19378243882

Committed 14 Nov 2025 09:34PM UTC coverage: 62.201% (+0.06%) from 62.146%

Build # 19378243882

Build Type

push

github

Committed by

vigna

Commit Message

Back to edition 2021

Run Details

5251 of 8442 relevant lines covered (62.2%)

29314157.11 hits per line

Source File
Press 'n' to go to next uncovered line, 'b' for previous

91.3

/webgraph/src/utils/batch_codec/grouped_gaps.rs

/*
 * SPDX-FileCopyrightText: 2025 Tommaso Fontana
 *
 * SPDX-License-Identifier: Apache-2.0 OR LGPL-2.1-or-later
 */

use super::{BitReader, BitWriter};
use crate::traits::SortedIterator;
use crate::utils::{ArcMmapHelper, MmapHelper, Triple};
use crate::{
    traits::{BitDeserializer, BitSerializer},
    utils::{humanize, BatchCodec},
};

use std::sync::Arc;

use anyhow::{Context, Result};
use dsi_bitstream::prelude::*;
use mmap_rs::MmapFlags;
use rdst::*;

#[derive(Clone, Debug)]
/// A codec for encoding and decoding batches of triples using grouped gap compression.
///
/// This codec encodes triples of the form `(src, dst, label)` by grouping edges
/// with the same source node, and encoding the gaps between consecutive sources
/// and destinations using a specified code (default: gamma). The outdegree
/// (number of edges for each source) is also encoded using the specified code.
///
/// # Type Parameters
///
/// - `S`: Serializer for the labels, implementing [`BitSerializer`] for the label type.
/// - `D`: Deserializer for the labels, implementing [`BitDeserializer`] for the label type.
/// - `OUTDEGREE_CODE`: Code used for encoding outdegrees (default: [ɣ](dsi_bitstream::codes::gamma)).
/// - `SRC_CODE`: Code used for encoding source gaps (default: [ɣ](dsi_bitstream::codes::gamma)).
/// - `DST_CODE`: Code used for encoding destination gaps (default: [ɣ](dsi_bitstream::codes::gamma)).
///
/// # Encoding Format
///
/// 1. The batch length is written using delta coding.
/// 2. For each group of triples with the same source:
///     - The gap from the previous source is encoded.
///     - The outdegree (number of edges for this source) is encoded.
///     - For each destination:
///         - The gap from the previous destination is encoded.
///         - The label is serialized.
///
/// The bit deserializer must be [`Clone`] because we need one for each
/// [`GroupedGapsIterator`], and there are possible scenarios in which the
/// deserializer might be stateful.
pub struct GroupedGapsCodec<
    E: Endianness = NE,
    S: BitSerializer<E, BitWriter<E>> = (),
    D: BitDeserializer<E, BitReader<E>, DeserType = S::SerType> + Clone = (),
    const OUTDEGREE_CODE: usize = { dsi_bitstream::dispatch::code_consts::GAMMA },
    const SRC_CODE: usize = { dsi_bitstream::dispatch::code_consts::GAMMA },
    const DST_CODE: usize = { dsi_bitstream::dispatch::code_consts::DELTA },
> where
    BitReader<E>: BitRead<E>,
    BitWriter<E>: BitWrite<E>,
{
    /// Serializer for the labels.
    pub serializer: S,
    /// Deserializer for the labels.
    pub deserializer: D,

    pub _marker: core::marker::PhantomData<E>,
}

impl<E, S, D, const OUTDEGREE_CODE: usize, const SRC_CODE: usize, const DST_CODE: usize>
    GroupedGapsCodec<E, S, D, OUTDEGREE_CODE, SRC_CODE, DST_CODE>
where
    E: Endianness,
    S: BitSerializer<E, BitWriter<E>> + Send + Sync,
    D: BitDeserializer<E, BitReader<E>, DeserType = S::SerType> + Send + Sync + Clone,
    BitReader<E>: BitRead<E>,
    BitWriter<E>: BitWrite<E>,
{
    /// Creates a new `GroupedGapsCodec` with the given serializer and deserializer.
    pub fn new(serializer: S, deserializer: D) -> Self {
        Self {
            serializer,
            deserializer,
            _marker: core::marker::PhantomData,
        }
    }
}

impl<
        E: Endianness,
        S: BitSerializer<E, BitWriter<E>> + Default,
        D: BitDeserializer<E, BitReader<E>, DeserType = S::SerType> + Clone + Default,
        const OUTDEGREE_CODE: usize,
        const SRC_CODE: usize,
        const DST_CODE: usize,
    > Default for GroupedGapsCodec<E, S, D, OUTDEGREE_CODE, SRC_CODE, DST_CODE>
where
    BitReader<E>: BitRead<E>,
    BitWriter<E>: BitWrite<E>,
{
    fn default() -> Self {
        Self {
            serializer: S::default(),
            deserializer: D::default(),
            _marker: core::marker::PhantomData,
        }
    }
}

#[derive(Debug, Clone, Copy)]
/// Statistics about the encoding performed by
/// [`GapsCodec`](crate::utils::gaps::GapsCodec).
pub struct GroupedGapsStats {
    /// Total number of triples encoded
    pub total_triples: usize,
    /// Number of bits used for outdegrees
    pub outdegree_bits: usize,
    /// Number of bits used for source gaps
    pub src_bits: usize,
    //// Number of bits used for destination gaps
    pub dst_bits: usize,
    /// Number of bits used for labels
    pub labels_bits: usize,
}

impl core::fmt::Display for GroupedGapsStats {
    fn fmt(&self, f: &mut core::fmt::Formatter<'_>) -> core::fmt::Result {
        write!(
            f,
            "outdegree: {}B ({:.3} bits / arc), src: {}B ({:.3} bits / arc), dst: {}B ({:.3} bits / arc), labels: {}B ({:.3} bits / arc)",
            humanize(self.outdegree_bits as f64 / 8.0),
            self.outdegree_bits as f64 / self.total_triples as f64,
            humanize(self.src_bits as f64 / 8.0),
            self.src_bits as f64 / self.total_triples as f64,
            humanize(self.dst_bits as f64 / 8.0),
            self.dst_bits as f64 / self.total_triples as f64,
            humanize(self.labels_bits as f64 / 8.0),
            self.labels_bits as f64 / self.total_triples as f64,
        )
    }
}

impl<E, S, D, const OUTDEGREE_CODE: usize, const SRC_CODE: usize, const DST_CODE: usize> BatchCodec
    for GroupedGapsCodec<E, S, D, OUTDEGREE_CODE, SRC_CODE, DST_CODE>
where
    E: Endianness,
    S: BitSerializer<E, BitWriter<E>> + Send + Sync,
    D: BitDeserializer<E, BitReader<E>, DeserType = S::SerType> + Send + Sync + Clone,
    S::SerType: Send + Sync + Copy + 'static, // needed by radix sort
    BitReader<E>: BitRead<E> + CodesRead<E>,
    BitWriter<E>: BitWrite<E> + CodesWrite<E>,
{
    type Label = S::SerType;
    type DecodedBatch = GroupedGapsIterator<E, D, OUTDEGREE_CODE, SRC_CODE, DST_CODE>;
    type EncodedBatchStats = GroupedGapsStats;

    fn encode_batch(
        &self,
        path: impl AsRef<std::path::Path>,
        batch: &mut [((usize, usize), Self::Label)],
    ) -> Result<(usize, Self::EncodedBatchStats)> {
        let start = std::time::Instant::now();
        Triple::cast_batch_mut(batch).radix_sort_unstable();
        log::debug!("Sorted {} arcs in {:?}", batch.len(), start.elapsed());
        self.encode_sorted_batch(path, batch)
    }

    fn encode_sorted_batch(
        &self,
        path: impl AsRef<std::path::Path>,
        batch: &[((usize, usize), Self::Label)],
    ) -> Result<(usize, Self::EncodedBatchStats)> {
        debug_assert!(Triple::cast_batch(batch).is_sorted(), "Batch is not sorted");
        // create a batch file where to dump
        let file_path = path.as_ref();
        let file = std::io::BufWriter::with_capacity(
            1 << 16,
            std::fs::File::create(file_path).with_context(|| {
                format!(
                    "Could not create BatchIterator temporary file {}",
                    file_path.display()
                )
            })?,
        );
        // create a bitstream to write to the file
        let mut stream = <BufBitWriter<E, _>>::new(<WordAdapter<usize, _>>::new(file));

        // prefix the stream with the length of the batch
        // we use a delta code since it'll be a big number most of the time
        stream
            .write_delta(batch.len() as u64)
            .context("Could not write length")?;

        let mut stats = GroupedGapsStats {
            total_triples: batch.len(),
            outdegree_bits: 0,
            src_bits: 0,
            dst_bits: 0,
            labels_bits: 0,
        };
        // dump the triples to the bitstream
        let mut prev_src = 0;
        let mut i = 0;
        while i < batch.len() {
            let ((src, _), _) = batch[i];
            // write the source gap as gamma
            stats.src_bits += ConstCode::<SRC_CODE>
                .write(&mut stream, (src - prev_src) as _)
                .with_context(|| format!("Could not write {src} after {prev_src}"))?;
            // figure out how many edges have this source
            let outdegree = batch[i..].iter().take_while(|t| t.0 .0 == src).count();
            // write the outdegree
            stats.outdegree_bits += ConstCode::<OUTDEGREE_CODE>
                .write(&mut stream, outdegree as _)
                .with_context(|| format!("Could not write outdegree {outdegree} for {src}"))?;

            // encode the destinations
            let mut prev_dst = 0;
            for _ in 0..outdegree {
                let ((_, dst), label) = &batch[i];
                // write the destination gap as gamma
                stats.dst_bits += ConstCode::<DST_CODE>
                    .write(&mut stream, (dst - prev_dst) as _)
                    .with_context(|| format!("Could not write {dst} after {prev_dst}"))?;
                // write the label
                stats.labels_bits += self
                    .serializer
                    .serialize(label, &mut stream)
                    .context("Could not serialize label")?;
                prev_dst = *dst;
                i += 1;
            }
            prev_src = src;
        }
        // flush the stream and reset the buffer
        stream.flush().context("Could not flush stream")?;

        let total_bits = stats.outdegree_bits + stats.src_bits + stats.dst_bits + stats.labels_bits;
        Ok((total_bits, stats))
    }

    fn decode_batch(&self, path: impl AsRef<std::path::Path>) -> Result<Self::DecodedBatch> {
        // open the file
        let mut stream = <BufBitReader<E, _>>::new(MemWordReader::new(ArcMmapHelper(Arc::new(
            MmapHelper::mmap(
                path.as_ref(),
                MmapFlags::TRANSPARENT_HUGE_PAGES | MmapFlags::SEQUENTIAL,
            )
            .with_context(|| format!("Could not mmap {}", path.as_ref().display()))?,
        ))));

        // read the length of the batch (first value in the stream)
        let len = stream.read_delta().context("Could not read length")? as usize;

        // create the iterator
        Ok(GroupedGapsIterator {
            deserializer: self.deserializer.clone(),
            stream,
            len,
            current: 0,
            src: 0,
            dst_left: 0,
            prev_dst: 0,
        })
    }
}

#[derive(Clone, Debug)]
/// An iterator over triples encoded with gaps, this is returned by [`GroupedGapsCodec`].
pub struct GroupedGapsIterator<
    E: Endianness = NE,
    D: BitDeserializer<E, BitReader<E>> = (),
    const OUTDEGREE_CODE: usize = { dsi_bitstream::dispatch::code_consts::GAMMA },
    const SRC_CODE: usize = { dsi_bitstream::dispatch::code_consts::GAMMA },
    const DST_CODE: usize = { dsi_bitstream::dispatch::code_consts::GAMMA },
> where
    BitReader<E>: BitRead<E>,
    BitWriter<E>: BitWrite<E>,
{
    /// Deserializer for the labels
    deserializer: D,
    /// Bitstream to read from
    stream: BitReader<E>,
    /// Length of the iterator (number of triples)
    len: usize,
    /// Current position in the iterator
    current: usize,
    /// Current source node
    src: usize,
    /// Number of destinations left for the current source
    dst_left: usize,
    /// Previous destination node
    prev_dst: usize,
}

unsafe impl<
        E: Endianness,
        D: BitDeserializer<E, BitReader<E>>,
        const OUTDEGREE_CODE: usize,
        const SRC_CODE: usize,
        const DST_CODE: usize,
    > SortedIterator for GroupedGapsIterator<E, D, OUTDEGREE_CODE, SRC_CODE, DST_CODE>
where
    BitReader<E>: BitRead<E> + CodesRead<E>,
    BitWriter<E>: BitWrite<E> + CodesWrite<E>,
{
}

impl<
        E: Endianness,
        D: BitDeserializer<E, BitReader<E>>,
        const OUTDEGREE_CODE: usize,
        const SRC_CODE: usize,
        const DST_CODE: usize,
    > Iterator for GroupedGapsIterator<E, D, OUTDEGREE_CODE, SRC_CODE, DST_CODE>
where
    BitReader<E>: BitRead<E> + CodesRead<E>,
    BitWriter<E>: BitWrite<E> + CodesWrite<E>,
{
    type Item = ((usize, usize), D::DeserType);
    fn next(&mut self) -> Option<Self::Item> {
        if self.current >= self.len {
            return None;
        }
        if self.dst_left == 0 {
            // read a new source
            let src_gap = ConstCode::<SRC_CODE>.read(&mut self.stream).ok()?;
            self.src += src_gap as usize;
            // read the outdegree
            self.dst_left = ConstCode::<OUTDEGREE_CODE>.read(&mut self.stream).ok()? as usize;
            self.prev_dst = 0;
        }

        let dst_gap = ConstCode::<DST_CODE>.read(&mut self.stream).ok()?;
        let label = self.deserializer.deserialize(&mut self.stream).ok()?;
        self.prev_dst += dst_gap as usize;
        self.current += 1;
        self.dst_left -= 1;
        Some(((self.src, self.prev_dst), label))
    }

    fn size_hint(&self) -> (usize, Option<usize>) {
        (self.len(), Some(self.len()))
    }
}

impl<
        E: Endianness,
        D: BitDeserializer<E, BitReader<E>>,
        const OUTDEGREE_CODE: usize,
        const SRC_CODE: usize,
        const DST_CODE: usize,
    > ExactSizeIterator for GroupedGapsIterator<E, D, OUTDEGREE_CODE, SRC_CODE, DST_CODE>
where
    BitReader<E>: BitRead<E> + CodesRead<E>,
    BitWriter<E>: BitWrite<E> + CodesWrite<E>,
{
    fn len(&self) -> usize {
        self.len - self.current
    }
}

1	/*
2	* SPDX-FileCopyrightText: 2025 Tommaso Fontana
3	*
4	* SPDX-License-Identifier: Apache-2.0 OR LGPL-2.1-or-later
5	*/
6
7	use super::{BitReader, BitWriter};
8	use crate::traits::SortedIterator;
9	use crate::utils::{ArcMmapHelper, MmapHelper, Triple};
10	use crate::{
11	traits::{BitDeserializer, BitSerializer},
12	utils::{humanize, BatchCodec},
13	};
14
15	use std::sync::Arc;
16
17	use anyhow::{Context, Result};
18	use dsi_bitstream::prelude::*;
19	use mmap_rs::MmapFlags;
20	use rdst::*;
21
22	#[derive(Clone, Debug)]
23	/// A codec for encoding and decoding batches of triples using grouped gap compression.
24	///
25	/// This codec encodes triples of the form `(src, dst, label)` by grouping edges
26	/// with the same source node, and encoding the gaps between consecutive sources
27	/// and destinations using a specified code (default: gamma). The outdegree
28	/// (number of edges for each source) is also encoded using the specified code.
29	///
30	/// # Type Parameters
31	///
32	/// - `S`: Serializer for the labels, implementing [`BitSerializer`] for the label type.
33	/// - `D`: Deserializer for the labels, implementing [`BitDeserializer`] for the label type.
34	/// - `OUTDEGREE_CODE`: Code used for encoding outdegrees (default: [ɣ](dsi_bitstream::codes::gamma)).
35	/// - `SRC_CODE`: Code used for encoding source gaps (default: [ɣ](dsi_bitstream::codes::gamma)).
36	/// - `DST_CODE`: Code used for encoding destination gaps (default: [ɣ](dsi_bitstream::codes::gamma)).
37	///
38	/// # Encoding Format
39	///
40	/// 1. The batch length is written using delta coding.
41	/// 2. For each group of triples with the same source:
42	/// - The gap from the previous source is encoded.
43	/// - The outdegree (number of edges for this source) is encoded.
44	/// - For each destination:
45	/// - The gap from the previous destination is encoded.
46	/// - The label is serialized.
47	///
48	/// The bit deserializer must be [`Clone`] because we need one for each
49	/// [`GroupedGapsIterator`], and there are possible scenarios in which the
50	/// deserializer might be stateful.
51	pub struct GroupedGapsCodec<
52	E: Endianness = NE,
53	S: BitSerializer<E, BitWriter<E>> = (),
54	D: BitDeserializer<E, BitReader<E>, DeserType = S::SerType> + Clone = (),
55	const OUTDEGREE_CODE: usize = { dsi_bitstream::dispatch::code_consts::GAMMA },
56	const SRC_CODE: usize = { dsi_bitstream::dispatch::code_consts::GAMMA },
57	const DST_CODE: usize = { dsi_bitstream::dispatch::code_consts::DELTA },
58	> where
59	BitReader<E>: BitRead<E>,
60	BitWriter<E>: BitWrite<E>,
61	{
62	/// Serializer for the labels.
63	pub serializer: S,
64	/// Deserializer for the labels.
65	pub deserializer: D,
66
67	pub _marker: core::marker::PhantomData<E>,
68	}
69
70	impl<E, S, D, const OUTDEGREE_CODE: usize, const SRC_CODE: usize, const DST_CODE: usize>
71	GroupedGapsCodec<E, S, D, OUTDEGREE_CODE, SRC_CODE, DST_CODE>
72	where
73	E: Endianness,
74	S: BitSerializer<E, BitWriter<E>> + Send + Sync,
75	D: BitDeserializer<E, BitReader<E>, DeserType = S::SerType> + Send + Sync + Clone,
76	BitReader<E>: BitRead<E>,
77	BitWriter<E>: BitWrite<E>,
78	{
79	/// Creates a new `GroupedGapsCodec` with the given serializer and deserializer.
80	pub fn new(serializer: S, deserializer: D) -> Self {	×
81	Self {
82	serializer,
83	deserializer,
84	_marker: core::marker::PhantomData,
85	}
86	}
87	}
88
89	impl<
90	E: Endianness,
91	S: BitSerializer<E, BitWriter<E>> + Default,
92	D: BitDeserializer<E, BitReader<E>, DeserType = S::SerType> + Clone + Default,
93	const OUTDEGREE_CODE: usize,
94	const SRC_CODE: usize,
95	const DST_CODE: usize,
96	> Default for GroupedGapsCodec<E, S, D, OUTDEGREE_CODE, SRC_CODE, DST_CODE>
97	where
98	BitReader<E>: BitRead<E>,
99	BitWriter<E>: BitWrite<E>,
100	{
101	fn default() -> Self {	131✔
102	Self {
103	serializer: S::default(),	262✔
104	deserializer: D::default(),	131✔
105	_marker: core::marker::PhantomData,
106	}
107	}
108	}
109
110	#[derive(Debug, Clone, Copy)]
111	/// Statistics about the encoding performed by
112	/// [`GapsCodec`](crate::utils::gaps::GapsCodec).
113	pub struct GroupedGapsStats {
114	/// Total number of triples encoded
115	pub total_triples: usize,
116	/// Number of bits used for outdegrees
117	pub outdegree_bits: usize,
118	/// Number of bits used for source gaps
119	pub src_bits: usize,
120	//// Number of bits used for destination gaps
121	pub dst_bits: usize,
122	/// Number of bits used for labels
123	pub labels_bits: usize,
124	}
125
126	impl core::fmt::Display for GroupedGapsStats {
127	fn fmt(&self, f: &mut core::fmt::Formatter<'_>) -> core::fmt::Result {	32✔
128	write!(	32✔
129	f,	32✔
130	"outdegree: {}B ({:.3} bits / arc), src: {}B ({:.3} bits / arc), dst: {}B ({:.3} bits / arc), labels: {}B ({:.3} bits / arc)",
131	humanize(self.outdegree_bits as f64 / 8.0),	64✔
132	self.outdegree_bits as f64 / self.total_triples as f64,	32✔
133	humanize(self.src_bits as f64 / 8.0),	64✔
134	self.src_bits as f64 / self.total_triples as f64,	32✔
135	humanize(self.dst_bits as f64 / 8.0),	64✔
136	self.dst_bits as f64 / self.total_triples as f64,	32✔
137	humanize(self.labels_bits as f64 / 8.0),	64✔
138	self.labels_bits as f64 / self.total_triples as f64,	32✔
139	)
140	}
141	}
142
143	impl<E, S, D, const OUTDEGREE_CODE: usize, const SRC_CODE: usize, const DST_CODE: usize> BatchCodec
144	for GroupedGapsCodec<E, S, D, OUTDEGREE_CODE, SRC_CODE, DST_CODE>
145	where
146	E: Endianness,
147	S: BitSerializer<E, BitWriter<E>> + Send + Sync,
148	D: BitDeserializer<E, BitReader<E>, DeserType = S::SerType> + Send + Sync + Clone,
149	S::SerType: Send + Sync + Copy + 'static, // needed by radix sort
150	BitReader<E>: BitRead<E> + CodesRead<E>,
151	BitWriter<E>: BitWrite<E> + CodesWrite<E>,
152	{
153	type Label = S::SerType;
154	type DecodedBatch = GroupedGapsIterator<E, D, OUTDEGREE_CODE, SRC_CODE, DST_CODE>;
155	type EncodedBatchStats = GroupedGapsStats;
156
157	fn encode_batch(	459✔
158	&self,
159	path: impl AsRef<std::path::Path>,
160	batch: &mut [((usize, usize), Self::Label)],
161	) -> Result<(usize, Self::EncodedBatchStats)> {
162	let start = std::time::Instant::now();	918✔
163	Triple::cast_batch_mut(batch).radix_sort_unstable();	918✔
164	log::debug!("Sorted {} arcs in {:?}", batch.len(), start.elapsed());	459✔
165	self.encode_sorted_batch(path, batch)	1,836✔
166	}
167
168	fn encode_sorted_batch(	459✔
169	&self,
170	path: impl AsRef<std::path::Path>,
171	batch: &[((usize, usize), Self::Label)],
172	) -> Result<(usize, Self::EncodedBatchStats)> {
173	debug_assert!(Triple::cast_batch(batch).is_sorted(), "Batch is not sorted");	1,377✔
174	// create a batch file where to dump
175	let file_path = path.as_ref();	1,377✔
176	let file = std::io::BufWriter::with_capacity(
177	1 << 16,	459✔
178	std::fs::File::create(file_path).with_context(\|\| {	1,377✔
179	format!(	×
180	"Could not create BatchIterator temporary file {}",	×
181	file_path.display()	×
182	)
183	})?,
184	);
185	// create a bitstream to write to the file
186	let mut stream = <BufBitWriter<E, _>>::new(<WordAdapter<usize, _>>::new(file));	1,836✔
187
188	// prefix the stream with the length of the batch
189	// we use a delta code since it'll be a big number most of the time
190	stream	459✔
191	.write_delta(batch.len() as u64)	918✔
192	.context("Could not write length")?;
193
194	let mut stats = GroupedGapsStats {
195	total_triples: batch.len(),	459✔
196	outdegree_bits: 0,
197	src_bits: 0,
198	dst_bits: 0,
199	labels_bits: 0,
200	};
201	// dump the triples to the bitstream
202	let mut prev_src = 0;	918✔
203	let mut i = 0;	918✔
204	while i < batch.len() {	4,726,460✔
205	let ((src, _), _) = batch[i];	4,725,542✔
206	// write the source gap as gamma
207	stats.src_bits += ConstCode::<SRC_CODE>	2,362,771✔
208	.write(&mut stream, (src - prev_src) as _)	7,088,313✔
209	.with_context(\|\| format!("Could not write {src} after {prev_src}"))?;	2,362,771✔
210	// figure out how many edges have this source
211	let outdegree = batch[i..].iter().take_while(\|t\| t.0 .0 == src).count();	93,710,435✔
212	// write the outdegree
213	stats.outdegree_bits += ConstCode::<OUTDEGREE_CODE>	2,362,771✔
214	.write(&mut stream, outdegree as _)	7,088,313✔
215	.with_context(\|\| format!("Could not write outdegree {outdegree} for {src}"))?;	2,362,771✔
216
217	// encode the destinations
218	let mut prev_dst = 0;	4,725,542✔
219	for _ in 0..outdegree {	2,362,771✔
220	let ((_, dst), label) = &batch[i];	115,757,895✔
221	// write the destination gap as gamma
222	stats.dst_bits += ConstCode::<DST_CODE>	38,585,965✔
223	.write(&mut stream, (dst - prev_dst) as _)	115,757,895✔
224	.with_context(\|\| format!("Could not write {dst} after {prev_dst}"))?;	38,585,965✔
225	// write the label
226	stats.labels_bits += self	38,585,965✔
227	.serializer	38,585,965✔
228	.serialize(label, &mut stream)	115,757,895✔
229	.context("Could not serialize label")?;	38,585,965✔
230	prev_dst = *dst;	38,585,965✔
231	i += 1;	38,585,965✔
232	}
233	prev_src = src;	2,362,771✔
234	}
235	// flush the stream and reset the buffer
236	stream.flush().context("Could not flush stream")?;	1,377✔
237
238	let total_bits = stats.outdegree_bits + stats.src_bits + stats.dst_bits + stats.labels_bits;	918✔
239	Ok((total_bits, stats))	459✔
240	}
241
242	fn decode_batch(&self, path: impl AsRef<std::path::Path>) -> Result<Self::DecodedBatch> {	162✔
243	// open the file
244	let mut stream = <BufBitReader<E, _>>::new(MemWordReader::new(ArcMmapHelper(Arc::new(	648✔
245	MmapHelper::mmap(	162✔
246	path.as_ref(),	324✔
247	MmapFlags::TRANSPARENT_HUGE_PAGES \| MmapFlags::SEQUENTIAL,	162✔
248	)
249	.with_context(\|\| format!("Could not mmap {}", path.as_ref().display()))?,	162✔
250	))));
251
252	// read the length of the batch (first value in the stream)
253	let len = stream.read_delta().context("Could not read length")? as usize;	648✔
254
255	// create the iterator
256	Ok(GroupedGapsIterator {	162✔
257	deserializer: self.deserializer.clone(),	486✔
258	stream,	162✔
259	len,	162✔
260	current: 0,	162✔
261	src: 0,	162✔
262	dst_left: 0,	162✔
263	prev_dst: 0,	162✔
264	})
265	}
266	}
267
268	#[derive(Clone, Debug)]
269	/// An iterator over triples encoded with gaps, this is returned by [`GroupedGapsCodec`].
270	pub struct GroupedGapsIterator<
271	E: Endianness = NE,
272	D: BitDeserializer<E, BitReader<E>> = (),
273	const OUTDEGREE_CODE: usize = { dsi_bitstream::dispatch::code_consts::GAMMA },
274	const SRC_CODE: usize = { dsi_bitstream::dispatch::code_consts::GAMMA },
275	const DST_CODE: usize = { dsi_bitstream::dispatch::code_consts::GAMMA },
276	> where
277	BitReader<E>: BitRead<E>,
278	BitWriter<E>: BitWrite<E>,
279	{
280	/// Deserializer for the labels
281	deserializer: D,
282	/// Bitstream to read from
283	stream: BitReader<E>,
284	/// Length of the iterator (number of triples)
285	len: usize,
286	/// Current position in the iterator
287	current: usize,
288	/// Current source node
289	src: usize,
290	/// Number of destinations left for the current source
291	dst_left: usize,
292	/// Previous destination node
293	prev_dst: usize,
294	}
295
296	unsafe impl<
297	E: Endianness,
298	D: BitDeserializer<E, BitReader<E>>,
299	const OUTDEGREE_CODE: usize,
300	const SRC_CODE: usize,
301	const DST_CODE: usize,
302	> SortedIterator for GroupedGapsIterator<E, D, OUTDEGREE_CODE, SRC_CODE, DST_CODE>
303	where
304	BitReader<E>: BitRead<E> + CodesRead<E>,
305	BitWriter<E>: BitWrite<E> + CodesWrite<E>,
306	{
307	}
308
309	impl<
310	E: Endianness,
311	D: BitDeserializer<E, BitReader<E>>,
312	const OUTDEGREE_CODE: usize,
313	const SRC_CODE: usize,
314	const DST_CODE: usize,
315	> Iterator for GroupedGapsIterator<E, D, OUTDEGREE_CODE, SRC_CODE, DST_CODE>
316	where
317	BitReader<E>: BitRead<E> + CodesRead<E>,
318	BitWriter<E>: BitWrite<E> + CodesWrite<E>,
319	{
320	type Item = ((usize, usize), D::DeserType);
321	fn next(&mut self) -> Option<Self::Item> {	67,942,779✔
322	if self.current >= self.len {	67,942,779✔
323	return None;	151✔
324	}
325	if self.dst_left == 0 {	67,942,628✔
326	// read a new source
327	let src_gap = ConstCode::<SRC_CODE>.read(&mut self.stream).ok()?;	20,350,085✔
328	self.src += src_gap as usize;	4,070,017✔
329	// read the outdegree
330	self.dst_left = ConstCode::<OUTDEGREE_CODE>.read(&mut self.stream).ok()? as usize;	16,280,068✔
331	self.prev_dst = 0;	4,070,017✔
332	}
333
334	let dst_gap = ConstCode::<DST_CODE>.read(&mut self.stream).ok()?;	339,713,140✔
335	let label = self.deserializer.deserialize(&mut self.stream).ok()?;	339,713,140✔
336	self.prev_dst += dst_gap as usize;	67,942,628✔
337	self.current += 1;	67,942,628✔
338	self.dst_left -= 1;	67,942,628✔
339	Some(((self.src, self.prev_dst), label))	67,942,628✔
340	}
341
342	fn size_hint(&self) -> (usize, Option<usize>) {	×
343	(self.len(), Some(self.len()))	×
344	}
345	}
346
347	impl<
348	E: Endianness,
349	D: BitDeserializer<E, BitReader<E>>,
350	const OUTDEGREE_CODE: usize,
351	const SRC_CODE: usize,
352	const DST_CODE: usize,
353	> ExactSizeIterator for GroupedGapsIterator<E, D, OUTDEGREE_CODE, SRC_CODE, DST_CODE>
354	where
355	BitReader<E>: BitRead<E> + CodesRead<E>,
356	BitWriter<E>: BitWrite<E> + CodesWrite<E>,
357	{
358	fn len(&self) -> usize {	×
359	self.len - self.current	×
360	}
361	}

vigna / webgraph-rs / 19378243882

Source File Press 'n' to go to next uncovered line, 'b' for previous

Source File
Press 'n' to go to next uncovered line, 'b' for previous