Summary: The canonical URL type in Rust, `http::Uri`, does not support manipulating URLs easily. (e.g., concatenating path components, etc.) As such, switch to using the `Url` type from the `url` crate, which does support URL manipulation, and convert to `http::Uri` before passing the resulting URL to Hyper.
Reviewed By: phillco
Differential Revision: D13738139
fbshipit-source-id: c7de67f1596ebc1bdde89d3fe87086f49c32b5db
Summary:
Directory listing is different in every OS, and due to the current repack
implementation, this directly affect the order in which the packfiles are added
to the new one. Since the resulting packfile name depends on the hash of its
content, the name was influenced by the directory order.
By sorting the files in list_packs, the packfile name will be independent of
the directory listing and thus be the same for all the OSes.
Reviewed By: singhsrb
Differential Revision: D13700935
fbshipit-source-id: 01e055a0c1bcf7fb2dc4faf614dfb20cd4499017
Summary: For now, combine all files smaller than 100MB that accumulate to less than 4GB.
Reviewed By: DurhamG
Differential Revision: D13603760
fbshipit-source-id: 3fa74f1ced3d3ccd463af8f187ef5e0254e1820b
Summary: Use the newly introduced PackWriter to write the {data,history}packs.
Reviewed By: markbt
Differential Revision: D13603759
fbshipit-source-id: 528a6af7c4ac3321aeec0559805de12114224cfd
Summary:
The packfiles are currently being written via an unbuffered file. This is
inefficient as every write to the file results results in a write(2) syscall.
By buffering these writes we can reduce the number of syscalls and thus
increase the throughput of pack writing operations.
Reviewed By: markbt
Differential Revision: D13603758
fbshipit-source-id: 649186a852d427a1473695b1d32cc9cd87a74a75
Summary:
Update pest to 2.1.0.
This version has a new behaviour for parser error messages: the line feed at
the end of the line is shown in the error output.
Reviewed By: wez
Differential Revision: D13671099
fbshipit-source-id: b8d1142a44a56a0b21b3b72cf027f3f8a30f421e
Summary:
The revisionstore crate currently consists of several public submodules,
each exposing several public types. The APIs exposed by each of the modules
require using types from the other modules. As such, users of this crate are
forced to have complex nested imports to use any of its functionality.
This diff helps ease this problem by reexporting the public types exposed from
each of the public submodules at the top level, thereby allowing crate users to
`use` all of the required types without needing nested imports.
Reviewed By: singhsrb
Differential Revision: D13686913
fbshipit-source-id: 9fb3cce8783787aa5f3f974c7168afada5952712
Summary:
The later tries to read from the disk, while the former is purely in memory and
thus more efficient.
Reviewed By: DurhamG, markbt
Differential Revision: D13603757
fbshipit-source-id: 5fd120ba4065d6a65cb2982db9ab81db3ea26524
Summary:
Use the `Fallible` type alias provided by `failure` rather than defining our
own.
Differential Revision: D13657313
fbshipit-source-id: ae249bc15037cc2be019ce7ce8a440c153aa31cc
Summary:
Use the `Fallible` type alias provided by `failure` rather than defining our
own.
Differential Revision: D13657312
fbshipit-source-id: 55134ee93f1f3aaaeefe5644a4a1f2285603bc1c
Summary:
Use the `Fallible` type alias provided by `failure` rather than defining our
own.
Differential Revision: D13657314
fbshipit-source-id: f1a379089972f7f0066c49ddedf606d36b7ac260
Summary:
Use the `Fallible` type alias provided by `failure` rather than defining our
own.
Differential Revision: D13657310
fbshipit-source-id: cae73fc239a6ad30bb6ef56a664d1ef5a2a19b5f
Summary:
On some platforms, removing a file can fail if it's still mapped or opened. In
mercurial, this can happen during repack as the datapacks are removed while
still being mapped.
Reviewed By: DurhamG
Differential Revision: D13615938
fbshipit-source-id: fdc1ff9370e2767e52ee1828552f4598105f784f
Summary:
After repacking the data/history packs, we need to cleanup the
repacked files. This was an omission from D13363853.
Reviewed By: markbt
Differential Revision: D13577592
fbshipit-source-id: 36e7d5b8e86affe47cdd10d33a769969f02b8a62
Summary:
The python version of the mutable packs set the permission to read-only after
writing them, while the rust version keeps them writeable. Let's make the rust
one more consistent.
Reviewed By: markbt
Differential Revision: D13573572
fbshipit-source-id: 61256994562aa09058a88a7935c16dfd7ddf9d18
Summary:
Use of `write!` requires checking for errors, however in this case, there is no
need to use `write!`, as we just want the error as a string.
Reviewed By: ikostia
Differential Revision: D13596497
fbshipit-source-id: 5892025344936936188cf3a8ca227e71eff57d55
Summary:
When I was debugging an eden importer issue with Puneet, we saw errors caused
by important extensions (ex. remotefilelog, lz4revlog) not being loaded. It
turned out that configpaser was checking the "exe dir" to decide where to
load "system configs". For example, If we run:
C:\open\fbsource\fbcode\scm\hg\build\pythonMSVC2015\python.exe eden_import_helper.py
The "exe dir" is "C:\open\fbsource\fbcode\scm\hg\build", and system config is
not there.
Instead of copying "mercurial.ini" to every possible "exe dir", this diff just
switches to a hard-coded system config path. It's now consistent with what we
do on POSIX systems.
The logic to copy "mercurial.ini" to "C:\open\fbsource\fbcode\scm\hg" or
"C:\tools\hg" become unnecessary and are removed.
Reviewed By: singhsrb
Differential Revision: D13542939
fbshipit-source-id: 5fb50d8e42d36ec6da28af29de89966628fe5549
Summary:
`test-check-fix-code.t` was failing due to copyright header missing
from certain files. This commit fixes the files by running
```
contrib/fix-code.py FILE
```
as suggested in the failure message.
Reviewed By: DurhamG
Differential Revision: D13538506
fbshipit-source-id: d8063c9a0e665377a9976abeccb68fbef6781950
Summary:
Unfortunately required symbols are not exposed by lz4-sys. So we just declare
them ourselves.
Make sure it compresses better:
In [1]: c=open('/bin/bash').read();
In [2]: from mercurial.rust import lz4
In [3]: len(lz4.compress(c))
Out[3]: 762906
In [4]: len(lz4.compresshc(c))
Out[4]: 626970
While it's much slower for larger data (and compresshc is slower than pylz4):
Benchmarking (easy to compress data, 20MB)...
pylz4.compress: 10328.03 MB/s
rustlz4.compress_py: 9373.84 MB/s
pylz4.compressHC: 1666.80 MB/s
rustlz4.compresshc_py: 8298.57 MB/s
pylz4.decompress: 3953.03 MB/s
rustlz4.decompress_py: 3935.57 MB/s
Benchmarking (hard to compress data, 0.2MB)...
pylz4.compress: 4357.88 MB/s
rustlz4.compress_py: 4193.34 MB/s
pylz4.compressHC: 3740.40 MB/s
rustlz4.compresshc_py: 2730.71 MB/s
pylz4.decompress: 5600.94 MB/s
rustlz4.decompress_py: 5362.96 MB/s
Benchmarking (hard to compress data, 20MB)...
pylz4.compress: 5156.72 MB/s
rustlz4.compress_py: 5447.00 MB/s
pylz4.compressHC: 33.70 MB/s
rustlz4.compresshc_py: 22.25 MB/s
pylz4.decompress: 2375.42 MB/s
rustlz4.decompress_py: 5755.46 MB/s
Note python-lz4 was using an ancient version of lz4. So there could be differences.
Reviewed By: DurhamG
Differential Revision: D13528200
fbshipit-source-id: 6be1c1dd71f57d40dcffcc8d212d40a853583254
Summary:
The `pybuf` provides a way to read `bytes`, `bytearray`, some `buffer` types in
a zero-copy way. The main benefit is to use same code to support different
input types. It's copied to a couple of places. Let's move it to `cpython-ext`.
Reviewed By: DurhamG
Differential Revision: D13516206
fbshipit-source-id: f58881c4bfe651a6fdb84cf317a74c3c8d7a4961
Summary: Make it possible to write content directly into a PyBytes buffer.
Reviewed By: DurhamG
Differential Revision: D13528202
fbshipit-source-id: 8c0a4ed030439a8dc40cdfbd72b1f6734a8b2036
Summary:
This allows decompressing into a pre-allocated buffer. After some experiments,
it seems `bytearray` will just break too many things, ex:
- bytearray is not hashable
- bytearray[index] returns an int
- a = bytearray('x'); b = a; b += '3' # will mutate 'a'
- ''.join([bytearray('')]) will raise TypeError
Therefore we have to use zero-copy `bytes` instead, which is less elegent. But
this API change is a step forward.
Reviewed By: DurhamG
Differential Revision: D13528201
fbshipit-source-id: 1cfaf5d55efdc0d6c0df85df9960fe9682028b08
Summary:
I need to convert `Vec<u8>` to a Python object in a zero-copy way for rustlz4
performacne.
Assuming Python and Rust use the same memory allocator, it's possible to transfer
the control of a malloc-ed pointer from Rust to Python. Use this to implement
zero-copy. PyByteArrayObject is chosen because its struct contains such a pointer.
PyBytes cannot be used as it embeds the bytes, without using a pointer.
Sadly there are no CPython APIs to do this job. So we have to write to the raw
structures. That means the code will crash if python is replaced by
python-debug (due to Python object header change). However, that seems less an
issue given the performance wins. If python-debug does become a problem, we can
try vendoring libpython directly.
I didn't implement a feature-rich `PyByteArray` Rust object. It's not easy to
do so outside the cpython crate. Most helper macros to declare types cannot be
reused, because they refer to `::python`, which is not available in the current
crate.
Reviewed By: DurhamG
Differential Revision: D13516209
fbshipit-source-id: 9aa089b309beb71d4d21f6c63fcb97dbc798b5f8
Summary:
This gives some sense about how fast it is.
Background: I was trying to get rid of python-lz4, by exposing this to Python.
However, I noticed it's 10x slower than python-lz4. Therefore I added some
benchmark here to test if it's the wrapper or the Rust lz4 code.
It does not seem to be this crate:
```
# Pure Rust
compress (100M) 77.170 ms
decompress (~100M) 67.043 ms
# python-lz4
In [1]: import lz4, os
In [2]: b=os.urandom(100000000);
In [3]: %timeit lz4.compress(b)
10 loops, best of 3: 87.4 ms per loop
```
Reviewed By: DurhamG
Differential Revision: D13516205
fbshipit-source-id: f55f94bbecc3b49667ed12174f7000b1aa29e7c4
Summary:
This exposes the underlying lookup functions from `Index`.
Alternatively we can allow access to `Index` and provide an `iter_started_from`
method on `Log` which takes a raw offset. I have been trying to avoid exposing
raw offsets in public interfaces, as they would change after `flush()` and cause
problems.
Reviewed By: markbt
Differential Revision: D13498303
fbshipit-source-id: 8b00a2a36a9383e3edb6fd7495a005bc985fd461
Summary:
This is the missing API before `indexedlog::Index` can fit in the
`changelog.partialmatch` case. It's actually more flexible as it can provide
some example commit hashes while the existing revlog.c or radixbuf
implementation just error out saying "ambiguous prefix".
It can be also "abused" for the semantics of sorted "sub-keys". By replace
"key" with "key + subkey" when inserting to the index. Looking up using "key"
would return a lazy result list (`PrefixIter`) sorted by "subkey". Note:
the radix tree is NOT efficient (both in time and space) when there are common
prefixes. So this use-case needs to be careful.
Reviewed By: markbt
Differential Revision: D13498301
fbshipit-source-id: 637856ebd761734d68b20c15866424b1d4518ad6
Summary: This will be used in prefix lookups.
Reviewed By: markbt
Differential Revision: D13498300
fbshipit-source-id: 3db7a21d6f35a18699d9dc3a0eca71a5410e0e61
Summary:
It makes testing duplicated - now `cargo test` would try running tests on 2 entry points:
lib.rs and indexedlog_dump.rs. Move it to a separate crate to solve the issue.
Reviewed By: markbt
Differential Revision: D13498266
fbshipit-source-id: 8abf07c1272dfa825ec7701fd8ea9e0d1310ec5f
Summary: `write!` result needs to be used.
Reviewed By: markbt
Differential Revision: D13471967
fbshipit-source-id: d48752bcac05dd33b112679d7faf990eb8ddd651
Summary: The former is deprecated and thus compiling revisionstore shows many warnings.
Reviewed By: markbt
Differential Revision: D13379278
fbshipit-source-id: d4b4662a1ad00997de4c46274deaf22f48487328
Summary:
Adds a new crate `cpython-result`, which provides a `ResultExt` trait, which
extends the failure `Result` type to allow coversion to `PyResult` by
converting the error to an appropriate Python Exception.
Reviewed By: quark-zju
Differential Revision: D12980782
fbshipit-source-id: 44a63d31f9ecf2f77efa3b37c68f9a99eaf6d6fa
Summary:
The mutationstore is a new store for recording records of commit mutations for
commits that are not in the local repository.
It uses an indexedlog to store the data. Each mutation entry corresponds to
the information the mutation that led to the creation of a particular commit,
which is recorded as the successor in the entry.
Entries can come from three possible places:
* `Commit` metadata for a commit not available locally
* `Obsmarkers` for repos that have been migrated from evolution tracking
* `Synthetic` for entries created synthetically, e.g. by a pullcreatemarkers
implementation.
The other commits referred to in an entry must predate the successor commit.
For entries that originated from commits, this is ensured, as the successor
commit hash includes the other commit hashes. For other entry types, it is
an error to refer to later commits, and any entry that causes a cycle will
be ignored.
Reviewed By: quark-zju
Differential Revision: D12980773
fbshipit-source-id: 040d3f7369a113e710ed8c9f61fabec6c5ec9258
Summary:
The derived debug for Node prints out each byte as a decimal number. Instead,
make the Debug output for nodes look like `Node("hexstring")`.
Reviewed By: DurhamG
Differential Revision: D12980775
fbshipit-source-id: 042cbf6eade8403759684969e1f69f7f4e335582
Summary:
Add a utility function for tests to generate a vector of random nodes. This
will be used in future tests.
Reviewed By: DurhamG
Differential Revision: D12980784
fbshipit-source-id: 73fc8643503e11a46a845671df94c912a5e49d23
Summary:
Add traits that extend `std::io::Read` and `std::io::Write` to implement new
`read_node` and `write_node` methods, allowing simple reading and writing of
binary nodes from and to streams.
Reviewed By: DurhamG
Differential Revision: D12980778
fbshipit-source-id: fc6751cd43a1693a5a5a3ac93aea74aec5fda4fe
Summary:
The future of mercurial is rust, and one of the missing piece is repacking of data/history packs. For now, let's implement a very basic packing strategy that just pulls all the packs into one, with one small optimization that puts all the delta chains close together in the output file.
At first, it's expected that this code will be driven by the existing python code, but more and more will be done in rust as time goes.
Reviewed By: DurhamG
Differential Revision: D13363853
fbshipit-source-id: ad1ac2039e1732f7141d99abf7f01804a9bde097
Summary:
Add a new entry type - INLINE_LEAF, which embeds the EXT_KEY and LINK entries
to save space.
The index size for referred keys is significantly reduced with little overhead:
index insertion (owned key) 3.732 ms
index insertion (referred key) 3.604 ms
index flush 11.868 ms
index lookup (memory) 1.159 ms
index lookup (disk, no verify) 2.175 ms
index lookup (disk, verified) 4.303 ms
index size (5M owned keys) 216626039
index size (5M referred keys) 96616431
11.87s user 2.96s system 98% cpu 15.107 total
The breakdown of the "5M referred keys" size is:
type count bytes
radixes 1729472 33835772
inline_leafs 5000000 62780651
There are no other kinds of entries stored.
Previously, the index size of referred keys is:
index size (5M referred keys) 136245815 bytes
So it's 136MB -> 96MB, 40% decrease.
Reviewed By: DurhamG
Differential Revision: D13036801
fbshipit-source-id: 27e68e4b6c332c1dc419abc6aba69271952e4b3d
Summary:
Replace the 20-byte "jump table" with 3-byte "flag + bitmap". This saves space
for indexes less than 4GB. There are some reserved bits in the "flag" so if we
run into space issues when indexes are larger than 4GB, we can try adding
6-byte integer, or VLQ back without breaking backwards-compatibility.
It seems to hurt flush performance a bit, because we have to scan the child
array twice. However, lookup (the most important performance) does not change
much. And the index is more compact.
After:
index flush 19.644 ms
index lookup (disk, no verify) 2.220 ms
index lookup (disk, verified) 4.067 ms
index size (5M owned keys) 216626039 bytes
index size (5M referred keys) 136245815 bytes
Before:
index flush 16.764 ms
index lookup (disk, no verify) 2.205 ms
index lookup (disk, verified) 4.030 ms
index size (5M owned keys) 240838647 bytes
index size (5M referred keys) 160458423 bytes
For the "referred key" case, it's 160->136MB, 17% decrease.
A detailed break down of components of index is:
After:
type count bytes (using owned keys)
radixes 1729472 33835772
links 5000000 27886336
leafs 5000000 44629384
keys 5000000 110000000
type count bytes (using referred keys)
radixes 1729472 33835772
links 5000000 27886336
leafs 5000000 44629384
ext_keys 5000000 29894315
Before:
type count bytes (using owned keys)
radixes 1729472 58048380
links 5000000 27886336
leafs 5000000 44903923
keys 5000000 110000000
type count bytes (using referred keys)
radixes 1729472 58048380
links 5000000 27886336
leafs 5000000 44629384
ext_keys 5000000 29894315
Leaf nodes are taking too much space. It seems the next big optimization might
be inlining ext_keys into leafs.
Reviewed By: DurhamG, markbt
Differential Revision: D13028196
fbshipit-source-id: 6043b16fd67a497eb52d20a17e153fcba5cb3e81
Summary:
Since the size test only runs once, we can use a larger number of keys. This is
closer to some production use-cases.
`cargo bench size` shows:
index size (5M owned keys) 240838647
index size (5M referred keys) 160458423
It currently uses 32 bytes per key for 5M referred keys.
Reviewed By: markbt
Differential Revision: D13027880
fbshipit-source-id: 726f5fb2da056e77ab93d82fda9f1afa500d0a8d
Summary:
Add benchmarks about index sizes, and a benchmark of insertion using key
references.
An example `cargo bench` result running on my devserver looks like:
index insertion (owned key) 3.551 ms
index insertion (referred key) 3.713 ms
index flush 20.648 ms
index lookup (memory) 1.087 ms
index lookup (disk, no verify) 2.041 ms
index lookup (disk, verified) 4.347 ms
index size (owned key) 886010
index size (referred key) 534298
Reviewed By: markbt
Differential Revision: D13027879
fbshipit-source-id: 70644c504026ffee2122d857d5035f5b7eea4f42
Summary:
For checksum values like xxhash, there is no benefit using big endian. Switch
to little endian so it's slightly slightly faster on the major platforms we
care about.
This is a breaking change. However, the format is not used in production yet.
So there is no migration code.
Reviewed By: markbt
Differential Revision: D13015465
fbshipit-source-id: ca83d19b3328370d089b03a33e848e64b728ef2a
Summary:
Previously, the format of an Log entry is hard-coded - length, xxhash, and
content. The xxhash always takes 8 bytes.
For small (ex. 40-byte) entries, xxhash32 is actually faster and takes less
disk space.
Introduce the "entry flags" concept so we can store some metadata about what
checksum function to use. The concept could be potentially used to support
other new format changes at per entry level in the future.
As we're here, also support data without checksums. That can be useful for
content with its own checksum, like a blob store with its own SHA1 integrity
check.
Performance-wise, log insertion is slower (but the majority insertaion overhead
would be on the index part), iteration is a little bit faster, perhaps because
the log can use less data.
Before:
log insertion 15.874 ms
log iteration (memory) 6.778 ms
log iteration (disk) 6.830 ms
After:
log insertion 18.114 ms
log iteration (memory) 6.403 ms
log iteration (disk) 6.307 ms
Reviewed By: DurhamG, markbt
Differential Revision: D13051386
fbshipit-source-id: 629c251633ecf85058ee7c3ce7a9f576dfac7bdf
Summary:
Xxhash result won't usually have leading zeros. So VLQ encoding is not an
efficient choice. Use non-VLQ encoding instead.
Performance wise, this is noticably faster than before:
log insertion 14.161 ms
log insertion with index 102.724 ms
log flush 11.336 ms
log iteration (memory) 6.351 ms
log iteration (disk) 7.922 ms
10.18s user 3.66s system 97% cpu 14.218 total
log insertion 13.377 ms
log insertion with index 97.422 ms
log flush 11.792 ms
log iteration (memory) 6.890 ms
log iteration (disk) 7.139 ms
10.20s user 3.56s system 97% cpu 14.117 total
log insertion 14.573 ms
log insertion with index 94.216 ms
log flush 18.993 ms
log iteration (memory) 7.867 ms
log iteration (disk) 7.567 ms
9.85s user 3.73s system 96% cpu 14.073 total
log insertion 15.526 ms
log insertion with index 98.868 ms
log flush 19.600 ms
log iteration (memory) 7.533 ms
log iteration (disk) 7.150 ms
10.13s user 4.02s system 96% cpu 14.647 total
log insertion 14.629 ms
log insertion with index 100.449 ms
log flush 20.997 ms
log iteration (memory) 7.299 ms
log iteration (disk) 7.518 ms
10.14s user 3.65s system 96% cpu 14.274 total
This is a format-breaking change. Fortunately we haven't really use the old
format in production yet.
Reviewed By: DurhamG, markbt
Differential Revision: D13015463
fbshipit-source-id: 6e7e4f7a845ea8dbf0904b3902740b65cc7467d5
Summary:
Some simple benchmark for "log". The initial result running from my devserver
looks like:
log insertion 33.146 ms
log insertion with index 106.449 ms
log flush 9.623 ms
log iteration (memory) 10.644 ms
log iteration (disk) 11.517 ms
13.75s user 3.61s system 97% cpu 17.778 total
log insertion 27.906 ms
log insertion with index 107.683 ms
log flush 19.204 ms
log iteration (memory) 10.239 ms
log iteration (disk) 11.118 ms
12.89s user 3.55s system 97% cpu 16.924 total
log insertion 31.645 ms
log insertion with index 109.403 ms
log flush 9.416 ms
log iteration (memory) 10.226 ms
log iteration (disk) 10.757 ms
13.07s user 3.02s system 97% cpu 16.423 total
log insertion 31.848 ms
log insertion with index 109.332 ms
log flush 18.345 ms
log iteration (memory) 10.709 ms
log iteration (disk) 11.346 ms
13.12s user 3.70s system 97% cpu 17.276 total
log insertion 29.665 ms
log insertion with index 106.041 ms
log flush 16.159 ms
log iteration (memory) 10.367 ms
log iteration (disk) 11.110 ms
12.99s user 3.27s system 97% cpu 16.717 total
Reviewed By: markbt
Differential Revision: D13015464
fbshipit-source-id: 035fee6c8b6d0bea4cfe194eed3d58ba4b5ebcb8
Summary:
An upcoming diff will need the ability to iterate over all the keys in
the store. So let's expose that functionality.
Reviewed By: quark-zju
Differential Revision: D13062575
fbshipit-source-id: a173fcdbbf44e2d3f09f7229266cca6f3e67944b
Summary:
Introduces a nodemap structure that stores the mapping between two
nodes with bidirectional indexes.
Reviewed By: quark-zju
Differential Revision: D13047698
fbshipit-source-id: 967bf4b26a4b57e4fa2421a342edb21d3a5adbf6
Summary:
You can currently iterate over indexlog entries, but there's no way to
iterate over the keys without keeping a copy of the index function with you.
Let's add a key iterator function.
Reviewed By: quark-zju
Differential Revision: D13010744
fbshipit-source-id: 1fcaf959ae82417e5cbafae7c1927c3ae8f8e76a