, len, regexp) !=
+                                contains(p->two, needle, len, regexp))
                                has_changes++;
                }
                if (has_changes)
@@ -80,16 +110,16 @@ void diffcore_pickaxe(const char *needle, int opts)
                                if (!DIFF_FILE_VALID(p->two))
                                        ; /* ignore unmerged */
                                /* created */
-                               else if (contains(p->two, needle, len))
+                               else if (contains(p->two, needle, len, regexp))
                                        has_changes = 1;
                        }
                        else if (!DIFF_FILE_VALID(p->two)) {
-                               if (contains(p->one, needle, len))
+                               if (contains(p->one, needle, len, regexp))
                                        has_changes = 1;
                        }
                        else if (!diff_unmodified_pair(p) &&
-                                contains(p->one, needle, len) !=
-                                contains(p->two, needle, len))
+                                contains(p->one, needle, len, regexp) !=
+                                contains(p->two, needle, len, regexp))
                                has_changes = 1;
 
                        if (has_changes)
@@ -98,6 +128,10 @@ void diffcore_pickaxe(const char *needle, int opts)
                                diff_free_filepair(p);
                }
 
+       if (opts & DIFF_PICKAXE_REGEX) {
+               regfree(&regex);
+       }
+
        free(q->queue);
        *q = outq;
        return;
diff --git a/git-clean.sh b/git-clean.sh
new file mode 100755 (executable)
index 0000000..b200868
--- /dev/null
@@ -0,0 +1,80 @@
+#!/bin/sh
+#
+# Copyright (c) 2005-2006 Pavel Roskin
+#
+
+USAGE="[-d] [-n] [-q] [-x | -X]"
+LONG_USAGE='Clean untracked files from the working directory
+       -d      remove directories as well
+       -n      don'\''t remove anything, just show what would be done
+       -q      be quiet, only report errors
+       -x      remove ignored files as well
+       -X      remove only ignored files as well'
+SUBDIRECTORY_OK=Yes
+. git-sh-setup
+
+ignored=
+ignoredonly=
+cleandir=
+quiet=
+rmf="rm -f"
+rmrf="rm -rf"
+rm_refuse="echo Not removing"
+echo1="echo"
+
+while case "$#" in 0) break ;; esac
+do
+       case "$1" in
+       -d)
+               cleandir=1
+               ;;
+       -n)
+               quiet=1
+               rmf="echo Would remove"
+               rmrf="echo Would remove"
+               rm_refuse="echo Would not remove"
+               echo1=":"
+               ;;
+       -q)
+               quiet=1
+               ;;
+       -x)
+               ignored=1
+               ;;
+       -X)
+               ignoredonly=1
+               ;;
+       *)
+               usage
+       esac
+       shift
+done
+
+case "$ignored,$ignoredonly" in
+       1,1) usage;;
+esac
+
+if [ -z "$ignored" ]; then
+       excl="--exclude-per-directory=.gitignore"
+       if [ -f "$GIT_DIR/info/exclude" ]; then
+               excl_info="--exclude-from=$GIT_DIR/info/exclude"
+       fi
+       if [ "$ignoredonly" ]; then
+               excl="$excl --ignored"
+       fi
+fi
+
+git-ls-files --others --directory $excl ${excl_info:+"$excl_info"} |
+while read -r file; do
+       if [ -d "$file" -a ! -L "$file" ]; then
+               if [ -z "$cleandir" ]; then
+                       $rm_refuse "$file"
+                       continue
+               fi
+               $echo1 "Removing $file"
+               $rmrf "$file"
+       else
+               $echo1 "Removing $file"
+               $rmf "$file"
+       fi
+done
index 823c74b..c013e48 100755 (executable)
@@ -52,7 +52,8 @@ Perhaps git-update-server-info needs to be run there?"
                git-http-fetch -v -a -w "$tname" "$name" "$1/" || exit 1
        done <"$clone_tmp/refs"
        rm -fr "$clone_tmp"
-       http_fetch "$1/HEAD" "$GIT_DIR/REMOTE_HEAD"
+       http_fetch "$1/HEAD" "$GIT_DIR/REMOTE_HEAD" ||
+       rm -f "$GIT_DIR/REMOTE_HEAD"
 }
 
 # Read git-fetch-pack -k output and store the remote branches.
@@ -324,7 +325,7 @@ test -d "$GIT_DIR/refs/reference-tmp" && rm -fr "$GIT_DIR/refs/reference-tmp"
 
 if test -f "$GIT_DIR/CLONE_HEAD"
 then
-       # Figure out where the remote HEAD points at.
+       # Read git-fetch-pack -k output and store the remote branches.
        perl -e "$copy_refs" "$GIT_DIR" "$use_separate_remote" "$origin"
 fi
 
@@ -332,22 +333,25 @@ cd "$D" || exit
 
 if test -z "$bare" && test -f "$GIT_DIR/REMOTE_HEAD"
 then
-       head_sha1=`cat "$GIT_DIR/REMOTE_HEAD"`
        # Figure out which remote branch HEAD points at.
        case "$use_separate_remote" in
        '')     remote_top=refs/heads ;;
        *)      remote_top="refs/remotes/$origin" ;;
        esac
 
-       # What to use to track the remote primary branch
-       if test -n "$use_separate_remote"
-       then
-               origin_tracking="remotes/$origin/master"
-       else
-               origin_tracking="heads/$origin"
-       fi
+       head_sha1=`cat "$GIT_DIR/REMOTE_HEAD"`
+       case "$head_sha1" in
+       'ref: refs/'*)
+               # Uh-oh, the remote told us (http transport done against
+               # new style repository with a symref HEAD).
+               # Ideally we should skip the guesswork but for now
+               # opt for minimum change.
+               head_sha1=`expr "$head_sha1" : 'ref: refs/heads/\(.*\)'`
+               head_sha1=`cat "$GIT_DIR/$remote_top/$head_sha1"`
+               ;;
+       esac
 
-       # The name under $remote_top the remote HEAD seems to point at
+       # The name under $remote_top the remote HEAD seems to point at.
        head_points_at=$(
                (
                        echo "master"
@@ -368,23 +372,28 @@ then
                )
        )
 
-       # Write out remotes/$origin file.
+       # Write out remotes/$origin file, and update our "$head_points_at".
        case "$head_points_at" in
        ?*)
                mkdir -p "$GIT_DIR/remotes" &&
-               echo >"$GIT_DIR/remotes/$origin" \
-               "URL: $repo
-Pull: refs/heads/$head_points_at:refs/$origin_tracking" &&
+               git-symbolic-ref HEAD "refs/heads/$head_points_at" &&
                case "$use_separate_remote" in
-               t) git-update-ref HEAD "$head_sha1" ;;
-               *) git-update-ref "refs/heads/$origin" $(git-rev-parse HEAD) ;;
+               t)      origin_track="$remote_top/$head_points_at"
+                       git-update-ref HEAD "$head_sha1" ;;
+               *)      origin_track="$remote_top/$origin"
+                       git-update-ref "refs/heads/$origin" "$head_sha1" ;;
                esac &&
+               echo >"$GIT_DIR/remotes/$origin" \
+               "URL: $repo
+Pull: refs/heads/$head_points_at:$origin_track" &&
                (cd "$GIT_DIR/$remote_top" && find . -type f -print) |
                while read dotslref
                do
                        name=`expr "$dotslref" : './\(.*\)'` &&
-                       test "$head_points_at" = "$name" ||
-                       test "$origin" = "$name" ||
+                       test "$use_separate_remote" = '' && {
+                               test "$head_points_at" = "$name" ||
+                               test "$origin" = "$name"
+                       } ||
                        echo "Pull: refs/heads/${name}:$remote_top/${name}"
                done >>"$GIT_DIR/remotes/$origin" &&
                case "$use_separate_remote" in
diff --git a/gitk b/gitk
index fa1e83c..f88c06e 100755 (executable)
--- a/gitk
+++ b/gitk
@@ -75,6 +75,7 @@ proc getcommitlines {commfd}  {
     global commitlisted nextupdate
     global leftover
     global displayorder commitidx commitrow commitdata
+    global parentlist childlist children
 
     set stuff [read $commfd]
     if {$stuff == {}} {
@@ -140,15 +141,26 @@ proc getcommitlines {commfd}  {
        set id [lindex $ids 0]
        if {$listed} {
            set olds [lrange $ids 1 end]
-           set commitlisted($id) 1
+           if {[llength $olds] > 1} {
+               set olds [lsort -unique $olds]
+           }
+           foreach p $olds {
+               lappend children($p) $id
+           }
        } else {
            set olds {}
        }
-       updatechildren $id $olds
+       lappend parentlist $olds
+       if {[info exists children($id)]} {
+           lappend childlist $children($id)
+       } else {
+           lappend childlist {}
+       }
        set commitdata($id) [string range $cmit [expr {$j + 1}] end]
        set commitrow($id) $commitidx
        incr commitidx
        lappend displayorder $id
+       lappend commitlisted $listed
        set gotsome 1
     }
     if {$gotsome} {
@@ -181,14 +193,12 @@ proc doupdate {reading} {
 
 proc readcommit {id} {
     if {[catch {set contents [exec git-cat-file commit $id]}]} return
-    updatechildren $id {}
     parsecommit $id $contents 0
 }
 
 proc updatecommits {rargs} {
     stopfindproc
-    foreach v {children nchildren parents nparents commitlisted
-       colormap selectedline matchinglines treediffs
+    foreach v {colormap selectedline matchinglines treediffs
        mergefilelist currentid rowtextx commitrow
        rowidlist rowoffsets idrowranges idrangedrawn iddrawn
        linesegends crossings cornercrossings} {
@@ -200,26 +210,6 @@ proc updatecommits {rargs} {
     getcommits $rargs
 }
 
-proc updatechildren {id olds} {
-    global children nchildren parents nparents
-
-    if {![info exists nchildren($id)]} {
-       set children($id) {}
-       set nchildren($id) 0
-    }
-    set parents($id) $olds
-    set nparents($id) [llength $olds]
-    foreach p $olds {
-       if {![info exists nchildren($p)]} {
-           set children($p) [list $id]
-           set nchildren($p) 1
-       } elseif {[lsearch -exact $children($p) $id] < 0} {
-           lappend children($p) $id
-           incr nchildren($p)
-       }
-    }
-}
-
 proc parsecommit {id contents listed} {
     global commitinfo cdate
 
@@ -274,7 +264,7 @@ proc parsecommit {id contents listed} {
 }
 
 proc getcommit {id} {
-    global commitdata commitinfo nparents
+    global commitdata commitinfo
 
     if {[info exists commitdata($id)]} {
        parsecommit $id $commitdata($id) 1
@@ -282,7 +272,6 @@ proc getcommit {id} {
        readcommit $id
        if {![info exists commitinfo($id)]} {
            set commitinfo($id) {"No commit information available"}
-           set nparents($id) 0
        }
     }
     return 1
@@ -295,7 +284,7 @@ proc readrefs {} {
     foreach v {tagids idtags headids idheads otherrefids idotherrefs} {
        catch {unset $v}
     }
-    set refd [open [list | git-ls-remote [gitdir]] r]
+    set refd [open [list | git ls-remote [gitdir]] r]
     while {0 <= [set n [gets $refd line]]} {
        if {![regexp {^([0-9a-f]{40})   refs/([^^]*)$} $line \
            match id path]} {
@@ -346,7 +335,7 @@ proc error_popup msg {
 }
 
 proc makewindow {rargs} {
-    global canv canv2 canv3 linespc charspc ctext cflist textfont
+    global canv canv2 canv3 linespc charspc ctext cflist textfont mainfont uifont
     global findtype findtypemenu findloc findstring fstring geometry
     global entries sha1entry sha1string sha1but
     global maincursor textcursor curtextcursor
@@ -354,16 +343,21 @@ proc makewindow {rargs} {
 
     menu .bar
     .bar add cascade -label "File" -menu .bar.file
+    .bar configure -font $uifont
     menu .bar.file
     .bar.file add command -label "Update" -command [list updatecommits $rargs]
     .bar.file add command -label "Reread references" -command rereadrefs
     .bar.file add command -label "Quit" -command doquit
+    .bar.file configure -font $uifont
     menu .bar.edit
     .bar add cascade -label "Edit" -menu .bar.edit
     .bar.edit add command -label "Preferences" -command doprefs
+    .bar.edit configure -font $uifont
     menu .bar.help
     .bar add cascade -label "Help" -menu .bar.help
     .bar.help add command -label "About gitk" -command about
+    .bar.help add command -label "Key bindings" -command keys
+    .bar.help configure -font $uifont
     . configure -menu .bar
 
     if {![info exists geometry(canv1)]} {
@@ -410,7 +404,7 @@ proc makewindow {rargs} {
     set entries $sha1entry
     set sha1but .ctop.top.bar.sha1label
     button $sha1but -text "SHA1 ID: " -state disabled -relief flat \
-       -command gotocommit -width 8
+       -command gotocommit -width 8 -font $uifont
     $sha1but conf -disabledforeground [$sha1but cget -foreground]
     pack .ctop.top.bar.sha1label -side left
     entry $sha1entry -width 40 -font $textfont -textvariable sha1string
@@ -440,19 +434,24 @@ proc makewindow {rargs} {
        -state disabled -width 26
     pack .ctop.top.bar.rightbut -side left -fill y
 
-    button .ctop.top.bar.findbut -text "Find" -command dofind
+    button .ctop.top.bar.findbut -text "Find" -command dofind -font $uifont
     pack .ctop.top.bar.findbut -side left
     set findstring {}
     set fstring .ctop.top.bar.findstring
     lappend entries $fstring
-    entry $fstring -width 30 -font $textfont -textvariable findstring
+    entry $fstring -width 30 -font $textfont -textvariable findstring -font $textfont
     pack $fstring -side left -expand 1 -fill x
     set findtype Exact
     set findtypemenu [tk_optionMenu .ctop.top.bar.findtype \
                          findtype Exact IgnCase Regexp]
+    .ctop.top.bar.findtype configure -font $uifont
+    .ctop.top.bar.findtype.menu configure -font $uifont
     set findloc "All fields"
     tk_optionMenu .ctop.top.bar.findloc findloc "All fields" Headline \
        Comments Author Committer Files Pickaxe
+    .ctop.top.bar.findloc configure -font $uifont
+    .ctop.top.bar.findloc.menu configure -font $uifont
+
     pack .ctop.top.bar.findloc -side right
     pack .ctop.top.bar.findtype -side right
     # for making sure type==Exact whenever loc==Pickaxe
@@ -499,7 +498,7 @@ proc makewindow {rargs} {
     frame .ctop.cdet.right
     set cflist .ctop.cdet.right.cfiles
     listbox $cflist -bg white -selectmode extended -width $geometry(cflistw) \
-       -yscrollcommand ".ctop.cdet.right.sb set"
+       -yscrollcommand ".ctop.cdet.right.sb set" -font $mainfont
     scrollbar .ctop.cdet.right.sb -command "$cflist yview"
     pack .ctop.cdet.right.sb -side right -fill y
     pack $cflist -side left -fill both -expand 1
@@ -514,12 +513,20 @@ proc makewindow {rargs} {
     bindall <ButtonRelease-5> "allcanvs yview scroll 5 units"
     bindall <2> "canvscan mark %W %x %y"
     bindall <B2-Motion> "canvscan dragto %W %x %y"
+    bindkey <Home> selfirstline
+    bindkey <End> sellastline
     bind . <Key-Up> "selnextline -1"
     bind . <Key-Down> "selnextline 1"
-    bind . <Key-Right> "goforw"
-    bind . <Key-Left> "goback"
-    bind . <Key-Prior> "allcanvs yview scroll -1 pages"
-    bind . <Key-Next> "allcanvs yview scroll 1 pages"
+    bindkey <Key-Right> "goforw"
+    bindkey <Key-Left> "goback"
+    bind . <Key-Prior> "selnextpage -1"
+    bind . <Key-Next> "selnextpage 1"
+    bind . <Control-Home> "allcanvs yview moveto 0.0"
+    bind . <Control-End> "allcanvs yview moveto 1.0"
+    bind . <Control-Key-Up> "allcanvs yview scroll -1 units"
+    bind . <Control-Key-Down> "allcanvs yview scroll 1 units"
+    bind . <Control-Key-Prior> "allcanvs yview scroll -1 pages"
+    bind . <Control-Key-Next> "allcanvs yview scroll 1 pages"
     bindkey <Key-Delete> "$ctext yview scroll -1 pages"
     bindkey <Key-BackSpace> "$ctext yview scroll -1 pages"
     bindkey <Key-space> "$ctext yview scroll 1 pages"
@@ -612,7 +619,7 @@ proc click {w} {
 }
 
 proc savestuff {w} {
-    global canv canv2 canv3 ctext cflist mainfont textfont
+    global canv canv2 canv3 ctext cflist mainfont textfont uifont
     global stuffsaved findmergefiles maxgraphpct
     global maxwidth
 
@@ -622,6 +629,7 @@ proc savestuff {w} {
        set f [open "~/.gitk-new" w]
        puts $f [list set mainfont $mainfont]
        puts $f [list set textfont $textfont]
+       puts $f [list set uifont $uifont]
        puts $f [list set findmergefiles $findmergefiles]
        puts $f [list set maxgraphpct $maxgraphpct]
        puts $f [list set maxwidth $maxwidth]
@@ -729,6 +737,55 @@ Use and redistribute under the terms of the GNU General Public License} \
     pack $w.ok -side bottom
 }
 
+proc keys {} {
+    set w .keys
+    if {[winfo exists $w]} {
+       raise $w
+       return
+    }
+    toplevel $w
+    wm title $w "Gitk key bindings"
+    message $w.m -text {
+Gitk key bindings:
+
+<Ctrl-Q>               Quit
+<Home>         Move to first commit
+<End>          Move to last commit
+<Up>, p, i     Move up one commit
+<Down>, n, k   Move down one commit
+<Left>, z, j   Go back in history list
+<Right>, x, l  Go forward in history list
+<PageUp>       Move up one page in commit list
+<PageDown>     Move down one page in commit list
+<Ctrl-Home>    Scroll to top of commit list
+<Ctrl-End>     Scroll to bottom of commit list
+<Ctrl-Up>      Scroll commit list up one line
+<Ctrl-Down>    Scroll commit list down one line
+<Ctrl-PageUp>  Scroll commit list up one page
+<Ctrl-PageDown>        Scroll commit list down one page
+<Delete>, b    Scroll diff view up one page
+<Backspace>    Scroll diff view up one page
+<Space>                Scroll diff view down one page
+u              Scroll diff view up 18 lines
+d              Scroll diff view down 18 lines
+<Ctrl-F>               Find
+<Ctrl-G>               Move to next find hit
+<Ctrl-R>               Move to previous find hit
+<Return>       Move to next find hit
+/              Move to next find hit, or redo find
+?              Move to previous find hit
+f              Scroll diff view to next file
+<Ctrl-KP+>     Increase font size
+<Ctrl-plus>    Increase font size
+<Ctrl-KP->     Decrease font size
+<Ctrl-minus>   Decrease font size
+} \
+           -justify left -bg white -border 2 -relief sunken
+    pack $w.m -side top -fill both
+    button $w.ok -text Close -command "destroy $w"
+    pack $w.ok -side bottom
+}
+
 proc shortids {ids} {
     set res {}
     foreach id $ids {
@@ -843,15 +900,20 @@ proc makeuparrow {oid x y z} {
 }
 
 proc initlayout {} {
-    global rowidlist rowoffsets displayorder
+    global rowidlist rowoffsets displayorder commitlisted
     global rowlaidout rowoptim
     global idinlist rowchk
     global commitidx numcommits canvxmax canv
     global nextcolor
+    global parentlist childlist children
 
     set commitidx 0
     set numcommits 0
     set displayorder {}
+    set commitlisted {}
+    set parentlist {}
+    set childlist {}
+    catch {unset children}
     set nextcolor 0
     set rowidlist {{}}
     set rowoffsets {{}}
@@ -950,7 +1012,7 @@ proc showstuff {canshow} {
 proc layoutrows {row endrow last} {
     global rowidlist rowoffsets displayorder
     global uparrowlen downarrowlen maxwidth mingaplen
-    global nchildren parents nparents
+    global childlist parentlist
     global idrowranges linesegends
     global commitidx
     global idinlist rowchk
@@ -961,7 +1023,7 @@ proc layoutrows {row endrow last} {
        set id [lindex $displayorder $row]
        set oldolds {}
        set newolds {}
-       foreach p $parents($id) {
+       foreach p [lindex $parentlist $row] {
            if {![info exists idinlist($p)]} {
                lappend newolds $p
            } elseif {!$idinlist($p)} {
@@ -1000,7 +1062,7 @@ proc layoutrows {row endrow last} {
            lappend idlist $id
            lset rowidlist $row $idlist
            set z {}
-           if {$nchildren($id) > 0} {
+           if {[lindex $childlist $row] ne {}} {
                set z [expr {[llength [lindex $rowidlist [expr {$row-1}]]] - $col}]
                unset idinlist($id)
            }
@@ -1053,16 +1115,22 @@ proc layoutrows {row endrow last} {
 }
 
 proc addextraid {id row} {
-    global displayorder commitrow commitinfo nparents
+    global displayorder commitrow commitinfo
     global commitidx
+    global parentlist childlist children
 
     incr commitidx
     lappend displayorder $id
+    lappend parentlist {}
     set commitrow($id) $row
     readcommit $id
     if {![info exists commitinfo($id)]} {
        set commitinfo($id) {"No commit information available"}
-       set nparents($id) 0
+    }
+    if {[info exists children($id)]} {
+       lappend childlist $children($id)
+    } else {
+       lappend childlist {}
     }
 }
 
@@ -1365,7 +1433,7 @@ proc drawparentlinks {id row col olds} {
 proc drawlines {id} {
     global colormap canv
     global idrowranges idrangedrawn
-    global children iddrawn commitrow rowidlist
+    global childlist iddrawn commitrow rowidlist
 
     $canv delete lines.$id
     set nr [expr {[llength $idrowranges($id)] / 2}]
@@ -1374,14 +1442,12 @@ proc drawlines {id} {
            drawlineseg $id $i
        }
     }
-    if {[info exists children($id)]} {
-       foreach child $children($id) {
-           if {[info exists iddrawn($child)]} {
-               set row $commitrow($child)
-               set col [lsearch -exact [lindex $rowidlist $row] $child]
-               if {$col >= 0} {
-                   drawparentlinks $child $row $col [list $id]
-               }
+    foreach child [lindex $childlist $commitrow($id)] {
+       if {[info exists iddrawn($child)]} {
+           set row $commitrow($child)
+           set col [lsearch -exact [lindex $rowidlist $row] $child]
+           if {$col >= 0} {
+               drawparentlinks $child $row $col [list $id]
            }
        }
     }
@@ -1394,7 +1460,7 @@ proc drawcmittext {id row col rmx} {
     global linehtag linentag linedtag
     global mainfont namefont canvxmax
 
-    set ofill [expr {[info exists commitlisted($id)]? "blue": "white"}]
+    set ofill [expr {[lindex $commitlisted $row]? "blue": "white"}]
     set x [xc $row $col]
     set y [yc $row]
     set orad [expr {$linespc / 3}]
@@ -1434,7 +1500,7 @@ proc drawcmittext {id row col rmx} {
 proc drawcmitrow {row} {
     global displayorder rowidlist
     global idrowranges idrangedrawn iddrawn
-    global commitinfo commitlisted parents numcommits
+    global commitinfo commitlisted parentlist numcommits
 
     if {$row >= $numcommits} return
     foreach id [lindex $rowidlist $row] {
@@ -1465,9 +1531,9 @@ proc drawcmitrow {row} {
        getcommit $id
     }
     assigncolor $id
-    if {[info exists commitlisted($id)] && [info exists parents($id)]
-       && $parents($id) ne {}} {
-       set rmx [drawparentlinks $id $row $col $parents($id)]
+    set olds [lindex $parentlist $row]
+    if {$olds ne {}} {
+       set rmx [drawparentlinks $id $row $col $olds]
     } else {
        set rmx 0
     }
@@ -1511,15 +1577,22 @@ proc clear_display {} {
 
 proc assigncolor {id} {
     global colormap colors nextcolor
-    global parents nparents children nchildren
+    global commitrow parentlist children childlist
     global cornercrossings crossings
 
     if {[info exists colormap($id)]} return
     set ncolors [llength $colors]
-    if {$nchildren($id) == 1} {
-       set child [lindex $children($id) 0]
+    if {[info exists commitrow($id)]} {
+       set kids [lindex $childlist $commitrow($id)]
+    } elseif {[info exists children($id)]} {
+       set kids $children($id)
+    } else {
+       set kids {}
+    }
+    if {[llength $kids] == 1} {
+       set child [lindex $kids 0]
        if {[info exists colormap($child)]
-           && $nparents($child) == 1} {
+           && [llength [lindex $parentlist $commitrow($child)]] == 1} {
            set colormap($id) $colormap($child)
            return
        }
@@ -1552,17 +1625,15 @@ proc assigncolor {id} {
        set origbad $badcolors
     }
     if {[llength $badcolors] < $ncolors - 1} {
-       foreach child $children($id) {
+       foreach child $kids {
            if {[info exists colormap($child)]
                && [lsearch -exact $badcolors $colormap($child)] < 0} {
                lappend badcolors $colormap($child)
            }
-           if {[info exists parents($child)]} {
-               foreach p $parents($child) {
-                   if {[info exists colormap($p)]
-                       && [lsearch -exact $badcolors $colormap($p)] < 0} {
-                       lappend badcolors $colormap($p)
-                   }
+           foreach p [lindex $parentlist $commitrow($child)] {
+               if {[info exists colormap($p)]
+                   && [lsearch -exact $badcolors $colormap($p)] < 0} {
+                   lappend badcolors $colormap($p)
                }
            }
        }
@@ -1657,14 +1728,14 @@ proc drawtags {id x xt y1} {
 }
 
 proc checkcrossings {row endrow} {
-    global displayorder parents rowidlist
+    global displayorder parentlist rowidlist
 
     for {} {$row < $endrow} {incr row} {
        set id [lindex $displayorder $row]
        set i [lsearch -exact [lindex $rowidlist $row] $id]
        if {$i < 0} continue
        set idlist [lindex $rowidlist [expr {$row+1}]]
-       foreach p $parents($id) {
+       foreach p [lindex $parentlist $row] {
            set j [lsearch -exact $idlist $p]
            if {$j > 0} {
                if {$j < $i - 1} {
@@ -2046,7 +2117,7 @@ proc insertmatch {l id} {
 
 proc findfiles {} {
     global selectedline numcommits displayorder ctext
-    global ffileline finddidsel parents nparents
+    global ffileline finddidsel parentlist
     global findinprogress findstartline findinsertpos
     global treediffs fdiffid fdiffsneeded fdiffpos
     global findmergefiles
@@ -2064,7 +2135,7 @@ proc findfiles {} {
     set fdiffsneeded {}
     while 1 {
        set id [lindex $displayorder $l]
-       if {$findmergefiles || $nparents($id) == 1} {
+       if {$findmergefiles || [llength [lindex $parentlist $l]] == 1} {
            if {![info exists treediffs($id)]} {
                append diffsneeded "$id\n"
                lappend fdiffsneeded $id
@@ -2096,7 +2167,7 @@ proc findfiles {} {
     . config -cursor watch
     settextcursor watch
     set findinprogress 1
-    findcont $id
+    findcont
     update
 }
 
@@ -2143,7 +2214,7 @@ proc donefilediff {} {
            set treediffs($nullid) {}
            if {[info exists findid] && $nullid eq $findid} {
                unset findid
-               findcont $nullid
+               findcont
            }
            incr fdiffpos
        }
@@ -2154,20 +2225,21 @@ proc donefilediff {} {
        }
        if {[info exists findid] && $fdiffid eq $findid} {
            unset findid
-           findcont $fdiffid
+           findcont
        }
     }
 }
 
-proc findcont {id} {
-    global findid treediffs parents nparents
+proc findcont {} {
+    global findid treediffs parentlist
     global ffileline findstartline finddidsel
     global displayorder numcommits matchinglines findinprogress
     global findmergefiles
 
     set l $ffileline
-    while 1 {
-       if {$findmergefiles || $nparents($id) == 1} {
+    while {1} {
+       set id [lindex $displayorder $l]
+       if {$findmergefiles || [llength [lindex $parentlist $l]] == 1} {
            if {![info exists treediffs($id)]} {
                set findid $id
                set ffileline $l
@@ -2189,7 +2261,6 @@ proc findcont {id} {
            set l 0
        }
        if {$l == $findstartline} break
-       set id [lindex $displayorder $l]
     }
     stopfindproc
     if {!$finddidsel} {
@@ -2286,10 +2357,26 @@ proc appendwithlinks {text} {
     $ctext tag bind link <Leave> { %W configure -cursor $curtextcursor }
 }
 
+proc viewnextline {dir} {
+    global canv linespc
+
+    $canv delete hover
+    set ymax [lindex [$canv cget -scrollregion] 3]
+    set wnow [$canv yview]
+    set wtop [expr {[lindex $wnow 0] * $ymax}]
+    set newtop [expr {$wtop + $dir * $linespc}]
+    if {$newtop < 0} {
+       set newtop 0
+    } elseif {$newtop > $ymax} {
+       set newtop $ymax
+    }
+    allcanvs yview moveto [expr {$newtop * 1.0 / $ymax}]
+}
+
 proc selectline {l isnew} {
     global canv canv2 canv3 ctext commitinfo selectedline
     global displayorder linehtag linentag linedtag
-    global canvy0 linespc parents nparents children
+    global canvy0 linespc parentlist childlist
     global cflist currentid sha1entry
     global commentend idtags linknum
     global mergemax numcommits
@@ -2379,9 +2466,10 @@ proc selectline {l isnew} {
     }
  
     set comment {}
-    if {$nparents($id) > 1} {
+    set olds [lindex $parentlist $l]
+    if {[llength $olds] > 1} {
        set np 0
-       foreach p $parents($id) {
+       foreach p $olds {
            if {$np >= $mergemax} {
                set tag mmax
            } else {
@@ -2392,17 +2480,13 @@ proc selectline {l isnew} {
            incr np
        }
     } else {
-       if {[info exists parents($id)]} {
-           foreach p $parents($id) {
-               append comment "Parent: [commit_descriptor $p]\n"
-           }
+       foreach p $olds {
+           append comment "Parent: [commit_descriptor $p]\n"
        }
     }
 
-    if {[info exists children($id)]} {
-       foreach c $children($id) {
-           append comment "Child:  [commit_descriptor $c]\n"
-       }
+    foreach c [lindex $childlist $l] {
+       append comment "Child:  [commit_descriptor $c]\n"
     }
     append comment "\n"
     append comment [lindex $info 5]
@@ -2417,13 +2501,25 @@ proc selectline {l isnew} {
 
     $cflist delete 0 end
     $cflist insert end "Comments"
-    if {$nparents($id) <= 1} {
+    if {[llength $olds] <= 1} {
        startdiff $id
     } else {
-       mergediff $id
+       mergediff $id $l
     }
 }
 
+proc selfirstline {} {
+    unmarkmatches
+    selectline 0 1
+}
+
+proc sellastline {} {
+    global numcommits
+    unmarkmatches
+    set l [expr {$numcommits - 1}]
+    selectline $l 1
+}
+
 proc selnextline {dir} {
     global selectedline
     if {![info exists selectedline]} return
@@ -2432,6 +2528,25 @@ proc selnextline {dir} {
     selectline $l 1
 }
 
+proc selnextpage {dir} {
+    global canv linespc selectedline numcommits
+
+    set lpp [expr {([winfo height $canv] - 2) / $linespc}]
+    if {$lpp < 1} {
+       set lpp 1
+    }
+    allcanvs yview scroll [expr {$dir * $lpp}] units
+    if {![info exists selectedline]} return
+    set l [expr {$selectedline + $dir * $lpp}]
+    if {$l < 0} {
+       set l 0
+    } elseif {$l >= $numcommits} {
+        set l [expr $numcommits - 1]
+    }
+    unmarkmatches
+    selectline $l 1    
+}
+
 proc unselectline {} {
     global selectedline
 
@@ -2489,9 +2604,10 @@ proc goforw {} {
     }
 }
 
-proc mergediff {id} {
-    global parents diffmergeid diffopts mdifffd
+proc mergediff {id l} {
+    global diffmergeid diffopts mdifffd
     global difffilestart diffids
+    global parentlist
 
     set diffmergeid $id
     set diffids $id
@@ -2505,12 +2621,13 @@ proc mergediff {id} {
     }
     fconfigure $mdf -blocking 0
     set mdifffd($id) $mdf
-    fileevent $mdf readable [list getmergediffline $mdf $id]
+    set np [llength [lindex $parentlist $l]]
+    fileevent $mdf readable [list getmergediffline $mdf $id $np]
     set nextupdate [expr {[clock clicks -milliseconds] + 100}]
 }
 
-proc getmergediffline {mdf id} {
-    global diffmergeid ctext cflist nextupdate nparents mergemax
+proc getmergediffline {mdf id np} {
+    global diffmergeid ctext cflist nextupdate mergemax
     global difffilestart mdifffd
 
     set n [gets $mdf line]
@@ -2543,7 +2660,6 @@ proc getmergediffline {mdf id} {
        # do nothing
     } else {
        # parse the prefix - one ' ', '-' or '+' for each parent
-       set np $nparents($id)
        set spaces {}
        set minuses {}
        set pluses {}
@@ -2584,7 +2700,7 @@ proc getmergediffline {mdf id} {
        incr nextupdate 100
        fileevent $mdf readable {}
        update
-       fileevent $mdf readable [list getmergediffline $mdf $id]
+       fileevent $mdf readable [list getmergediffline $mdf $id $np]
     }
 }
 
@@ -2611,7 +2727,7 @@ proc addtocflist {ids} {
 }
 
 proc gettreediffs {ids} {
-    global treediff parents treepending
+    global treediff treepending
     set treepending $ids
     set treediff {}
     if {[catch \
@@ -2846,13 +2962,15 @@ proc sha1change {n1 n2 op} {
 }
 
 proc gotocommit {} {
-    global sha1string currentid commitrow tagids
+    global sha1string currentid commitrow tagids headids
     global displayorder numcommits
 
     if {$sha1string == {}
        || ([info exists currentid] && $sha1string == $currentid)} return
     if {[info exists tagids($sha1string)]} {
        set id $tagids($sha1string)
+    } elseif {[info exists headids($sha1string)]} {
+       set id $headids($sha1string)
     } else {
        set id [string tolower $sha1string]
        if {[regexp {^[0-9a-f]{4,39}$} $id]} {
@@ -2878,7 +2996,7 @@ proc gotocommit {} {
     if {[regexp {^[0-9a-fA-F]{4,}$} $sha1string]} {
        set type "SHA1 id"
     } else {
-       set type "Tag"
+       set type "Tag/Head"
     }
     error_popup "$type $sha1string is not known"
 }
@@ -2979,7 +3097,7 @@ proc arrowjump {id n y} {
 }
 
 proc lineclick {x y id isnew} {
-    global ctext commitinfo children cflist canv thickerline
+    global ctext commitinfo childlist commitrow cflist canv thickerline
 
     if {![info exists commitinfo($id)] && ![getcommit $id]} return
     unmarkmatches
@@ -3018,10 +3136,11 @@ proc lineclick {x y id isnew} {
     $ctext insert end "\tAuthor:\t[lindex $info 1]\n"
     set date [formatdate [lindex $info 2]]
     $ctext insert end "\tDate:\t$date\n"
-    if {[info exists children($id)]} {
+    set kids [lindex $childlist $commitrow($id)]
+    if {$kids ne {}} {
        $ctext insert end "\nChildren:"
        set i 0
-       foreach child $children($id) {
+       foreach child $kids {
            incr i
            if {![info exists commitinfo($child)] && ![getcommit $child]} continue
            set info $commitinfo($child)
@@ -3368,7 +3487,6 @@ proc listrefs {id} {
 
 proc rereadrefs {} {
     global idtags idheads idotherrefs
-    global tagids headids otherrefids
 
     set refids [concat [array names idtags] \
                    [array names idheads] [array names idotherrefs]]
@@ -3777,6 +3895,7 @@ if {$tclencoding == {}} {
 
 set mainfont {Helvetica 9}
 set textfont {Courier 9}
+set uifont {Helvetica 9 bold}
 set findmergefiles 0
 set maxgraphpct 50
 set maxwidth 16
index f3cb79b..52e2400 100644 (file)
@@ -1202,6 +1202,7 @@ read_message( FILE *f, msg_data_t *msg )
                        p = xrealloc(msg->data, len+1);
                        if (!p)
                                break;
+                       msg->data = p;
                }
                r = fread( &msg->data[msg->len], 1, len - msg->len, f );
                if (r <= 0)
@@ -1332,6 +1333,12 @@ main(int argc, char **argv)
                return 1;
        }
 
+       total = count_messages( &all_msgs );
+       if (!total) {
+               fprintf(stderr,"no messages to send\n");
+               return 1;
+       }
+
        /* write it to the imap server */
        ctx = imap_open_store( &server );
        if (!ctx) {
@@ -1339,7 +1346,6 @@ main(int argc, char **argv)
                return 1;
        }
 
-       total = count_messages( &all_msgs );
        fprintf( stderr, "sending %d message%s\n", total, (total!=1)?"s":"" );
        ctx->name = imap_folder;
        while (1) {
index c0e1311..d95f0d9 100644 (file)
@@ -28,12 +28,12 @@ void *patch_delta(void *src_buf, unsigned long src_size,
        top = delta_buf + delta_size;
 
        /* make sure the orig file size matches what we expect */
-       size = get_delta_hdr_size(&data);
+       size = get_delta_hdr_size(&data, top);
        if (size != src_size)
                return NULL;
 
        /* now the result size */
-       size = get_delta_hdr_size(&data);
+       size = get_delta_hdr_size(&data, top);
        dst_buf = malloc(size + 1);
        if (!dst_buf)
                return NULL;
@@ -52,21 +52,37 @@ void *patch_delta(void *src_buf, unsigned long src_size,
                        if (cmd & 0x20) cp_size |= (*data++ << 8);
                        if (cmd & 0x40) cp_size |= (*data++ << 16);
                        if (cp_size == 0) cp_size = 0x10000;
+                       if (cp_off + cp_size < cp_size ||
+                           cp_off + cp_size > src_size ||
+                           cp_size > size)
+                               goto bad;
                        memcpy(out, src_buf + cp_off, cp_size);
                        out += cp_size;
-               } else {
+                       size -= cp_size;
+               } else if (cmd) {
+                       if (cmd > size)
+                               goto bad;
                        memcpy(out, data, cmd);
                        out += cmd;
                        data += cmd;
+                       size -= cmd;
+               } else {
+                       /*
+                        * cmd == 0 is reserved for future encoding
+                        * extensions. In the mean time we must fail when
+                        * encountering them (might be data corruption).
+                        */
+                       goto bad;
                }
        }
 
        /* sanity check */
-       if (data != top || out - dst_buf != size) {
+       if (data != top || size != 0) {
+               bad:
                free(dst_buf);
                return NULL;
        }
 
-       *dst_size = size;
+       *dst_size = out - dst_buf;
        return dst_buf;
 }
index 22141e2..1301502 100644 (file)
@@ -30,6 +30,7 @@ static const char rev_list_usage[] =
 "    --unpacked\n"
 "    --header | --pretty\n"
 "    --abbrev=nr | --no-abbrev\n"
+"    --abbrev-commit\n"
 "  special purpose:\n"
 "    --bisect"
 ;
@@ -39,6 +40,7 @@ struct rev_info revs;
 static int bisect_list = 0;
 static int verbose_header = 0;
 static int abbrev = DEFAULT_ABBREV;
+static int abbrev_commit = 0;
 static int show_timestamp = 0;
 static int hdr_termination = 0;
 static const char *commit_prefix = "";
@@ -52,7 +54,10 @@ static void show_commit(struct commit *commit)
                fputs(commit_prefix, stdout);
        if (commit->object.flags & BOUNDARY)
                putchar('-');
-       fputs(sha1_to_hex(commit->object.sha1), stdout);
+       if (abbrev_commit && abbrev)
+               fputs(find_unique_abbrev(commit->object.sha1, abbrev), stdout);
+       else
+               fputs(sha1_to_hex(commit->object.sha1), stdout);
        if (revs.parents) {
                struct commit_list *parents = commit->parents;
                while (parents) {
@@ -319,6 +324,14 @@ int main(int argc, const char **argv)
                        abbrev = 0;
                        continue;
                }
+               if (!strcmp(arg, "--abbrev")) {
+                       abbrev = DEFAULT_ABBREV;
+                       continue;
+               }
+               if (!strcmp(arg, "--abbrev-commit")) {
+                       abbrev_commit = 1;
+                       continue;
+               }
                if (!strncmp(arg, "--abbrev=", 9)) {
                        abbrev = strtoul(arg + 9, NULL, 10);
                        if (abbrev && abbrev < MINIMUM_ABBREV)
index ba8c4f7..e3d0113 100644 (file)
@@ -808,10 +808,12 @@ static int packed_delta_info(unsigned char *base_sha1,
                 * the result size.
                 */
                data = delta_head;
-               get_delta_hdr_size(&data); /* ignore base size */
+
+               /* ignore base size */
+               get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
 
                /* Read the result size */
-               result_size = get_delta_hdr_size(&data);
+               result_size = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
                *sizep = result_size;
        }
        return 0;
diff --git a/xdiff-interface.c b/xdiff-interface.c
new file mode 100644 (file)
index 0000000..6a82da7
--- /dev/null
@@ -0,0 +1,104 @@
+#include "cache.h"
+#include "xdiff-interface.h"
+
+static int parse_num(char **cp_p, int *num_p)
+{
+       char *cp = *cp_p;
+       int num = 0;
+       int read_some;
+
+       while ('0' <= *cp && *cp <= '9')
+               num = num * 10 + *cp++ - '0';
+       if (!(read_some = cp - *cp_p))
+               return -1;
+       *cp_p = cp;
+       *num_p = num;
+       return 0;
+}
+
+int parse_hunk_header(char *line, int len,
+                     int *ob, int *on,
+                     int *nb, int *nn)
+{
+       char *cp;
+       cp = line + 4;
+       if (parse_num(&cp, ob)) {
+       bad_line:
+               return error("malformed diff output: %s", line);
+       }
+       if (*cp == ',') {
+               cp++;
+               if (parse_num(&cp, on))
+                       goto bad_line;
+       }
+       else
+               *on = 1;
+       if (*cp++ != ' ' || *cp++ != '+')
+               goto bad_line;
+       if (parse_num(&cp, nb))
+               goto bad_line;
+       if (*cp == ',') {
+               cp++;
+               if (parse_num(&cp, nn))
+                       goto bad_line;
+       }
+       else
+               *nn = 1;
+       return -!!memcmp(cp, " @@", 3);
+}
+
+static void consume_one(void *priv_, char *s, unsigned long size)
+{
+       struct xdiff_emit_state *priv = priv_;
+       char *ep;
+       while (size) {
+               unsigned long this_size;
+               ep = memchr(s, '\n', size);
+               this_size = (ep == NULL) ? size : (ep - s + 1);
+               priv->consume(priv, s, this_size);
+               size -= this_size;
+               s += this_size;
+       }
+}
+
+int xdiff_outf(void *priv_, mmbuffer_t *mb, int nbuf)
+{
+       struct xdiff_emit_state *priv = priv_;
+       int i;
+
+       for (i = 0; i < nbuf; i++) {
+               if (mb[i].ptr[mb[i].size-1] != '\n') {
+                       /* Incomplete line */
+                       priv->remainder = realloc(priv->remainder,
+                                                 priv->remainder_size +
+                                                 mb[i].size);
+                       memcpy(priv->remainder + priv->remainder_size,
+                              mb[i].ptr, mb[i].size);
+                       priv->remainder_size += mb[i].size;
+                       continue;
+               }
+
+               /* we have a complete line */
+               if (!priv->remainder) {
+                       consume_one(priv, mb[i].ptr, mb[i].size);
+                       continue;
+               }
+               priv->remainder = realloc(priv->remainder,
+                                         priv->remainder_size +
+                                         mb[i].size);
+               memcpy(priv->remainder + priv->remainder_size,
+                      mb[i].ptr, mb[i].size);
+               consume_one(priv, priv->remainder,
+                           priv->remainder_size + mb[i].size);
+               free(priv->remainder);
+               priv->remainder = NULL;
+               priv->remainder_size = 0;
+       }
+       if (priv->remainder) {
+               consume_one(priv, priv->remainder, priv->remainder_size);
+               free(priv->remainder);
+               priv->remainder = NULL;
+               priv->remainder_size = 0;
+       }
+       return 0;
+}
diff --git a/xdiff-interface.h b/xdiff-interface.h
new file mode 100644 (file)
index 0000000..1346908
--- /dev/null
@@ -0,0 +1,21 @@
+#ifndef XDIFF_INTERFACE_H
+#define XDIFF_INTERFACE_H
+
+#include "xdiff/xdiff.h"
+
+struct xdiff_emit_state;
+
+typedef void (*xdiff_emit_consume_fn)(void *, char *, unsigned long);
+
+struct xdiff_emit_state {
+       xdiff_emit_consume_fn consume;
+       char *remainder;
+       unsigned long remainder_size;
+};
+
+int xdiff_outf(void *priv_, mmbuffer_t *mb, int nbuf);
+int parse_hunk_header(char *line, int len,
+                     int *ob, int *on,
+                     int *nb, int *nn);
+
+#endif