We've had this notion of a "object_list" for a long time, which eventually
grew a "name" member because some users (notably git-rev-list) wanted to
name each object as it is generated.
That object_list is great for some things, but it isn't all that wonderful
for others, and the "name" member is generally not used by everybody.
This patch splits the users of the object_list array up into two: the
traditional list users, who want the list-like format, and who don't
actually use or want the name. And another class of users that really used
the list as an extensible array, and generally wanted to name the objects.
The patch is fairly straightforward, but it's also biggish. Most of it
really just cleans things up: switching the revision parsing and listing
over to the array makes things like the builtin-diff usage much simpler
(we now see exactly how many members the array has, and we don't get the
objects reversed from the order they were on the command line).
One of the main reasons for doing this at all is that the malloc overhead
of the simple object list was actually pretty high, and the array is just
a lot denser. So this patch brings down memory usage by git-rev-list by
just under 3% (on top of all the other memory use optimizations) on the
mozilla archive.
It does add more lines than it removes, and more importantly, it adds a
whole new infrastructure for maintaining lists of objects, but on the
other hand, the new dynamic array code is pretty obvious. The change to
builtin-diff-tree.c shows a fairly good example of why an array interface
is sometimes more natural, and just much simpler for everybody.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This is really the dregs of my effort to not waste memory in git-rev-list,
and makes barely one percent of a difference in the memory footprint, but
hey, it's also a pretty small patch.
It discards the parent lists and the commit buffer after the commit has
been shown by git-rev-list (and "git log" - which already did the commit
buffer part), and frees the commit list entry that was used by the
revision walker.
The big win would be to get rid of the "refs" pointer in the object
structure (another 5%), because it's only used by fsck. That would require
some pretty major surgery to fsck, though, so I'm timid and did the less
interesting but much easier part instead.
This (percentually) makes a bigger difference to "git log" and friends,
since those are walking _just_ commits, and thus the list entries tend to
be a bigger percentage of the memory use. But the "list all objects" case
does improve too.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This shrinks "struct object" by a small amount, by getting rid of the
"struct type *" pointer and replacing it with a 3-bit bitfield instead.
In addition, we merge the bitfields and the "flags" field, which
incidentally should also remove a useless 4-byte padding from the object
when in 64-bit mode.
Now, our "struct object" is still too damn large, but it's now less
obviously bloated, and of the remaining fields, only the "util" (which is
not used by most things) is clearly something that should be eventually
discarded.
This shrinks the "git-rev-list --all" memory use by about 2.5% on the
kernel archive (and, perhaps more importantly, on the larger mozilla
archive). That may not sound like much, but I suspect it's more on a
64-bit platform.
There are other remaining inefficiencies (the parent lists, for example,
probably have horrible malloc overhead), but this was pretty obvious.
Most of the patch is just changing the comparison of the "type" pointer
from one of the constant string pointers to the appropriate new TYPE_xxx
small integer constant.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
The tree-walking conversion of the "process_tree()" function
broke packing by using an unrelated variable from outer scope.
Signed-off-by: Junio C Hamano <junkio@cox.net>
This adds a "tree_entry()" function that combines the common operation of
doing a "tree_entry_extract()" + "update_tree_entry()".
It also has a simplified calling convention, designed for simple loops
that traverse over a whole tree: the arguments are pointers to the tree
descriptor and a name_entry structure to fill in, and it returns a boolean
"true" if there was an entry left to be gotten in the tree.
This allows tree traversal with
struct tree_desc desc;
struct name_entry entry;
desc.buf = tree->buffer;
desc.size = tree->size;
while (tree_entry(&desc, &entry) {
... use "entry.{path, sha1, mode, pathlen}" ...
}
which is not only shorter than writing it out in full, it's hopefully less
error prone too.
[ It's actually a tad faster too - we don't need to recalculate the entry
pathlength in both extract and update, but need to do it only once.
Also, some callers can avoid doing a "strlen()" on the result, since
it's returned as part of the name_entry structure.
However, by now we're talking just 1% speedup on "git-rev-list --objects
--all", and we're definitely at the point where tree walking is no
longer the issue any more. ]
NOTE! Not everybody wants to use this new helper function, since some of
the tree walkers very much on purpose do the descriptor update separately
from the entry extraction. So the "extract + update" sequence still
remains as the core sequence, this is just a simplified interface.
We should probably add a silly two-line inline helper function for
initializing the descriptor from the "struct tree" too, just to cut down
on the noise from that common "desc" initializer.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Instead, just use the tree buffer directly, and use the tree-walk
infrastructure to walk the buffers instead of the tree-entry list.
The tree-entry list is inefficient, and generates tons of small
allocations for no good reason. The tree-walk infrastructure is
generally no harder to use than following a linked list, and allows
us to do most tree parsing in-place.
Some programs still use the old tree-entry lists, and are a bit
painful to convert without major surgery. For them we have a helper
function that creates a temporary tree-entry list on demand.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This is preparatory work for further cleanups, where we try to make
tree_entry look more like the more efficient tree-walk descriptor.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This allows us to avoid allocating information for names etc, because
we can just use the information from the tree buffer directly.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This finally removes the tree-entry list from "struct tree", since most of
the users can just use the tree-walk infrastructure to walk the raw tree
buffers instead of the tree-entry list.
The tree-entry list is inefficient, and generates tons of small
allocations for no good reason. The tree-walk infrastructure is generally
no harder to use than following a linked list, and allows us to do most
tree parsing in-place.
Some programs still use the old tree-entry lists, and are a bit painful to
convert without major surgery. For them we have a helper function that
creates a temporary tree-entry list on demand. We can convert those too
eventually, but with this they no longer affect any users who don't need
the explicit lists.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This is preparatory work for further cleanups, where we try to make
tree_entry look more like the more efficient tree-walk descriptor.
Instead of having a union of pointers to blob/tree/objects, this just
makes "struct tree_entry" have the raw SHA1, and makes all the users use
that instead (often that implies adding a "lookup_tree(..)" on the sha1,
but sometimes the user just wanted the SHA1 in the first place, and it
just avoids an unnecessary indirection).
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This allows us to avoid allocating information for names etc, because
we can just use the information from the tree buffer directly.
We still keep the old "tree_entry_list" in struct tree as well, so old
users aren't affected, apart from the fact that the allocations are
different (if you free a tree entry, you should no longer free the name
allocation for it, since it's allocated as part of "tree->buffer")
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Martin Langhoff points out that "git repack -a" ends up using up a lot of
memory for big archives, and that git cvsimport probably should do only
incremental repacks in order to avoid having repacking flush all the
caches.
The big majority of the memory usage of repacking is from git rev-list
tracking all objects, and this patch should go a long way in avoiding the
excessive memory usage: the bulk of it was due to the object names being
leaked from the tree parser.
For the historic Linux kernel archive, this simple patch does:
Before:
/usr/bin/time git-rev-list --all --objects > /dev/null
72.45user 0.82system 1:13.55elapsed 99%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+125376minor)pagefaults 0swaps
After:
/usr/bin/time git-rev-list --all --objects > /dev/null
75.22user 0.48system 1:16.34elapsed 99%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+43921minor)pagefaults 0swaps
where we do end up wasting a bit of time on some extra strdup()s (which
could be avoided, but that would require tracking where the pathnames came
from), but we avoid a lot of memory usage.
Minor page faults track maximum RSS very closely (each page fault maps in
one page into memory), so the reduction from 125376 page faults to 43921
means a rough reduction of VM footprint from almost half a gigabyte to
about a third of that. Those numbers were also double-checked by looking
at "top" while the process was running.
(Side note: at least part of the remaining VM footprint is the mapping of
the 177MB pack-file, so the remaining memory use is at least partly "well
behaved" from a project caching perspective).
For the current git archive itself, the memory usage for a "--all
--objects" rev-list invocation dropped from 7128 pages to 2318 (27MB to
9MB), so the reduction seems to hold for much smaller projects too.
For regular "git-rev-list" usage (ie without the "--objects" flag) this
patch has no impact.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This patch touches a couple of files, because it adds options to print a
custom text just after the subject of a commit, and just after the
diffstat.
[jc: made "many dashes" used as the boundary leader into a single
variable, to reduce the possibility of later tweaks to miscount the
number of dashes to break it.]
Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This was surprisingly easy. The diff is truly minimal: rename "main()" to
"cmd_rev_list()" in rev-list.c, and rename the whole file to reflect its
new built-in status.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
On Sun, 16 Apr 2006, Junio C Hamano wrote:
>
> In the mid-term, I am hoping we can drop the generate_header()
> callchain _and_ the custom code that formats commit log in-core,
> found in cmd_log_wc().
Ok, this was nastier than expected, just because the dependencies between
the different log-printing stuff were absolutely _everywhere_, but here's
a patch that does exactly that.
The patch is not very easy to read, and the "--patch-with-stat" thing is
still broken (it does not call the "show_log()" thing properly for
merges). That's not a new bug. In the new world order it _should_ do
something like
if (rev->logopt)
show_log(rev, rev->logopt, "---\n");
but it doesn't. I haven't looked at the --with-stat logic, so I left it
alone.
That said, this patch removes more lines than it adds, and in particular,
the "cmd_log_wc()" loop is now a very clean:
while ((commit = get_revision(rev)) != NULL) {
log_tree_commit(rev, commit);
free(commit->buffer);
commit->buffer = NULL;
}
so it doesn't get much prettier than this. All the complexity is entirely
hidden in log-tree.c, and any code that needs to flush the log literally
just needs to do the "if (rev->logopt) show_log(...)" incantation.
I had to make the combined_diff() logic take a "struct rev_info" instead
of just a "struct diff_options", but that part is pretty clean.
This does change "git whatchanged" from using "diff-tree" as the commit
descriptor to "commit", and I changed one of the tests to reflect that new
reality. Otherwise everything still passes, and my other tests look fine
too.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
The boundary commits are shown for UI like gitk to draw them as
soon as topo-order sorting allows, and should not be omitted by
get_revision() filtering logic. As long as their immediate
child commits are shown, we should not filter them out.
Signed-off-by: Junio C Hamano <junkio@cox.net>
The big option parser unification broke rev-list the big way;
this makes it use options from the parsed revs structure.
Signed-off-by: Junio C Hamano <junkio@cox.net>
I noticed bisect does not work well without both good and bad.
Running this script in git.git repository would give you quite
different results:
#!/bin/sh
initial=e83c5163316f89bfbde7d9ab23ca2e25604af290
mid0=`git rev-list --bisect ^$initial --all`
git rev-list $mid0 | wc -l
git rev-list ^$mid0 --all | wc -l
mid1=`git rev-list --bisect --all`
git rev-list $mid1 | wc -l
git rev-list ^$mid1 --all | wc -l
The $initial commit is the very first commit you made. The
first midpoint bisects things evenly as designed, but the latter
does not.
The reason I got interested in this was because I was wondering
if something like the following would help people converting a
huge repository from foreign SCM, or preparing a repository to
be fetched over plain dumb HTTP only:
#!/bin/sh
N=4
P=.git/objects/pack
bottom=
while test 0 \< $N
do
N=$((N-1))
if test -z "$bottom"
then
newbottom=`git rev-list --bisect --all`
else
newbottom=`git rev-list --bisect ^$bottom --all`
fi
if test -z "$bottom"
then
rev_list="$newbottom"
elif test 0 = $N
then
rev_list="^$bottom --all"
else
rev_list="^$bottom $newbottom"
fi
p=$(git rev-list --unpacked --objects $rev_list |
git pack-objects $P/pack)
git show-index <$P/pack-$p.idx | wc -l
bottom=$newbottom
done
The idea is to pack older half of the history to one pack, then
older half of the remaining history to another, to continue a
few times, using finer granularity as we get closer to the tip.
This may not matter, since for a truly huge history, running
bisect number of times could be quite time consuming, and we
might be better off running "git rev-list --all" once into a
temporary file, and manually pick cut-off points from the
resulting list of commits. After all we are talking about
"approximately half" for such an usage, and older history does
not matter much.
Signed-off-by: Junio C Hamano <junkio@cox.net>
This makes things that include revision.h build again.
Blame is also built, but I am not sure how well it works (or how
well it worked to begin with) -- it was relying on tree-diff to
be using whatever pathspec was used the last time, which smells
a bit suspicious.
Signed-off-by: Junio C Hamano <junkio@cox.net>
The parent rewriting feature caused us to create the whole history in one
go, and then simplify it later, because of how rewrite_parents() had been
written. However, with a little tweaking, it's perfectly possible to do
even that one incrementally.
Right now, this doesn't really much matter, because every user of
"--parents" will probably generally _also_ use "--topo-order", which will
cause the old non-incremental behaviour anyway. However, I'm hopeful that
we could make even the topological sort incremental, or at least
_partially_ so (for example, make it incremental up to the first merge).
In the meantime, this at least moves things in the right direction, and
removes a strange special case.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This should make --pretty=oneline a whole lot more readable for
people using 80-column terminals. Originally from Eric Wong.
Signed-off-by: Junio C Hamano <junkio@cox.net>
Not only do we do it in both rev-list.c and git.c, the revision walking
code will soon want to know whether we should rewrite parenthood
information or not.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Introduce tree-walk.[ch] and move "struct tree_desc" and
associated functions from various places.
Rename DIFF_FILE_CANON_MODE(mode) macro to canon_mode(mode) and
move it to cache.h. This macro returns the canonicalized
st_mode value in the host byte order for files, symlinks and
directories -- to be compared with a tree_desc entry.
create_ce_mode(mode) in cache.h is similar but is intended to be
used for index entries (so it does not work for directories) and
returns the value in the network byte order.
Signed-off-by: Junio C Hamano <junkio@cox.net>
With the new --boundary flag, the output from rev-list includes
the UNINTERESING commits at the boundary, which are usually not
shown. Their object names are prefixed with '-'.
For example, with this graph:
C side
/
A---B---D master
You would get something like this:
$ git rev-list --boundary --header --parents side..master
D B
tree D^{tree}
parent B
... log message for commit D here ...
\0-B A
tree B^{tree}
parent A
... log message for commit B here ...
\0
Signed-off-by: Junio C Hamano <junkio@cox.net>
We do not need to track object refs, neither we need to save commit
unless we are doing verbose header. A lot of traversal happens
inside prepare_revision_walk() these days so setting things up before
calling that function is necessary.
Signed-off-by: Junio C Hamano <junkio@cox.net>
Acked-by: Linus Torvalds <torvalds@osdl.org>
prune_fn in the rev_info structure is called in place of
try_to_simplify_commit. This makes it possible to do rename tracking
with a custom try_to_simplify_commit-like function.
This commit also introduces init_revisions which initialises the rev_info
structure with default values.
Signed-off-by: Fredrik Kuivinen <freku045@student.liu.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This ports the following options from rev-list based git-log
implementation:
* -<n>, -n<n>, and -n <n>. I am still wondering if we want
this natively supported by setup_revisions(), which already
takes --max-count. We may want to move them in the next
round. Also I am not sure if we can get away with not
setting revs->limited when we set max-count. The latest
rev-list.c and revision.c in this series do not, so I left
them as they are.
* --pretty and --pretty=<fmt>.
* --abbrev=<n> and --no-abbrev.
The previous commit already handles time-based limiters
(--since, --until and friends). The remaining things that
rev-list based git-log happens to do are not useful in a pure
log-viewing purposes, and not ported:
* --bisect (obviously).
* --header. I am actually in favor of doing the NUL
terminated record format, but rev-list based one always
passed --pretty, which defeated this option. Maybe next
round.
* --parents. I do not think of a reason a log viewer wants
this. The flag is primarily for feeding squashed history
via pipe to downstream tools.
Signed-off-by: Junio C Hamano <junkio@cox.net>
Well, assuming breaking --merge-order is fine, here's a patch (on top of
the other ones) that makes
git log <filename>
actually work, as far as I can tell.
I didn't add the logic for --before/--after flags, but that should be
pretty trivial, and is independent of this anyway.
Signed-off-by: Junio C Hamano <junkio@cox.net>
This actually moves the "meat" of the revision walking from rev-list.c
to the new library code in revision.h. It introduces the new functions
void prepare_revision_walk(struct rev_info *revs);
struct commit *get_revision(struct rev_info *revs);
to prepare and then walk the revisions that we have.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This makes the rewrite easier to validate in that revision flag
parsing and warlking part are now all in rev_info structure.
Signed-off-by: Junio C Hamano <junkio@cox.net>
This fixes "the other end has commit X but since then we tagged
that commit with tag T, and he says he wants T -- what is the
list of objects we need to send him?" question:
git-rev-list --objects ^X T
We ended up sending everything since the beginning of time X-<.
Signed-off-by: Junio C Hamano <junkio@cox.net>
This really just splits things up partially, and creates the
interface to set things up by parsing the command line.
No real code changes so far, although the parsing of filenames is a bit
stricter. In particular, if there is a "--", then we do not accept any
filenames before it, and if there isn't any "--", then we check that _all_
paths listed are valid, not just the first one.
The new argument parsing automatically also gives us "--default" and
"--not" handling as in git-rev-parse.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This helps to group the same files from different revs together,
while spreading files with the same basename in different
directories, to help pack-object.
Signed-off-by: Junio C Hamano <junkio@cox.net>
This new flag is similar to --objects, but causes rev-list to
show list of "uninteresting" commits that appear on the edge
commit prefixed with '-'.
Downstream pack-objects will be changed to take these as hints
to use the trees and blobs contained with them as base objects
of resulting pack, producing an incomplete (not self-contained)
pack.
Such a pack cannot be used in .git/objects/pack (it is prevented
by git-index-pack erroring out if it is fed to git-fetch-pack -k
or git-clone-pack), but would be useful when transferring only
small changes to huge blobs.
Signed-off-by: Junio C Hamano <junkio@cox.net>
This adds --date-order to rev-list; it is similar to topo order
in the sense that no parent comes before all of its children,
but otherwise things are still ordered in the commit timestamp
order.
The same flag is also added to show-branch.
Signed-off-by: Junio C Hamano <junkio@cox.net>
When we prettyprint commit log messages, merge parent names were
often very long and there was no way to abbreviate it.
This changes them to be abbreviated by default, and non-default
abbreviations can be specified with --no-abbrev or --abbrev=<n>
options.
Note that this affects only the prettyprinted parent names. The
output from --show-parents is meant for machine consumption and
is not affected by this flag.
Showing the same parent more than once for a commit does not
make much sense downstream, so stop it.
This can happen with an incorrectly made merge commit that
merges the same parent twice, but can happen in an otherwise
sane development history while squishing the history by taking
into account only commits that touch specified paths.
For example,
$ git rev-list --max-count=1 --parents addafaf -- rev-list.c
would have to show this commit ancestry graph:
.---o---.
/ \
.---*---o---.
/ 93b74bc \
---*---o---o-----o---o-----o addafafd8f6b34 \ /
.---o---o---.
\ /
.---*---.
3815f42
where 5 independent development tracks, only two of which have
changes in the specified paths since they forked. The last
change for the other three development tracks was done by the
same commit before they forked, and we were showing that three
times.
Signed-off-by: Junio C Hamano <junkio@cox.net>
This builds on top of the previous one.
Traditionally, head(1) and tail(1) allow their line limits to be
parsed this way.
Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Both -n<n> and -n <n> are supported. POSIX versions of head(1) and
tail(1) allow their line limits to be parsed this way. I find
--max-count to be a commonly used option, and also similar in spirit to
head/tail, so I decided to make life easier on my worn out (and lazy :)
fingers with this patch.
Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
The one thing I've considered doing (I really should) is to add a "stop
when you don't find the file" option to "git-rev-list". This patch does
some of the work towards that: it removes the "parent" thing when the
file disappears, so a "git annotate" could do do something like
git-rev-list --remove-empty --parents HEAD -- "$filename"
and it would get a good graph that stops when the filename disappears
(it's not perfect though: it won't remove all the unintersting commits).
It also simplifies the logic of finding tree differences a bit, at the
cost of making it a tad less efficient.
The old logic was two-phase: it would first simplify _only_ merges tree as
it traversed the tree, and then simplify the linear parts of the remainder
independently. That was pretty optimal from an efficiency standpoint
because it avoids doing any comparisons that we can see are unnecessary,
but it made it much harder to understand than it really needed to be.
The new logic is a lot more straightforward, and compares the trees as it
traverses the graph (ie everything is a single phase). That makes it much
easier to stop graph traversal at any point where a file disappears.
As an example, let's say that you have a git repository that has had a
file called "A" some time in the past. That file gets renamed to B, and
then gets renamed back again to A. The old "git-rev-list" would show two
commits: the commit that renames B to A (because it changes A) _and_ as
its parent the commit that renames A to B (because it changes A).
With the new --remove-empty flag, git-rev-list will show just the commit
that renames B to A as the "root" commit, and stop traversal there
(because that's what you want for "annotate" - you want to stop there, and
for every "root" commit you then separately see if it really is a new
file, or if the paths history disappeared because it was renamed from some
other file).
With this patch, you should be able to basically do a "poor mans 'git
annotate'" with a fairly simple loop:
push("HEAD", "$filename")
while (revision,filename = pop()) {
for each i in $(git-rev-list --parents --remove-empty $revision -- "$filename")
pseudo-parents($i) = git-rev-list parents for that line
if (pseudo-parents($i) is non-empty) {
show diff of $i against pseudo-parents
continue
}
/* See if the _real_ parents of $i had a rename */
parent($i) = real-parent($i)
if (find-rename in $parent($i)->$i)
push $parent($i), "old-name"
}
which should be doable in perl or something (doing stacks in shell is just
too painful to be worth it, so I'm not going to do this).
Anybody want to try?
Linus
Linus Torvalds