The classic diff adds only the lines that it's going to consider,
during the diff, to an array. A mapping between the compacted
array, and the lines of the file that they reference, is
facilitated by this array.
Signed-off-by: Ezekiel Newren <ezekielnewren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The field rindex describes an index offset for other arrays. Change it
to size_t.
Changing the type of rindex from long to size_t has no cascading
refactor impact because it is only ever used to directly index other
arrays.
Signed-off-by: Ezekiel Newren <ezekielnewren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
size_t is used because nreff describes the number of elements in memory
for rindex.
Signed-off-by: Ezekiel Newren <ezekielnewren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
size_t is used because nrec describes the number of elements for both
recs, and for 'changed' + 2.
Signed-off-by: Ezekiel Newren <ezekielnewren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The ha field is serving two different purposes, which makes the code
harder to read. At first glance, it looks like many places assume
there could never be hash collisions between lines of the two input
files. In reality, line_hash is used together with xdl_recmatch() to
ensure correct comparisons of lines, even when collisions occur.
To make this clearer, the old ha field has been split:
* line_hash: a straightforward hash of a line, independent of any
external context. Its type is uint64_t, as it comes from a fixed
width hash function.
* minimal_perfect_hash: Not a new concept, but now a separate
field. It comes from the classifier's general-purpose hash table,
which assigns each line a unique and minimal hash across the two
files. A size_t is used here because it's meant to be used to
index an array. This also avoids ` as usize` casts on the Rust
side when using it to index a slice.
Signed-off-by: Ezekiel Newren <ezekielnewren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
size_t is the appropriate type because size is describing the number of
elements, bytes in this case, in memory.
Signed-off-by: Ezekiel Newren <ezekielnewren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Make xrecord_t.ptr uint8_t because it's referring to bytes in memory.
In order to avoid a refactor avalanche, many uses of this field were
cast to char* or similar.
Places where casting was unnecessary:
xemit.c:156
xmerge.c:124
xmerge.c:127
xmerge.c:164
xmerge.c:169
xmerge.c:172
xmerge.c:178
Signed-off-by: Ezekiel Newren <ezekielnewren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
ptrdiff_t is appropriate for dstart and dend because they both describe
positive or negative offsets relative to a pointer.
Signed-off-by: Ezekiel Newren <ezekielnewren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The only values possible for 'changed' is 1 and 0, which exactly maps
to a bool type. It might not look like this because action1 and action2
(which use to be dis1, and dis2) were also of type char and were
assigned numerical values within a few lines of 'changed' (what used to
be rchg).
Using DISCARD/KEEP/INVESTIGATE for action1[i]/action2[j], and true/false
for changed[k] makes it clear to future readers that these are
logically separate concepts.
Best-viewed-with: --color-words
Signed-off-by: Ezekiel Newren <ezekielnewren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The field rchg (now 'changed') declares if a line in a file is changed
or not. A later commit will change it's type from 'char' to 'bool'
to make its purpose even more clear.
Best-viewed-with: --color-words
Signed-off-by: Ezekiel Newren <ezekielnewren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
xdfile_t currently uses chastore_t which is an arena allocator. I
think that xrecord_t used to be a linked list and recs didn't exist
originally. When recs was added I think they forgot to remove
xdfile_t.next, but was overlooked. This dual data structure setup
makes the code somewhat confusing.
Additionally the C type chastore_t isn't FFI friendly, and provides
little to no performance benefit over using realloc to grow an array.
Performance impact of deleting fields from xdfile_t:
Deleting ha is about 5% slower.
Deleting cha is about 5% faster.
Delete ha, but keep cha
time hyperfine --warmup 3 -L exe build_v2.51.0/git,build_delete_ha/git '{exe} log --oneline --shortstat --diff-algorithm=myers -3000 v2.39.1 >/dev/null'
Benchmark 1: build_v2.51.0/git log --oneline --shortstat --diff-algorithm=myers -3000 v2.39.1 >/dev/null
Time (mean ± σ): 1.269 s ± 0.017 s [User: 1.135 s, System: 0.128 s]
Range (min … max): 1.249 s … 1.286 s 10 runs
Benchmark 2: build_delete_ha/git log --oneline --shortstat --diff-algorithm=myers -3000 v2.39.1 >/dev/null
Time (mean ± σ): 1.339 s ± 0.017 s [User: 1.234 s, System: 0.099 s]
Range (min … max): 1.320 s … 1.358 s 10 runs
Summary
build_v2.51.0/git log --oneline --shortstat --diff-algorithm=myers -3000 v2.39.1 >/dev/null ran
1.06 ± 0.02 times faster than build_delete_ha/git log --oneline --shortstat --diff-algorithm=myers -3000 v2.39.1 >/dev/null
Delete cha, but keep ha
time hyperfine --warmup 3 -L exe build_v2.51.0/git,build_delete_chastore/git '{exe} log --oneline --shortstat --diff-algorithm=myers -3000 v2.39.1 >/dev/null'
Benchmark 1: build_v2.51.0/git log --oneline --shortstat --diff-algorithm=myers -3000 v2.39.1 >/dev/null
Time (mean ± σ): 1.290 s ± 0.001 s [User: 1.154 s, System: 0.130 s]
Range (min … max): 1.288 s … 1.292 s 10 runs
Benchmark 2: build_delete_chastore/git log --oneline --shortstat --diff-algorithm=myers -3000 v2.39.1 >/dev/null
Time (mean ± σ): 1.232 s ± 0.017 s [User: 1.105 s, System: 0.121 s]
Range (min … max): 1.205 s … 1.249 s 10 runs
Summary
build_delete_chastore/git log --oneline --shortstat --diff-algorithm=myers -3000 v2.39.1 >/dev/null ran
1.05 ± 0.01 times faster than build_v2.51.0/git log --oneline --shortstat --diff-algorithm=myers -3000 v2.39.1 >/dev/null
Delete ha AND chastore
time hyperfine --warmup 3 -L exe build_v2.51.0/git,build_delete_ha_and_chastore/git '{exe} log --oneline --shortstat --diff-algorithm=myers -3000 v2.39.1 >/dev/null'
Benchmark 1: build_v2.51.0/git log --oneline --shortstat --diff-algorithm=myers -3000 v2.39.1 >/dev/null
Time (mean ± σ): 1.291 s ± 0.002 s [User: 1.156 s, System: 0.129 s]
Range (min … max): 1.287 s … 1.295 s 10 runs
Benchmark 2: build_delete_ha_and_chastore/git log --oneline --shortstat --diff-algorithm=myers -3000 v2.39.1 >/dev/null
Time (mean ± σ): 1.306 s ± 0.001 s [User: 1.195 s, System: 0.105 s]
Range (min … max): 1.305 s … 1.308 s 10 runs
Summary
build_v2.51.0/git log --oneline --shortstat --diff-algorithm=myers -3000 v2.39.1 >/dev/null ran
1.01 ± 0.00 times faster than build_delete_ha_and_chastore/git log --oneline --shortstat --diff-algorithm=myers -3000 v2.39.1 >/dev/null
Best-viewed-with: --color-words
Signed-off-by: Ezekiel Newren <ezekielnewren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When 0 <= i < xdfile_t.nreff the following is true:
xdfile_t.ha[i] == xdfile_t.recs[xdfile_t.rindex[i]]
This makes the code about 5% slower. The fields rindex and ha are
specific to the classic diff (myers and minimal). I plan on creating a
struct for classic diff, but there's a lot of cleanup that needs to be
done before that can happen and leaving ha in would make those cleanups
harder to follow.
A subsequent commit will delete the chastore cha from xdfile_t. That
later commit will investigate deleting ha and cha independently and
together.
Signed-off-by: Ezekiel Newren <ezekielnewren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
xrecord_t.next, xdfile_t.hbits, xdfile_t.rhash are initialized,
but never used for anything by the code. Remove them.
Best-viewed-with: --color-words
Signed-off-by: Ezekiel Newren <ezekielnewren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The mailing address for the FSF has changed over the years. Rather than
updating the address across all files, refer readers to gnu.org, as the
GNU GPL documentation now suggests for license notices. The mailing
address is retained in the full license files (COPYING and LGPL-2.1).
The old address is still present in t/diff-lib/COPYING. This is
intentional, as the file is used in tests and the contents are not
expected to change.
Signed-off-by: Todd Zullinger <tmz@pobox.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This uses "git-apply --whitespace=strip" to fix whitespace errors that have
crept in to our source files over time. There are a few files that need
to have trailing whitespaces (most notably, test vectors). The results
still passes the test, and build result in Documentation/ area is unchanged.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This uses a simplified libxdiff setup to generate unified diffs _without_
doing fork/execve of GNU "diff".
This has several huge advantages, for example:
Before:
[torvalds@g5 linux]$ time git diff v2.6.16.. > /dev/null
real 0m24.818s
user 0m13.332s
sys 0m8.664s
After:
[torvalds@g5 linux]$ time git diff v2.6.16.. > /dev/null
real 0m4.563s
user 0m2.944s
sys 0m1.580s
and the fact that this should be a lot more portable (ie we can ignore all
the issues with doing fork/execve under Windows).
Perhaps even more importantly, this allows us to do diffs without actually
ever writing out the git file contents to a temporary file (and without
any of the shell quoting issues on filenames etc etc).
NOTE! THIS PATCH DOES NOT DO THAT OPTIMIZATION YET! I was lazy, and the
current "diff-core" code actually will always write the temp-files,
because it used to be something that you simply had to do. So this current
one actually writes a temp-file like before, and then reads it into memory
again just to do the diff. Stupid.
But if this basic infrastructure is accepted, we can start switching over
diff-core to not write temp-files, which should speed things up even
further, especially when doing big tree-to-tree diffs.
Now, in the interest of full disclosure, I should also point out a few
downsides:
- the libxdiff algorithm is different, and I bet GNU diff has gotten a
lot more testing. And the thing is, generating a diff is not an exact
science - you can get two different diffs (and you will), and they can
both be perfectly valid. So it's not possible to "validate" the
libxdiff output by just comparing it against GNU diff.
- GNU diff does some nice eye-candy, like trying to figure out what the
last function was, and adding that information to the "@@ .." line.
libxdiff doesn't do that.
- The libxdiff thing has some known deficiencies. In particular, it gets
the "\No newline at end of file" case wrong. So this is currently for
the experimental branch only. I hope Davide will help fix it.
That said, I think the huge performance advantage, and the fact that it
integrates better is definitely worth it. But it should go into a
development branch at least due to the missing newline issue.
Technical note: this is based on libxdiff-0.17, but I did some surgery to
get rid of the extraneous fat - stuff that git doesn't need, and seriously
cutting down on mmfile_t, which had much more capabilities than the diff
algorithm either needed or used. In this version, "mmfile_t" is just a
trivial <pointer,length> tuple.
That said, I tried to keep the differences to simple removals, so that you
can do a diff between this and the libxdiff origin, and you'll basically
see just things getting deleted. Even the mmfile_t simplifications are
left in a state where the diffs should be readable.
Apologies to Davide, whom I'd love to get feedback on this all from (I
wrote my own "fill_mmfile()" for the new simpler mmfile_t format: the old
complex format had a helper function for that, but I did my surgery with
the goal in mind that eventually we _should_ just do
mmfile_t mf;
buf = read_sha1_file(sha1, type, &size);
mf->ptr = buf;
mf->size = size;
.. use "mf" directly ..
which was really a nightmare with the old "helpful" mmfile_t, and really
is that easy with the new cut-down interfaces).
[ Btw, as any hawk-eye can see from the diff, this was actually generated
with itself, so it is "self-hosting". That's about all the testing it
has gotten, along with the above kernel diff, which eye-balls correctly,
but shows the newline issue when you double-check it with "git-apply" ]
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
