The most important optimization for performance when repacking is the
ability to reuse data from a previous pack as is and bypass any delta
or even SHA1 computation by simply copying the raw data from one pack
to another directly.
The problem with this is that any data corruption within a copied object
would go unnoticed and the new (repacked) pack would be self-consistent
with its own checksum despite containing a corrupted object. This is a
real issue that already happened at least once in the past.
In some attempt to prevent this, we validate the copied data by inflating
it and making sure no error is signaled by zlib. But this is still not
perfect as a significant portion of a pack content is made of object
headers and references to delta base objects which are not deflated and
therefore not validated when repacking actually making the pack data reuse
still not as safe as it could be.
Of course a full SHA1 validation could be performed, but that implies
full data inflating and delta replaying which is extremely costly, which
cost the data reuse optimization was designed to avoid in the first place.
So the best solution to this is simply to store a CRC32 of the raw pack
data for each object in the pack index. This way any object in a pack can
be validated before being copied as is in another pack, including header
and any other non deflated data.
Why CRC32 instead of a faster checksum like Adler32? Quoting Wikipedia:
Jonathan Stone discovered in 2001 that Adler-32 has a weakness for very
short messages. He wrote "Briefly, the problem is that, for very short
packets, Adler32 is guaranteed to give poor coverage of the available
bits. Don't take my word for it, ask Mark Adler. :-)" The problem is
that sum A does not wrap for short messages. The maximum value of A for
a 128-byte message is 32640, which is below the value 65521 used by the
modulo operation. An extended explanation can be found in RFC 3309,
which mandates the use of CRC32 instead of Adler-32 for SCTP, the
Stream Control Transmission Protocol.
In the context of a GIT pack, we have lots of small objects, especially
deltas, which are likely to be quite small and in a size range for which
Adler32 is dimed not to be sufficient. Another advantage of CRC32 is the
possibility for recovery from certain types of small corruptions like
single bit errors which are the most probable type of corruptions.
OK what this patch does is to compute the CRC32 of each object written to
a pack within pack-objects. It is not written to the index yet and it is
obviously not validated when reusing pack data yet either.
Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This abstracts away the size of the hash values when copying them
from memory location to memory location, much as the introduction
of hashcmp abstracted away hash value comparsion.
A few call sites were using char* rather than unsigned char* so
I added the cast rather than open hashcpy to be void*. This is a
reasonable tradeoff as most call sites already use unsigned char*
and the existing hashcmp is also declared to be unsigned char*.
[jc: Splitted the patch to "master" part, to be followed by a
patch for merge-recursive.c which is not in "master" yet.
Fixed the cast in the latter hunk to combine-diff.c which was
wrong in the original.
Also converted ones left-over in combine-diff.c, diff-lib.c and
upload-pack.c ]
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
With the change in default, "git add ." on kernel dir is about
twice as fast as before, with only minimal (0.5%) change in
object size. The speed difference is even more noticeable
when committing large files, which is now up to 8 times faster.
The configurability is through setting core.compression = [-1..9]
which maps to the zlib constants; -1 is the default, 0 is no
compression, and 1..9 are various speed/size tradeoffs, 9
being slowest.
Signed-off-by: Joachim B Haga (cjhaga@fys.uio.no)
Acked-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
ANSI C99 doesn't allow void-pointer arithmetic. This patch fixes this in
various ways. Usually the strategy that required the least changes was used.
Signed-off-by: Florian Forster <octo@verplant.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
We had errno==EINTR check after read(2)/write(2) sprinkled all
over the places, always doing continue. Consolidate them into
xread()/xwrite() wrapper routines.
Credits for suggestion goes to HPA -- bugs are mine.
Signed-off-by: Junio C Hamano <junkio@cox.net>
sha1create() and sha1fd() malloc the returned struct sha1file;
sha1close() should free it.
Signed-off-by: Sergey Vlasov <vsu@altlinux.ru>
Signed-off-by: Junio C Hamano <junkio@cox.net>
IIRC our strategy was to let the users' umask take care of the
final mode bits. This patch fixes places that deviate from it.
Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Also, make the writing of the SHA1 as a end-header be conditional: not
every user will necessarily want to write the SHA1 to the file itself,
even though current users do (but we migh end up using the same helper
functions for the object files themselves, that don't do this).
This also makes the packed index file contain the SHA1 of the packed
data file at the end (just before its own SHA1). That way you can
validate the pairing of the two if you want to.
We want to be able to check their integrity later, and putting the
sha1-sum of the contents at the end is a good thing. The writing
routines are generic, so we could try to re-use them for the index file,
instead of having the same logic duplicated.
Update unpack-objects to know about the extra 20 bytes at the end
of the index.
