If our size computation overflows size_t, we may allocate a
much smaller buffer than we expected and overflow it. It's
probably impossible to trigger an overflow in most of these
sites in practice, but it is easy enough convert their
additions and multiplications into overflow-checking
variants. This may be fixing real bugs, and it makes
auditing the code easier.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
What used to happen is that diffcore_count_changes() simply ignored any
hashes in the destination that didn't match hashes in the source. EXCEPT
if the source hash didn't exist at all, in which case it would count _one_
destination hash that happened to have the "next" hash value. As a
consequence, newly added material was often undercounted, making output
from --dirstat and "complete rewrite" detection used by -B unrelialble.
This changes it so that:
- whenever it bypasses a destination hash (because it doesn't match a
source), it counts the bytes associated with that as "literal added"
- at the end (once we have used up all the source hashes), we do the same
thing with the remaining destination hashes.
- when hashes do match, and we use the difference in counts as a value,
we also use up that destination hash entry (the 'd++').
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Here's a test-patch. I don't guarantee anything, except that when I did
the timings I also did a "wc" on the result, and they matched..
Before:
[torvalds@woody linux]$ time git diff -l0 --stat -C v2.6.22.. | wc
7104 28574 438020
real 0m10.526s
user 0m10.401s
sys 0m0.136s
After:
[torvalds@woody linux]$ time ~/git/git diff -l0 --stat -C v2.6.22.. | wc
7104 28574 438020
real 0m8.876s
user 0m8.761s
sys 0m0.128s
but the diff is fairly simple, so if somebody will go over it and say
whether it's likely to be *correct* too, that 15% may well be worth it.
[ Side note, without rename detection, that diff takes just under three
seconds for me, so in that sense the improvement to the rename detection
itself is larger than the overall 15% - it brings the cost of just
rename detection from 7.5s to 5.9s, which would be on the order of just
over a 20% performance improvement. ]
Hmm. The patch depends on half-way subtle issues like the fact that the
hashtables are guaranteed to not be full => we're guaranteed to have zero
counts at the end => we don't need to do any steenking iterator count in
the loop. A few comments might in order.
Linus
This replaces an explicit initialization of filespec->is_binary
field used for rename/break followed by direct access to that
field with a wrapper function that lazily iniaitlizes and
accesses the field. We would add more attribute accesses for
the use of diff routines, and it would be better to make this
abstraction earlier.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This ignores CR byte in CRLF sequence in text file when
computing similarity of two blobs.
Usually this should not matter as nobody sane would be checking
in a file with CRLF line endings to the repository (they would
use autocrlf so that the repository copy would have LF line
endings).
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The comment at the top of the file described an old algorithm
that was neutral to text/binary differences (it hashed sliding
window of N-byte sequences and counted overlaps), but long time
ago we switched to a new heuristics that are more suitable for
line oriented (read: text) files that are much faster.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We may want to use richer information on the data we are dealing
with in this function, so instead of passing a buffer address
and length, just pass the diffcore_filespec structure. Existing
callers always call this function with parameters taken from a
filespec anyway, so there is no functionality changes.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The rotating 64-bit number was not really rotating, and worse
yet ulong was longer than 64-bit on 64-bit architectures X-<.
Signed-off-by: Junio C Hamano <junkio@cox.net>
This tweaks the maximum hashvalue we use to hash the string into
without making the maximum size of the hashtable can grow from
the current limit. With this, the renames detected becomes a
bit more precise without incurring additional paging cost.
Signed-off-by: Junio C Hamano <junkio@cox.net>
To reduce wasted memory, wait until the hash fills up more
densely before we rehash. This reduces the working set size a
bit further.
Signed-off-by: Junio C Hamano <junkio@cox.net>
This changes diffcore-rename to reuse statistics information
gathered during similarity estimation, and updates the hashtable
implementation used to keep track of the statistics to be
denser. This seems to give better performance.
Signed-off-by: Junio C Hamano <junkio@cox.net>
The textual line oriented change counter was fun but was not
very effective. It tended to overcount the changes. This one
changes it to a simple N-letter substring based implementation.
Signed-off-by: Junio C Hamano <junkio@cox.net>
This is to rework diffcore break/rename/copy detection code
so that it does not affected when deltifier code gets improved.
Signed-off-by: Junio C Hamano <junkio@cox.net>
Jeff King
Linus Torvalds
Junio C Hamano
Linus Torvalds
Junio C Hamano