llm_readability 0.0.15

Readability library for LLM's built in Rust
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201

use super::dom;
use super::error::Error;
use super::scorer;
use crate::rcdom::{RcDom, SerializableHandle};
use html5ever::tendril::stream::TendrilSink;
use html5ever::{parse_document, serialize};
use std::cell::Cell;
use std::collections::BTreeMap;
use std::default::Default;
use std::io::Read;
use std::path::Path;
use url::Url;

#[cfg(feature = "tokio")]
use html5ever::tendril::stream::Utf8LossyDecoder;
#[cfg(feature = "tokio")]
use html5ever::tendril::{fmt as tendril_fmt, Tendril};
#[cfg(feature = "tokio")]
use html5ever::Parser;
#[cfg(feature = "tokio")]
use tokio::io::{AsyncRead, AsyncReadExt};

/// Above this size the async variants chunk input into html5ever's sink and
/// yield cooperatively between chunks, so the executor isn't starved by a
/// long synchronous parse. Below it, parsing is fast enough that yielding
/// would cost more than it saves.
#[cfg(feature = "tokio")]
pub const ASYNC_BYTE_THRESHOLD: usize = 128 * 1024;

/// Read / feed buffer size used when streaming bytes into html5ever's sink.
#[cfg(feature = "tokio")]
const STREAM_READ_CHUNK: usize = 32 * 1024;

#[derive(Debug)]
pub struct Product {
    /// The HTML content.
    pub content: String,
    /// The text content raw.
    pub text: String,
}

/// Readability alg extract a website url.
pub fn extract<R>(input: &mut R, url: &Url) -> Result<Product, Error>
where
    R: Read,
{
    let dom = parse_document(RcDom::default(), Default::default())
        .from_utf8()
        .read_from(input)?;
    process_dom(dom, url)
}

/// Run the readability pipeline on an already-parsed DOM. Used by both the
/// sync entry point and the async streaming variant after the sink finishes.
pub fn process_dom(mut dom: RcDom, url: &Url) -> Result<Product, Error> {
    let mut candidates = BTreeMap::new();
    let mut nodes = BTreeMap::new();
    let mut id: &str = "/";
    let mut bytes = vec![];
    let mut text: String = String::new();
    let mut title: String = String::new();
    let mut lang: String = String::new();

    let handle = dom.document.clone();

    scorer::preprocess(&mut dom, &handle, &mut title, &mut lang);
    scorer::find_candidates(Path::new(id), &handle, &mut candidates, &mut nodes);

    let mut top_candidate: &scorer::Candidate = &scorer::Candidate {
        node: handle,
        score: Cell::new(0.0),
    };

    for (i, c) in candidates.iter() {
        let score = c.score.get() * (1.0 - scorer::get_link_density(&c.node));
        c.score.set(score);
        if score <= top_candidate.score.get() {
            continue;
        }
        id = i;
        top_candidate = c;
    }

    let node = &top_candidate.node;

    scorer::clean(&mut dom, Path::new(id), &node, url, &candidates);

    serialize(
        &mut bytes,
        &SerializableHandle::from(node.clone()),
        Default::default(),
    )
    .ok();

    let content = auto_encoder::auto_encode_bytes(bytes.as_slice());

    dom::extract_text(&node, &mut text, true);

    let html_content = format!(
        r#"<html class="paper"{}><head>
<meta name="disabled-adaptations" content="watch">
<meta http-equiv="Content-Type" content="text/html; charset=utf-8">
<meta name="viewport" content="initial-scale=1">
<base href="{url}">
{}
<script>window.isReaderPage = true;</script>
</head><body>
"#,
        if !lang.is_empty() {
            format!(r#" lang="{}""#, &lang)
        } else {
            "".into()
        },
        if title.is_empty() {
            "".into()
        } else {
            format!("<title>{title}</title>")
        }
    );

    let formatted_content = format!("{}{}</body></html>", html_content, content);

    Ok(Product {
        content: formatted_content,
        text,
    })
}

/// Async variant of [`extract`] for callers that already have the body buffered.
///
/// Below [`ASYNC_BYTE_THRESHOLD`] the body is small enough that a single
/// straight-through parse won't meaningfully starve the executor, so we feed
/// it to the sink in one shot. Above the threshold we feed chunks into
/// html5ever's [`TendrilSink`] and call [`tokio::task::yield_now`] between
/// each chunk so the executor stays responsive.
///
/// The returned future is `!Send` because [`RcDom`] uses `Rc<RefCell<…>>`.
/// Run it on a `current_thread` runtime, in a [`tokio::task::LocalSet`],
/// or via [`tokio::task::spawn_local`].
#[cfg(feature = "tokio")]
pub async fn extract_async(bytes: Vec<u8>, url: Url) -> Result<Product, Error> {
    let mut sink: Utf8LossyDecoder<Parser<RcDom>> =
        parse_document(RcDom::default(), Default::default()).from_utf8();

    if bytes.len() < ASYNC_BYTE_THRESHOLD {
        // Small enough to feed in one shot — yields would cost more than they save.
        sink.process(Tendril::<tendril_fmt::Bytes>::from_slice(&bytes));
        let dom = sink.finish();
        return process_dom(dom, &url);
    }

    // Large: chunk through the sink and yield between chunks so we don't
    // monopolize the executor for a multi-millisecond CPU burst.
    let mut offset = 0;
    while offset < bytes.len() {
        let end = (offset + STREAM_READ_CHUNK).min(bytes.len());
        sink.process(Tendril::<tendril_fmt::Bytes>::from_slice(
            &bytes[offset..end],
        ));
        offset = end;
        tokio::task::yield_now().await;
    }
    let dom = sink.finish();
    process_dom(dom, &url)
}

/// Streaming async variant: pulls bytes from `reader` and feeds them straight
/// into html5ever's [`TendrilSink`] on the current task, chunk by chunk. The
/// `await` on each `reader.read()` is the cooperative yield point — there is
/// no thread hop, no `spawn_blocking`, no channel.
///
/// The returned future is `!Send` because the parser/[`RcDom`] use
/// `Rc<RefCell<…>>`. Run it on a `current_thread` runtime, in a
/// [`tokio::task::LocalSet`], or via [`tokio::task::spawn_local`].
#[cfg(feature = "tokio")]
pub async fn extract_async_reader<R>(mut reader: R, url: Url) -> Result<Product, Error>
where
    R: AsyncRead + Unpin,
{
    let mut sink: Utf8LossyDecoder<Parser<RcDom>> =
        parse_document(RcDom::default(), Default::default()).from_utf8();
    let mut buf = vec![0u8; STREAM_READ_CHUNK];
    let mut since_yield = 0usize;
    loop {
        let n = reader.read(&mut buf).await?;
        if n == 0 {
            break;
        }
        sink.process(Tendril::<tendril_fmt::Bytes>::from_slice(&buf[..n]));
        since_yield += n;
        // If the underlying reader keeps returning Ready (e.g. an in-memory
        // AsyncRead), insert an explicit yield once we've accumulated more
        // than the threshold so we don't starve the executor.
        if since_yield >= ASYNC_BYTE_THRESHOLD {
            tokio::task::yield_now().await;
            since_yield = 0;
        }
    }
    let dom = sink.finish();
    process_dom(dom, &url)
}