16e652d70Sopenharmony_ci//! [![github]](https://github.com/dtolnay/unicode-ident) [![crates-io]](https://crates.io/crates/unicode-ident) [![docs-rs]](https://docs.rs/unicode-ident)
26e652d70Sopenharmony_ci//!
36e652d70Sopenharmony_ci//! [github]: https://img.shields.io/badge/github-8da0cb?style=for-the-badge&labelColor=555555&logo=github
46e652d70Sopenharmony_ci//! [crates-io]: https://img.shields.io/badge/crates.io-fc8d62?style=for-the-badge&labelColor=555555&logo=rust
56e652d70Sopenharmony_ci//! [docs-rs]: https://img.shields.io/badge/docs.rs-66c2a5?style=for-the-badge&labelColor=555555&logo=docs.rs
66e652d70Sopenharmony_ci//!
76e652d70Sopenharmony_ci//! <br>
86e652d70Sopenharmony_ci//!
96e652d70Sopenharmony_ci//! Implementation of [Unicode Standard Annex #31][tr31] for determining which
106e652d70Sopenharmony_ci//! `char` values are valid in programming language identifiers.
116e652d70Sopenharmony_ci//!
126e652d70Sopenharmony_ci//! [tr31]: https://www.unicode.org/reports/tr31/
136e652d70Sopenharmony_ci//!
146e652d70Sopenharmony_ci//! This crate is a better optimized implementation of the older `unicode-xid`
156e652d70Sopenharmony_ci//! crate. This crate uses less static storage, and is able to classify both
166e652d70Sopenharmony_ci//! ASCII and non-ASCII codepoints with better performance, 2&ndash;10&times;
176e652d70Sopenharmony_ci//! faster than `unicode-xid`.
186e652d70Sopenharmony_ci//!
196e652d70Sopenharmony_ci//! <br>
206e652d70Sopenharmony_ci//!
216e652d70Sopenharmony_ci//! ## Comparison of performance
226e652d70Sopenharmony_ci//!
236e652d70Sopenharmony_ci//! The following table shows a comparison between five Unicode identifier
246e652d70Sopenharmony_ci//! implementations.
256e652d70Sopenharmony_ci//!
266e652d70Sopenharmony_ci//! - `unicode-ident` is this crate;
276e652d70Sopenharmony_ci//! - [`unicode-xid`] is a widely used crate run by the "unicode-rs" org;
286e652d70Sopenharmony_ci//! - `ucd-trie` and `fst` are two data structures supported by the
296e652d70Sopenharmony_ci//!   [`ucd-generate`] tool;
306e652d70Sopenharmony_ci//! - [`roaring`] is a Rust implementation of Roaring bitmap.
316e652d70Sopenharmony_ci//!
326e652d70Sopenharmony_ci//! The *static storage* column shows the total size of `static` tables that the
336e652d70Sopenharmony_ci//! crate bakes into your binary, measured in 1000s of bytes.
346e652d70Sopenharmony_ci//!
356e652d70Sopenharmony_ci//! The remaining columns show the **cost per call** to evaluate whether a
366e652d70Sopenharmony_ci//! single `char` has the XID\_Start or XID\_Continue Unicode property,
376e652d70Sopenharmony_ci//! comparing across different ratios of ASCII to non-ASCII codepoints in the
386e652d70Sopenharmony_ci//! input data.
396e652d70Sopenharmony_ci//!
406e652d70Sopenharmony_ci//! [`unicode-xid`]: https://github.com/unicode-rs/unicode-xid
416e652d70Sopenharmony_ci//! [`ucd-generate`]: https://github.com/BurntSushi/ucd-generate
426e652d70Sopenharmony_ci//! [`roaring`]: https://github.com/RoaringBitmap/roaring-rs
436e652d70Sopenharmony_ci//!
446e652d70Sopenharmony_ci//! | | static storage | 0% nonascii | 1% | 10% | 100% nonascii |
456e652d70Sopenharmony_ci//! |---|---|---|---|---|---|
466e652d70Sopenharmony_ci//! | **`unicode-ident`** | 9.75 K | 0.96 ns | 0.95 ns | 1.09 ns | 1.55 ns |
476e652d70Sopenharmony_ci//! | **`unicode-xid`** | 11.34 K | 1.88 ns | 2.14 ns | 3.48 ns | 15.63 ns |
486e652d70Sopenharmony_ci//! | **`ucd-trie`** | 9.95 K | 1.29 ns | 1.28 ns | 1.36 ns | 2.15 ns |
496e652d70Sopenharmony_ci//! | **`fst`** | 133 K | 55.1 ns | 54.9 ns | 53.2 ns | 28.5 ns |
506e652d70Sopenharmony_ci//! | **`roaring`** | 66.1 K | 2.78 ns | 3.09 ns | 3.37 ns | 4.70 ns |
516e652d70Sopenharmony_ci//!
526e652d70Sopenharmony_ci//! Source code for the benchmark is provided in the *bench* directory of this
536e652d70Sopenharmony_ci//! repo and may be repeated by running `cargo criterion`.
546e652d70Sopenharmony_ci//!
556e652d70Sopenharmony_ci//! <br>
566e652d70Sopenharmony_ci//!
576e652d70Sopenharmony_ci//! ## Comparison of data structures
586e652d70Sopenharmony_ci//!
596e652d70Sopenharmony_ci//! #### unicode-xid
606e652d70Sopenharmony_ci//!
616e652d70Sopenharmony_ci//! They use a sorted array of character ranges, and do a binary search to look
626e652d70Sopenharmony_ci//! up whether a given character lands inside one of those ranges.
636e652d70Sopenharmony_ci//!
646e652d70Sopenharmony_ci//! ```rust
656e652d70Sopenharmony_ci//! # const _: &str = stringify! {
666e652d70Sopenharmony_ci//! static XID_Continue_table: [(char, char); 763] = [
676e652d70Sopenharmony_ci//!     ('\u{30}', '\u{39}'),  // 0-9
686e652d70Sopenharmony_ci//!     ('\u{41}', '\u{5a}'),  // A-Z
696e652d70Sopenharmony_ci//! # "
706e652d70Sopenharmony_ci//!     …
716e652d70Sopenharmony_ci//! # "
726e652d70Sopenharmony_ci//!     ('\u{e0100}', '\u{e01ef}'),
736e652d70Sopenharmony_ci//! ];
746e652d70Sopenharmony_ci//! # };
756e652d70Sopenharmony_ci//! ```
766e652d70Sopenharmony_ci//!
776e652d70Sopenharmony_ci//! The static storage used by this data structure scales with the number of
786e652d70Sopenharmony_ci//! contiguous ranges of identifier codepoints in Unicode. Every table entry
796e652d70Sopenharmony_ci//! consumes 8 bytes, because it consists of a pair of 32-bit `char` values.
806e652d70Sopenharmony_ci//!
816e652d70Sopenharmony_ci//! In some ranges of the Unicode codepoint space, this is quite a sparse
826e652d70Sopenharmony_ci//! representation &ndash; there are some ranges where tens of thousands of
836e652d70Sopenharmony_ci//! adjacent codepoints are all valid identifier characters. In other places,
846e652d70Sopenharmony_ci//! the representation is quite inefficient. A characater like `µ` (U+00B5)
856e652d70Sopenharmony_ci//! which is surrounded by non-identifier codepoints consumes 64 bits in the
866e652d70Sopenharmony_ci//! table, while it would be just 1 bit in a dense bitmap.
876e652d70Sopenharmony_ci//!
886e652d70Sopenharmony_ci//! On a system with 64-byte cache lines, binary searching the table touches 7
896e652d70Sopenharmony_ci//! cache lines on average. Each cache line fits only 8 table entries.
906e652d70Sopenharmony_ci//! Additionally, the branching performed during the binary search is probably
916e652d70Sopenharmony_ci//! mostly unpredictable to the branch predictor.
926e652d70Sopenharmony_ci//!
936e652d70Sopenharmony_ci//! Overall, the crate ends up being about 10&times; slower on non-ASCII input
946e652d70Sopenharmony_ci//! compared to the fastest crate.
956e652d70Sopenharmony_ci//!
966e652d70Sopenharmony_ci//! A potential improvement would be to pack the table entries more compactly.
976e652d70Sopenharmony_ci//! Rust's `char` type is a 21-bit integer padded to 32 bits, which means every
986e652d70Sopenharmony_ci//! table entry is holding 22 bits of wasted space, adding up to 3.9 K. They
996e652d70Sopenharmony_ci//! could instead fit every table entry into 6 bytes, leaving out some of the
1006e652d70Sopenharmony_ci//! padding, for a 25% improvement in space used. With some cleverness it may be
1016e652d70Sopenharmony_ci//! possible to fit in 5 bytes or even 4 bytes by storing a low char and an
1026e652d70Sopenharmony_ci//! extent, instead of low char and high char. I don't expect that performance
1036e652d70Sopenharmony_ci//! would improve much but this could be the most efficient for space across all
1046e652d70Sopenharmony_ci//! the libraries, needing only about 7 K to store.
1056e652d70Sopenharmony_ci//!
1066e652d70Sopenharmony_ci//! #### ucd-trie
1076e652d70Sopenharmony_ci//!
1086e652d70Sopenharmony_ci//! Their data structure is a compressed trie set specifically tailored for
1096e652d70Sopenharmony_ci//! Unicode codepoints. The design is credited to Raph Levien in
1106e652d70Sopenharmony_ci//! [rust-lang/rust#33098].
1116e652d70Sopenharmony_ci//!
1126e652d70Sopenharmony_ci//! [rust-lang/rust#33098]: https://github.com/rust-lang/rust/pull/33098
1136e652d70Sopenharmony_ci//!
1146e652d70Sopenharmony_ci//! ```rust
1156e652d70Sopenharmony_ci//! pub struct TrieSet {
1166e652d70Sopenharmony_ci//!     tree1_level1: &'static [u64; 32],
1176e652d70Sopenharmony_ci//!     tree2_level1: &'static [u8; 992],
1186e652d70Sopenharmony_ci//!     tree2_level2: &'static [u64],
1196e652d70Sopenharmony_ci//!     tree3_level1: &'static [u8; 256],
1206e652d70Sopenharmony_ci//!     tree3_level2: &'static [u8],
1216e652d70Sopenharmony_ci//!     tree3_level3: &'static [u64],
1226e652d70Sopenharmony_ci//! }
1236e652d70Sopenharmony_ci//! ```
1246e652d70Sopenharmony_ci//!
1256e652d70Sopenharmony_ci//! It represents codepoint sets using a trie to achieve prefix compression. The
1266e652d70Sopenharmony_ci//! final states of the trie are embedded in leaves or "chunks", where each
1276e652d70Sopenharmony_ci//! chunk is a 64-bit integer. Each bit position of the integer corresponds to
1286e652d70Sopenharmony_ci//! whether a particular codepoint is in the set or not. These chunks are not
1296e652d70Sopenharmony_ci//! just a compact representation of the final states of the trie, but are also
1306e652d70Sopenharmony_ci//! a form of suffix compression. In particular, if multiple ranges of 64
1316e652d70Sopenharmony_ci//! contiguous codepoints have the same Unicode properties, then they all map to
1326e652d70Sopenharmony_ci//! the same chunk in the final level of the trie.
1336e652d70Sopenharmony_ci//!
1346e652d70Sopenharmony_ci//! Being tailored for Unicode codepoints, this trie is partitioned into three
1356e652d70Sopenharmony_ci//! disjoint sets: tree1, tree2, tree3. The first set corresponds to codepoints
1366e652d70Sopenharmony_ci//! \[0, 0x800), the second \[0x800, 0x10000) and the third \[0x10000,
1376e652d70Sopenharmony_ci//! 0x110000). These partitions conveniently correspond to the space of 1 or 2
1386e652d70Sopenharmony_ci//! byte UTF-8 encoded codepoints, 3 byte UTF-8 encoded codepoints and 4 byte
1396e652d70Sopenharmony_ci//! UTF-8 encoded codepoints, respectively.
1406e652d70Sopenharmony_ci//!
1416e652d70Sopenharmony_ci//! Lookups in this data structure are significantly more efficient than binary
1426e652d70Sopenharmony_ci//! search. A lookup touches either 1, 2, or 3 cache lines based on which of the
1436e652d70Sopenharmony_ci//! trie partitions is being accessed.
1446e652d70Sopenharmony_ci//!
1456e652d70Sopenharmony_ci//! One possible performance improvement would be for this crate to expose a way
1466e652d70Sopenharmony_ci//! to query based on a UTF-8 encoded string, returning the Unicode property
1476e652d70Sopenharmony_ci//! corresponding to the first character in the string. Without such an API, the
1486e652d70Sopenharmony_ci//! caller is required to tokenize their UTF-8 encoded input data into `char`,
1496e652d70Sopenharmony_ci//! hand the `char` into `ucd-trie`, only for `ucd-trie` to undo that work by
1506e652d70Sopenharmony_ci//! converting back into the variable-length representation for trie traversal.
1516e652d70Sopenharmony_ci//!
1526e652d70Sopenharmony_ci//! #### fst
1536e652d70Sopenharmony_ci//!
1546e652d70Sopenharmony_ci//! Uses a [finite state transducer][fst]. This representation is built into
1556e652d70Sopenharmony_ci//! [ucd-generate] but I am not aware of any advantage over the `ucd-trie`
1566e652d70Sopenharmony_ci//! representation. In particular `ucd-trie` is optimized for storing Unicode
1576e652d70Sopenharmony_ci//! properties while `fst` is not.
1586e652d70Sopenharmony_ci//!
1596e652d70Sopenharmony_ci//! [fst]: https://github.com/BurntSushi/fst
1606e652d70Sopenharmony_ci//! [ucd-generate]: https://github.com/BurntSushi/ucd-generate
1616e652d70Sopenharmony_ci//!
1626e652d70Sopenharmony_ci//! As far as I can tell, the main thing that causes `fst` to have large size
1636e652d70Sopenharmony_ci//! and slow lookups for this use case relative to `ucd-trie` is that it does
1646e652d70Sopenharmony_ci//! not specialize for the fact that only 21 of the 32 bits in a `char` are
1656e652d70Sopenharmony_ci//! meaningful. There are some dense arrays in the structure with large ranges
1666e652d70Sopenharmony_ci//! that could never possibly be used.
1676e652d70Sopenharmony_ci//!
1686e652d70Sopenharmony_ci//! #### roaring
1696e652d70Sopenharmony_ci//!
1706e652d70Sopenharmony_ci//! This crate is a pure-Rust implementation of [Roaring Bitmap], a data
1716e652d70Sopenharmony_ci//! structure designed for storing sets of 32-bit unsigned integers.
1726e652d70Sopenharmony_ci//!
1736e652d70Sopenharmony_ci//! [Roaring Bitmap]: https://roaringbitmap.org/about/
1746e652d70Sopenharmony_ci//!
1756e652d70Sopenharmony_ci//! Roaring bitmaps are compressed bitmaps which tend to outperform conventional
1766e652d70Sopenharmony_ci//! compressed bitmaps such as WAH, EWAH or Concise. In some instances, they can
1776e652d70Sopenharmony_ci//! be hundreds of times faster and they often offer significantly better
1786e652d70Sopenharmony_ci//! compression.
1796e652d70Sopenharmony_ci//!
1806e652d70Sopenharmony_ci//! In this use case the performance was reasonably competitive but still
1816e652d70Sopenharmony_ci//! substantially slower than the Unicode-optimized crates. Meanwhile the
1826e652d70Sopenharmony_ci//! compression was significantly worse, requiring 6&times; as much storage for
1836e652d70Sopenharmony_ci//! the data structure.
1846e652d70Sopenharmony_ci//!
1856e652d70Sopenharmony_ci//! I also benchmarked the [`croaring`] crate which is an FFI wrapper around the
1866e652d70Sopenharmony_ci//! C reference implementation of Roaring Bitmap. This crate was consistently
1876e652d70Sopenharmony_ci//! about 15% slower than pure-Rust `roaring`, which could just be FFI overhead.
1886e652d70Sopenharmony_ci//! I did not investigate further.
1896e652d70Sopenharmony_ci//!
1906e652d70Sopenharmony_ci//! [`croaring`]: https://crates.io/crates/croaring
1916e652d70Sopenharmony_ci//!
1926e652d70Sopenharmony_ci//! #### unicode-ident
1936e652d70Sopenharmony_ci//!
1946e652d70Sopenharmony_ci//! This crate is most similar to the `ucd-trie` library, in that it's based on
1956e652d70Sopenharmony_ci//! bitmaps stored in the leafs of a trie representation, achieving both prefix
1966e652d70Sopenharmony_ci//! compression and suffix compression.
1976e652d70Sopenharmony_ci//!
1986e652d70Sopenharmony_ci//! The key differences are:
1996e652d70Sopenharmony_ci//!
2006e652d70Sopenharmony_ci//! - Uses a single 2-level trie, rather than 3 disjoint partitions of different
2016e652d70Sopenharmony_ci//!   depth each.
2026e652d70Sopenharmony_ci//! - Uses significantly larger chunks: 512 bits rather than 64 bits.
2036e652d70Sopenharmony_ci//! - Compresses the XID\_Start and XID\_Continue properties together
2046e652d70Sopenharmony_ci//!   simultaneously, rather than duplicating identical trie leaf chunks across
2056e652d70Sopenharmony_ci//!   the two.
2066e652d70Sopenharmony_ci//!
2076e652d70Sopenharmony_ci//! The following diagram show the XID\_Start and XID\_Continue Unicode boolean
2086e652d70Sopenharmony_ci//! properties in uncompressed form, in row-major order:
2096e652d70Sopenharmony_ci//!
2106e652d70Sopenharmony_ci//! <table>
2116e652d70Sopenharmony_ci//! <tr><th>XID_Start</th><th>XID_Continue</th></tr>
2126e652d70Sopenharmony_ci//! <tr>
2136e652d70Sopenharmony_ci//! <td><img alt="XID_Start bitmap" width="256" src="https://user-images.githubusercontent.com/1940490/168647353-c6eeb922-afec-49b2-9ef5-c03e9d1e0760.png"></td>
2146e652d70Sopenharmony_ci//! <td><img alt="XID_Continue bitmap" width="256" src="https://user-images.githubusercontent.com/1940490/168647367-f447cca7-2362-4d7d-8cd7-d21c011d329b.png"></td>
2156e652d70Sopenharmony_ci//! </tr>
2166e652d70Sopenharmony_ci//! </table>
2176e652d70Sopenharmony_ci//!
2186e652d70Sopenharmony_ci//! Uncompressed, these would take 140 K to store, which is beyond what would be
2196e652d70Sopenharmony_ci//! reasonable. However, as you can see there is a large degree of similarity
2206e652d70Sopenharmony_ci//! between the two bitmaps and across the rows, which lends well to
2216e652d70Sopenharmony_ci//! compression.
2226e652d70Sopenharmony_ci//!
2236e652d70Sopenharmony_ci//! This crate stores one 512-bit "row" of the above bitmaps in the leaf level
2246e652d70Sopenharmony_ci//! of a trie, and a single additional level to index into the leafs. It turns
2256e652d70Sopenharmony_ci//! out there are 124 unique 512-bit chunks across the two bitmaps so 7 bits are
2266e652d70Sopenharmony_ci//! sufficient to index them.
2276e652d70Sopenharmony_ci//!
2286e652d70Sopenharmony_ci//! The chunk size of 512 bits is selected as the size that minimizes the total
2296e652d70Sopenharmony_ci//! size of the data structure. A smaller chunk, like 256 or 128 bits, would
2306e652d70Sopenharmony_ci//! achieve better deduplication but require a larger index. A larger chunk
2316e652d70Sopenharmony_ci//! would increase redundancy in the leaf bitmaps. 512 bit chunks are the
2326e652d70Sopenharmony_ci//! optimum for total size of the index plus leaf bitmaps.
2336e652d70Sopenharmony_ci//!
2346e652d70Sopenharmony_ci//! In fact since there are only 124 unique chunks, we can use an 8-bit index
2356e652d70Sopenharmony_ci//! with a spare bit to index at the half-chunk level. This achieves an
2366e652d70Sopenharmony_ci//! additional 8.5% compression by eliminating redundancies between the second
2376e652d70Sopenharmony_ci//! half of any chunk and the first half of any other chunk. Note that this is
2386e652d70Sopenharmony_ci//! not the same as using chunks which are half the size, because it does not
2396e652d70Sopenharmony_ci//! necessitate raising the size of the trie's first level.
2406e652d70Sopenharmony_ci//!
2416e652d70Sopenharmony_ci//! In contrast to binary search or the `ucd-trie` crate, performing lookups in
2426e652d70Sopenharmony_ci//! this data structure is straight-line code with no need for branching.
2436e652d70Sopenharmony_ci
2446e652d70Sopenharmony_ci#![no_std]
2456e652d70Sopenharmony_ci#![doc(html_root_url = "https://docs.rs/unicode-ident/1.0.8")]
2466e652d70Sopenharmony_ci#![allow(clippy::doc_markdown, clippy::must_use_candidate)]
2476e652d70Sopenharmony_ci
2486e652d70Sopenharmony_ci#[rustfmt::skip]
2496e652d70Sopenharmony_cimod tables;
2506e652d70Sopenharmony_ci
2516e652d70Sopenharmony_ciuse crate::tables::{ASCII_CONTINUE, ASCII_START, CHUNK, LEAF, TRIE_CONTINUE, TRIE_START};
2526e652d70Sopenharmony_ci
2536e652d70Sopenharmony_cipub fn is_xid_start(ch: char) -> bool {
2546e652d70Sopenharmony_ci    if ch.is_ascii() {
2556e652d70Sopenharmony_ci        return ASCII_START.0[ch as usize];
2566e652d70Sopenharmony_ci    }
2576e652d70Sopenharmony_ci    let chunk = *TRIE_START.0.get(ch as usize / 8 / CHUNK).unwrap_or(&0);
2586e652d70Sopenharmony_ci    let offset = chunk as usize * CHUNK / 2 + ch as usize / 8 % CHUNK;
2596e652d70Sopenharmony_ci    unsafe { LEAF.0.get_unchecked(offset) }.wrapping_shr(ch as u32 % 8) & 1 != 0
2606e652d70Sopenharmony_ci}
2616e652d70Sopenharmony_ci
2626e652d70Sopenharmony_cipub fn is_xid_continue(ch: char) -> bool {
2636e652d70Sopenharmony_ci    if ch.is_ascii() {
2646e652d70Sopenharmony_ci        return ASCII_CONTINUE.0[ch as usize];
2656e652d70Sopenharmony_ci    }
2666e652d70Sopenharmony_ci    let chunk = *TRIE_CONTINUE.0.get(ch as usize / 8 / CHUNK).unwrap_or(&0);
2676e652d70Sopenharmony_ci    let offset = chunk as usize * CHUNK / 2 + ch as usize / 8 % CHUNK;
2686e652d70Sopenharmony_ci    unsafe { LEAF.0.get_unchecked(offset) }.wrapping_shr(ch as u32 % 8) & 1 != 0
2696e652d70Sopenharmony_ci}
270