##// END OF EJS Templates
copies: use the rust code for `combine_changeset_copies`...
copies: use the rust code for `combine_changeset_copies` Changeset centric copy tracing now use the rust code. The rust code focussed on simplicity and will be optimised later. So the performance is not great yet. Now that all the pieces are in place we can start working on performance in the coming changesets. Below is a table that summarize how slower we got: Repo Cases Source-Rev Dest-Rev Py-time Rust-time Difference Factor ------------------------------------------------------------------------------------------------------------------------------------ mercurial x_revs_x_added_0_copies ad6b123de1c7 39cfcef4f463 : 0.000049 s, 0.000046 s, -0.000003 s, × 0.9388 mercurial x_revs_x_added_x_copies 2b1c78674230 0c1d10351869 : 0.000112 s, 0.000173 s, +0.000061 s, × 1.5446 mercurial x000_revs_x000_added_x_copies 81f8ff2a9bf2 dd3267698d84 : 0.004216 s, 0.006303 s, +0.002087 s, × 1.4950 pypy x_revs_x_added_0_copies aed021ee8ae8 099ed31b181b : 0.000204 s, 0.000229 s, +0.000025 s, × 1.1225 pypy x_revs_x000_added_0_copies 4aa4e1f8e19a 359343b9ac0e : 0.000058 s, 0.000056 s, -0.000002 s, × 0.9655 pypy x_revs_x_added_x_copies ac52eb7bbbb0 72e022663155 : 0.000112 s, 0.000143 s, +0.000031 s, × 1.2768 pypy x_revs_x00_added_x_copies c3b14617fbd7 ace7255d9a26 : 0.000339 s, 0.001166 s, +0.000827 s, × 3.4395 pypy x_revs_x000_added_x000_copies df6f7a526b60 a83dc6a2d56f : 0.010214 s, 0.022931 s, +0.012717 s, × 2.2451 pypy x000_revs_xx00_added_0_copies 89a76aede314 2f22446ff07e : 0.047497 s, 0.852446 s, +0.804949 s, × 17.9474 pypy x000_revs_x000_added_x_copies 8a3b5bfd266e 2c68e87c3efe : 0.075297 s, 2.221824 s, +2.146527 s, × 29.5075 pypy x000_revs_x000_added_x000_copies 89a76aede314 7b3dda341c84 : 0.057322 s, 1.194162 s, +1.136840 s, × 20.8325 pypy x0000_revs_x_added_0_copies d1defd0dc478 c9cb1334cc78 : 0.796264 s, 62.468362 s, +61.672098 s, × 78.4518 pypy x0000_revs_xx000_added_0_copies bf2c629d0071 4ffed77c095c : 0.020491 s, 0.022116 s, +0.001625 s, × 1.0793 pypy x0000_revs_xx000_added_x000_copies 08ea3258278e d9fa043f30c0 : 0.121612 s, 2.972788 s, +2.851176 s, × 24.4449 netbeans x_revs_x_added_0_copies fb0955ffcbcd a01e9239f9e7 : 0.000143 s, 0.000180 s, +0.000037 s, × 1.2587 netbeans x_revs_x000_added_0_copies 6f360122949f 20eb231cc7d0 : 0.000112 s, 0.000123 s, +0.000011 s, × 1.0982 netbeans x_revs_x_added_x_copies 1ada3faf6fb6 5a39d12eecf4 : 0.000232 s, 0.000315 s, +0.000083 s, × 1.3578 netbeans x_revs_x00_added_x_copies 35be93ba1e2c 9eec5e90c05f : 0.000721 s, 0.001297 s, +0.000576 s, × 1.7989 netbeans x000_revs_xx00_added_0_copies eac3045b4fdd 51d4ae7f1290 : 0.010115 s, 0.024884 s, +0.014769 s, × 2.4601 netbeans x000_revs_x000_added_x_copies e2063d266acd 6081d72689dc : 0.015461 s, 0.032653 s, +0.017192 s, × 2.1120 netbeans x000_revs_x000_added_x000_copies ff453e9fee32 411350406ec2 : 0.060756 s, 4.230118 s, +4.169362 s, × 69.6247 netbeans x0000_revs_xx000_added_x000_copies 588c2d1ced70 1aad62e59ddd : 0.605842 s, killed mozilla-central x_revs_x_added_0_copies 3697f962bb7b 7015fcdd43a2 : 0.000164 s, 0.000197 s, +0.000033 s, × 1.2012 mozilla-central x_revs_x000_added_0_copies dd390860c6c9 40d0c5bed75d : 0.000331 s, 0.000622 s, +0.000291 s, × 1.8792 mozilla-central x_revs_x_added_x_copies 8d198483ae3b 14207ffc2b2f : 0.000249 s, 0.000296 s, +0.000047 s, × 1.1888 mozilla-central x_revs_x00_added_x_copies 98cbc58cc6bc 446a150332c3 : 0.000711 s, 0.001626 s, +0.000915 s, × 2.2869 mozilla-central x_revs_x000_added_x000_copies 3c684b4b8f68 0a5e72d1b479 : 0.003438 s, 0.006218 s, +0.002780 s, × 1.8086 mozilla-central x_revs_x0000_added_x0000_copies effb563bb7e5 c07a39dc4e80 : 0.069869 s, 0.132760 s, +0.062891 s, × 1.9001 mozilla-central x000_revs_xx00_added_0_copies 6100d773079a 04a55431795e : 0.005701 s, 0.029001 s, +0.023300 s, × 5.0870 mozilla-central x000_revs_x000_added_x_copies 9f17a6fc04f9 2d37b966abed : 0.005757 s, 0.005886 s, +0.000129 s, × 1.0224 mozilla-central x000_revs_x000_added_x000_copies 7c97034feb78 4407bd0c6330 : 0.061826 s, 3.619850 s, +3.558024 s, × 58.5490 mozilla-central x0000_revs_xx000_added_0_copies 9eec5917337d 67118cc6dcad : 0.043354 s, 0.058678 s, +0.015324 s, × 1.3535 mozilla-central x0000_revs_xx000_added_x000_copies f78c615a656c 96a38b690156 : 0.198979 s, 11.926587 s, +11.727608 s, × 59.9389 mozilla-central x00000_revs_x0000_added_x0000_copies 6832ae71433c 4c222a1d9a00 : 2.067096 s, killed mozilla-central x00000_revs_x00000_added_x000_copies 76caed42cf7c 1daa622bbe42 : 3.102616 s, killed mozilla-try x_revs_x_added_0_copies aaf6dde0deb8 9790f499805a : 0.001212 s, 0.001204 s, -0.000008 s, × 0.9934 mozilla-try x_revs_x000_added_0_copies d8d0222927b4 5bb8ce8c7450 : 0.001237 s, 0.001217 s, -0.000020 s, × 0.9838 mozilla-try x_revs_x_added_x_copies 092fcca11bdb 936255a0384a : 0.000557 s, 0.000605 s, +0.000048 s, × 1.0862 mozilla-try x_revs_x00_added_x_copies b53d2fadbdb5 017afae788ec : 0.001532 s, 0.001876 s, +0.000344 s, × 1.2245 mozilla-try x_revs_x000_added_x000_copies 20408ad61ce5 6f0ee96e21ad : 0.035166 s, 0.078190 s, +0.043024 s, × 2.2235 mozilla-try x_revs_x0000_added_x0000_copies effb563bb7e5 c07a39dc4e80 : 0.070336 s, 0.135428 s, +0.065092 s, × 1.9254 mozilla-try x000_revs_xx00_added_0_copies 6100d773079a 04a55431795e : 0.006080 s, 0.029123 s, +0.023043 s, × 4.7900 mozilla-try x000_revs_x000_added_x_copies 9f17a6fc04f9 2d37b966abed : 0.006099 s, 0.006141 s, +0.000042 s, × 1.0069 mozilla-try x000_revs_x000_added_x000_copies 1346fd0130e4 4c65cbdabc1f : 0.064317 s, 4.857827 s, +4.793510 s, × 75.5294 mozilla-try x0000_revs_x_added_0_copies 63519bfd42ee a36a2a865d92 : 0.303263 s, 10.674920 s, +10.371657 s, × 35.2002 mozilla-try x0000_revs_x_added_x_copies 9fe69ff0762d bcabf2a78927 : 0.292804 s, 9.789462 s, +9.496658 s, × 33.4335 mozilla-try x0000_revs_xx000_added_x_copies 156f6e2674f2 4d0f2c178e66 : 0.107594 s, 1.087890 s, +0.980296 s, × 10.1111 mozilla-try x0000_revs_xx000_added_0_copies 9eec5917337d 67118cc6dcad : 0.045202 s, 0.060556 s, +0.015354 s, × 1.3397 mozilla-try x0000_revs_xx000_added_x000_copies 89294cd501d9 7ccb2fc7ccb5 : 1.926277 s, killed mozilla-try x0000_revs_x0000_added_x0000_copies e928c65095ed e951f4ad123a : 0.794492 s, killed mozilla-try x00000_revs_x_added_0_copies 6a320851d377 1ebb79acd503 : 84.521497 s, killed mozilla-try x00000_revs_x00000_added_0_copies dc8a3ca7010e d16fde900c9c : 0.965937 s, 19.647038 s, +18.681101 s, × 20.3399 mozilla-try x00000_revs_x_added_x_copies 5173c4b6f97c 95d83ee7242d : 83.367146 s, killed mozilla-try x00000_revs_x000_added_x_copies 9126823d0e9c ca82787bb23c : 84.260895 s, killed mozilla-try x00000_revs_x0000_added_x0000_copies 8d3fafa80d4b eb884023b810 : 3.274537 s, killed mozilla-try x00000_revs_x00000_added_x0000_copies 1b661134e2ca 1ae03d022d6d : 42.235843 s, killed mozilla-try x00000_revs_x00000_added_x000_copies 9b2a99adc05e 8e29777b48e6 : 49.872829 s, killed Differential Revision: https://phab.mercurial-scm.org/D9299

File last commit:

r46174:b68b1910 default
r46576:a66568f2 default
Show More
patch.rs
369 lines | 11.8 KiB | application/rls-services+xml | RustLexer
use byteorder::{BigEndian, ByteOrder};
/// A chunk of data to insert, delete or replace in a patch
///
/// A chunk is:
/// - an insertion when `!data.is_empty() && start == end`
/// - an deletion when `data.is_empty() && start < end`
/// - a replacement when `!data.is_empty() && start < end`
/// - not doing anything when `data.is_empty() && start == end`
#[derive(Debug, Clone)]
struct Chunk<'a> {
/// The start position of the chunk of data to replace
start: u32,
/// The end position of the chunk of data to replace (open end interval)
end: u32,
/// The data replacing the chunk
data: &'a [u8],
}
impl Chunk<'_> {
/// Adjusted start of the chunk to replace.
///
/// The offset, taking into account the growth/shrinkage of data
/// induced by previously applied chunks.
fn start_offset_by(&self, offset: i32) -> u32 {
let start = self.start as i32 + offset;
assert!(start >= 0, "negative chunk start should never happen");
start as u32
}
/// Adjusted end of the chunk to replace.
///
/// The offset, taking into account the growth/shrinkage of data
/// induced by previously applied chunks.
fn end_offset_by(&self, offset: i32) -> u32 {
self.start_offset_by(offset) + self.data.len() as u32
}
/// Length of the replaced chunk.
fn replaced_len(&self) -> u32 {
self.end - self.start
}
/// Length difference between the replacing data and the replaced data.
fn len_diff(&self) -> i32 {
self.data.len() as i32 - self.replaced_len() as i32
}
}
/// The delta between two revisions data.
#[derive(Debug, Clone)]
pub struct PatchList<'a> {
/// A collection of chunks to apply.
///
/// Those chunks are:
/// - ordered from the left-most replacement to the right-most replacement
/// - non-overlapping, meaning that two chucks can not change the same
/// chunk of the patched data
chunks: Vec<Chunk<'a>>,
}
impl<'a> PatchList<'a> {
/// Create a `PatchList` from bytes.
pub fn new(data: &'a [u8]) -> Self {
let mut chunks = vec![];
let mut data = data;
while !data.is_empty() {
let start = BigEndian::read_u32(&data[0..]);
let end = BigEndian::read_u32(&data[4..]);
let len = BigEndian::read_u32(&data[8..]);
assert!(start <= end);
chunks.push(Chunk {
start,
end,
data: &data[12..12 + (len as usize)],
});
data = &data[12 + (len as usize)..];
}
PatchList { chunks }
}
/// Return the final length of data after patching
/// given its initial length .
fn size(&self, initial_size: i32) -> i32 {
self.chunks
.iter()
.fold(initial_size, |acc, chunk| acc + chunk.len_diff())
}
/// Apply the patch to some data.
pub fn apply(&self, initial: &[u8]) -> Vec<u8> {
let mut last: usize = 0;
let mut vec =
Vec::with_capacity(self.size(initial.len() as i32) as usize);
for Chunk { start, end, data } in self.chunks.iter() {
vec.extend(&initial[last..(*start as usize)]);
vec.extend(data.iter());
last = *end as usize;
}
vec.extend(&initial[last..]);
vec
}
/// Combine two patch lists into a single patch list.
///
/// Applying consecutive patches can lead to waste of time and memory
/// as the changes introduced by one patch can be overridden by the next.
/// Combining patches optimizes the whole patching sequence.
fn combine(&mut self, other: &mut Self) -> Self {
let mut chunks = vec![];
// Keep track of each growth/shrinkage resulting from applying a chunk
// in order to adjust the start/end of subsequent chunks.
let mut offset = 0i32;
// Keep track of the chunk of self.chunks to process.
let mut pos = 0;
// For each chunk of `other`, chunks of `self` are processed
// until they start after the end of the current chunk.
for Chunk { start, end, data } in other.chunks.iter() {
// Add chunks of `self` that start before this chunk of `other`
// without overlap.
while pos < self.chunks.len()
&& self.chunks[pos].end_offset_by(offset) <= *start
{
let first = self.chunks[pos].clone();
offset += first.len_diff();
chunks.push(first);
pos += 1;
}
// The current chunk of `self` starts before this chunk of `other`
// with overlap.
// The left-most part of data is added as an insertion chunk.
// The right-most part data is kept in the chunk.
if pos < self.chunks.len()
&& self.chunks[pos].start_offset_by(offset) < *start
{
let first = &mut self.chunks[pos];
let (data_left, data_right) = first.data.split_at(
(*start - first.start_offset_by(offset)) as usize,
);
let left = Chunk {
start: first.start,
end: first.start,
data: data_left,
};
first.data = data_right;
offset += left.len_diff();
chunks.push(left);
// There is no index incrementation because the right-most part
// needs further examination.
}
// At this point remaining chunks of `self` starts after
// the current chunk of `other`.
// `start_offset` will be used to adjust the start of the current
// chunk of `other`.
// Offset tracking continues with `end_offset` to adjust the end
// of the current chunk of `other`.
let mut next_offset = offset;
// Discard the chunks of `self` that are totally overridden
// by the current chunk of `other`
while pos < self.chunks.len()
&& self.chunks[pos].end_offset_by(next_offset) <= *end
{
let first = &self.chunks[pos];
next_offset += first.len_diff();
pos += 1;
}
// Truncate the left-most part of chunk of `self` that overlaps
// the current chunk of `other`.
if pos < self.chunks.len()
&& self.chunks[pos].start_offset_by(next_offset) < *end
{
let first = &mut self.chunks[pos];
let how_much_to_discard =
*end - first.start_offset_by(next_offset);
first.data = &first.data[(how_much_to_discard as usize)..];
next_offset += how_much_to_discard as i32;
}
// Add the chunk of `other` with adjusted position.
chunks.push(Chunk {
start: (*start as i32 - offset) as u32,
end: (*end as i32 - next_offset) as u32,
data,
});
// Go back to normal offset tracking for the next `o` chunk
offset = next_offset;
}
// Add remaining chunks of `self`.
for elt in &self.chunks[pos..] {
chunks.push(elt.clone());
}
PatchList { chunks }
}
}
/// Combine a list of patch list into a single patch optimized patch list.
pub fn fold_patch_lists<'a>(lists: &[PatchList<'a>]) -> PatchList<'a> {
if lists.len() <= 1 {
if lists.is_empty() {
PatchList { chunks: vec![] }
} else {
lists[0].clone()
}
} else {
let (left, right) = lists.split_at(lists.len() / 2);
let mut left_res = fold_patch_lists(left);
let mut right_res = fold_patch_lists(right);
left_res.combine(&mut right_res)
}
}
#[cfg(test)]
mod tests {
use super::*;
struct PatchDataBuilder {
data: Vec<u8>,
}
impl PatchDataBuilder {
pub fn new() -> Self {
Self { data: vec![] }
}
pub fn replace(
&mut self,
start: usize,
end: usize,
data: &[u8],
) -> &mut Self {
assert!(start <= end);
self.data.extend(&(start as i32).to_be_bytes());
self.data.extend(&(end as i32).to_be_bytes());
self.data.extend(&(data.len() as i32).to_be_bytes());
self.data.extend(data.iter());
self
}
pub fn get(&mut self) -> &[u8] {
&self.data
}
}
#[test]
fn test_ends_before() {
let data = vec![0u8, 0u8, 0u8];
let mut patch1_data = PatchDataBuilder::new();
patch1_data.replace(0, 1, &[1, 2]);
let mut patch1 = PatchList::new(patch1_data.get());
let mut patch2_data = PatchDataBuilder::new();
patch2_data.replace(2, 4, &[3, 4]);
let mut patch2 = PatchList::new(patch2_data.get());
let patch = patch1.combine(&mut patch2);
let result = patch.apply(&data);
assert_eq!(result, vec![1u8, 2, 3, 4]);
}
#[test]
fn test_starts_after() {
let data = vec![0u8, 0u8, 0u8];
let mut patch1_data = PatchDataBuilder::new();
patch1_data.replace(2, 3, &[3]);
let mut patch1 = PatchList::new(patch1_data.get());
let mut patch2_data = PatchDataBuilder::new();
patch2_data.replace(1, 2, &[1, 2]);
let mut patch2 = PatchList::new(patch2_data.get());
let patch = patch1.combine(&mut patch2);
let result = patch.apply(&data);
assert_eq!(result, vec![0u8, 1, 2, 3]);
}
#[test]
fn test_overridden() {
let data = vec![0u8, 0, 0];
let mut patch1_data = PatchDataBuilder::new();
patch1_data.replace(1, 2, &[3, 4]);
let mut patch1 = PatchList::new(patch1_data.get());
let mut patch2_data = PatchDataBuilder::new();
patch2_data.replace(1, 4, &[1, 2, 3]);
let mut patch2 = PatchList::new(patch2_data.get());
let patch = patch1.combine(&mut patch2);
let result = patch.apply(&data);
assert_eq!(result, vec![0u8, 1, 2, 3]);
}
#[test]
fn test_right_most_part_is_overridden() {
let data = vec![0u8, 0, 0];
let mut patch1_data = PatchDataBuilder::new();
patch1_data.replace(0, 1, &[1, 3]);
let mut patch1 = PatchList::new(patch1_data.get());
let mut patch2_data = PatchDataBuilder::new();
patch2_data.replace(1, 4, &[2, 3, 4]);
let mut patch2 = PatchList::new(patch2_data.get());
let patch = patch1.combine(&mut patch2);
let result = patch.apply(&data);
assert_eq!(result, vec![1u8, 2, 3, 4]);
}
#[test]
fn test_left_most_part_is_overridden() {
let data = vec![0u8, 0, 0];
let mut patch1_data = PatchDataBuilder::new();
patch1_data.replace(1, 3, &[1, 3, 4]);
let mut patch1 = PatchList::new(patch1_data.get());
let mut patch2_data = PatchDataBuilder::new();
patch2_data.replace(0, 2, &[1, 2]);
let mut patch2 = PatchList::new(patch2_data.get());
let patch = patch1.combine(&mut patch2);
let result = patch.apply(&data);
assert_eq!(result, vec![1u8, 2, 3, 4]);
}
#[test]
fn test_mid_is_overridden() {
let data = vec![0u8, 0, 0];
let mut patch1_data = PatchDataBuilder::new();
patch1_data.replace(0, 3, &[1, 3, 3, 4]);
let mut patch1 = PatchList::new(patch1_data.get());
let mut patch2_data = PatchDataBuilder::new();
patch2_data.replace(1, 3, &[2, 3]);
let mut patch2 = PatchList::new(patch2_data.get());
let patch = patch1.combine(&mut patch2);
let result = patch.apply(&data);
assert_eq!(result, vec![1u8, 2, 3, 4]);
}
}