auto_encoder 0.2.4

Auto encoding library
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
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392

use crate::meta::{HtmlMetadata, ASSET_NUMBERS, FIRST_BYTE_MAP};

/// Magic byte signatures grouped by first byte for single-pass matching.
/// Sorted longest-first within each group so longer signatures match before shorter prefixes.
static MAGIC_TABLE: &[(u8, &[&[u8]])] = &[
    (0x00, &[&[0x00, 0x00, 0x01, 0xBA], &[0x00, 0x00, 0x01, 0xB3], &[0x00, 0x00, 0x00, 0x18]]),
    (0x1A, &[&[0x1A, 0x45, 0xDF, 0xA3]]),
    (0x1F, &[&[0x1F, 0x8B]]),
    (0x25, &[b"%PDF"]),
    (0x42, &[&[0x42, 0x5A, 0x68], &[0x42, 0x4D]]),
    (0x46, &[&[0x46, 0x4C, 0x56, 0x01]]),
    (0x47, &[&[0x47, 0x49, 0x46, 0x38]]),
    (0x49, &[&[0x49, 0x49, 0x2A, 0x00], &[0x49, 0x49, 0x2B, 0x00], &[0x49, 0x44, 0x33]]),
    (0x4C, &[&[0x4C]]),
    (0x4D, &[&[0x4D, 0x4D, 0x00, 0x2A], &[0x4D, 0x4D, 0x00, 0x2B]]),
    (0x4F, &[&[0x4F, 0x67, 0x67, 0x53]]),
    (0x50, &[&[0x50, 0x4B, 0x03, 0x04]]),
    (0x52, &[&[0x52, 0x49, 0x46, 0x46]]),
    (0x66, &[&[0x66, 0x4C, 0x61, 0x43]]),
    (0x7F, &[&[0x7F, 0x45, 0x4C, 0x46]]),
    (0x89, &[&[0x89, 0x50, 0x4E, 0x47]]),
    (0xCA, &[&[0xCA, 0xFE, 0xBA, 0xBE]]),
    (0xFF, &[&[0xFF, 0xD8, 0xFF], &[0xFF, 0xFB]]),
];

/// 256-entry classifier built at compile time.  A byte is "interesting"
/// iff it's the first byte of a magic signature in [`MAGIC_TABLE`] or
/// ASCII whitespace (` `, `\t`, `\n`, `\r`, `\x0c` — matches
/// `u8::is_ascii_whitespace`).  All other bytes short-circuit to
/// `false` without touching the magic search or the whitespace retry.
///
/// One byte per entry, so the table is a single 256-byte cache line —
/// the lookup is one L1-resident load plus a branch.  This is faster
/// than the original `binary_search_by_key` for the dominant
/// "definitely-not-binary" HTML case (`<` etc.) by a factor of ~5x in
/// micro-benches, more than amortizing the small overhead added to the
/// magic-hit path for actual binary content.
static INTERESTING_FIRST_BYTE: [bool; 256] = {
    let mut t = [false; 256];
    // Magic-table first bytes — keep in sync with `MAGIC_TABLE`.
    let magic = [
        0x00u8, 0x1A, 0x1F, 0x25, 0x42, 0x46, 0x47, 0x49, 0x4C, 0x4D, 0x4F, 0x50, 0x52, 0x66,
        0x7F, 0x89, 0xCA, 0xFF,
    ];
    let mut i = 0;
    while i < magic.len() {
        t[magic[i] as usize] = true;
        i += 1;
    }
    // ASCII whitespace bytes.
    t[b' ' as usize] = true;
    t[b'\t' as usize] = true;
    t[b'\n' as usize] = true;
    t[b'\r' as usize] = true;
    t[0x0C] = true;
    t
};

/// Checks if the file is a known binary format using its initial bytes.
///
/// Performance shape:
/// * **Not binary, no whitespace** (the dominant HTML case): one
///   bounds-check, one `INTERESTING_FIRST_BYTE` lookup, return false.
/// * **Magic hit** (PNG/JPEG/PDF/...): same lookup then a binary search
///   plus per-signature compare, identical to the original logic
///   except for the leading single-cycle table load.
/// * **Whitespace-padded** (Aestiva HTML/OS:
///   `\n\n\n\n\n\n\n\n\n<!doctype html>`): hands off to a `#[cold]`
///   helper that trims and re-runs the search.  Catches binaries with
///   stray leading whitespace and stops whitespace-padded text bodies
///   from being mis-classified as binary.
#[inline]
pub fn is_binary_file(content: &[u8]) -> bool {
    let first = match content.first() {
        Some(&b) => b,
        None => return false,
    };
    if !INTERESTING_FIRST_BYTE[first as usize] {
        return false;
    }
    if let Ok(idx) = MAGIC_TABLE.binary_search_by_key(&first, |&(b, _)| b) {
        let (_, signatures) = MAGIC_TABLE[idx];
        for sig in signatures.iter() {
            if content.len() >= sig.len() && &content[..sig.len()] == *sig {
                return true;
            }
        }
        return false;
    }
    // Reached only when the byte is in `INTERESTING_FIRST_BYTE` but not
    // in `MAGIC_TABLE` — i.e. ASCII whitespace.  Cold path.
    is_binary_file_ws_cold(content)
}

/// Checks if the file is a known binary format using its initial bytes.
/// Uses the original PHF map implementation for backwards compatibility.
///
/// Same fast-path / cold-path shape as [`is_binary_file`].
#[inline]
pub fn is_binary_file_phf(content: &[u8]) -> bool {
    let first = match content.first() {
        Some(&b) => b,
        None => return false,
    };
    if !INTERESTING_FIRST_BYTE[first as usize] {
        return false;
    }
    if let Some(&keys) = FIRST_BYTE_MAP.get(&first) {
        for &key in keys {
            if let Some(&k) = ASSET_NUMBERS.get(key) {
                if content.len() >= k.len() && &content[..k.len()] == k {
                    return true;
                }
            }
        }
        return false;
    }
    is_binary_file_phf_ws_cold(content)
}

/// Cold whitespace-retry helper for [`is_binary_file`].  Reached only
/// when the first byte is ASCII whitespace.
#[cold]
#[inline(never)]
fn is_binary_file_ws_cold(content: &[u8]) -> bool {
    let mut i = 1;
    while i < content.len() && content[i].is_ascii_whitespace() {
        i += 1;
    }
    if i >= content.len() {
        return false;
    }
    // SAFETY: i ∈ [1, content.len()).
    let trimmed = unsafe { content.get_unchecked(i..) };
    let first = trimmed[0];
    if let Ok(idx) = MAGIC_TABLE.binary_search_by_key(&first, |&(b, _)| b) {
        let (_, signatures) = MAGIC_TABLE[idx];
        for sig in signatures.iter() {
            if trimmed.len() >= sig.len() && &trimmed[..sig.len()] == *sig {
                return true;
            }
        }
    }
    false
}

/// Cold whitespace-retry helper for [`is_binary_file_phf`].
#[cold]
#[inline(never)]
fn is_binary_file_phf_ws_cold(content: &[u8]) -> bool {
    let mut i = 1;
    while i < content.len() && content[i].is_ascii_whitespace() {
        i += 1;
    }
    if i >= content.len() {
        return false;
    }
    // SAFETY: i ∈ [1, content.len()).
    let trimmed = unsafe { content.get_unchecked(i..) };
    let first = trimmed[0];
    if let Some(&keys) = FIRST_BYTE_MAP.get(&first) {
        for &key in keys {
            if let Some(&k) = ASSET_NUMBERS.get(key) {
                if trimmed.len() >= k.len() && &trimmed[..k.len()] == k {
                    return true;
                }
            }
        }
    }
    false
}

/// Find first byte using memchr SIMD.
#[inline(always)]
fn find_byte(haystack: &[u8], needle: u8) -> Option<usize> {
    memchr::memchr(needle, haystack)
}

/// Fast subsequence search with adaptive strategy.
/// Scalar loop for small haystacks (< 128 bytes) to avoid SIMD setup overhead.
/// memchr + verify for larger haystacks where SIMD amortizes.
#[inline(always)]
fn find_short(haystack: &[u8], needle: &[u8]) -> Option<usize> {
    let nlen = needle.len();
    if nlen == 0 {
        return Some(0);
    }
    if nlen > haystack.len() {
        return None;
    }
    let first = needle[0];
    let rest = &needle[1..];
    let end = haystack.len() - nlen + 1;

    if haystack.len() < 128 {
        // Scalar: tight loop avoids memchr SIMD setup cost on small inputs
        let mut i = 0;
        while i < end {
            if haystack[i] == first && haystack[i + 1..i + nlen] == *rest {
                return Some(i);
            }
            i += 1;
        }
        None
    } else {
        // SIMD: memchr finds first byte fast, then verify remainder
        let mut offset = 0;
        while offset < end {
            match memchr::memchr(first, &haystack[offset..end]) {
                Some(pos) => {
                    let abs = offset + pos;
                    if haystack[abs + 1..abs + nlen] == *rest {
                        return Some(abs);
                    }
                    offset = abs + 1;
                }
                None => return None,
            }
        }
        None
    }
}

#[inline(always)]
fn extract_quoted_or_unquoted(after_attr: &[u8]) -> Option<String> {
    let &quote = after_attr.get(0)?;
    if quote == b'"' || quote == b'\'' {
        let quote_close = find_byte(&after_attr[1..], quote)?;
        std::str::from_utf8(&after_attr[1..quote_close + 1])
            .ok()
            .map(String::from)
    } else {
        let end = after_attr
            .iter()
            .position(|&c| c.is_ascii_whitespace() || c == b'>')?;
        std::str::from_utf8(&after_attr[..end])
            .ok()
            .map(String::from)
    }
}

#[inline(always)]
fn extract_charset_quoted(after_charset: &[u8]) -> Option<String> {
    let (&quote, remaining) = after_charset.split_first()?;
    if quote != b'"' && quote != b'\'' {
        return None;
    }
    let quote_close = find_byte(remaining, quote)?;
    std::str::from_utf8(&remaining[..quote_close])
        .ok()
        .map(String::from)
}

/// Detect the language of a HTML resource. This does nothing without the "encoding" flag enabled.
#[inline]
pub fn detect_language(html_content: &[u8]) -> Option<String> {
    if html_content.is_empty() {
        return None;
    }
    let search_area = &html_content[..html_content.len().min(1024)];
    let html_start = find_short(search_area, b"<html")?;
    let rest = &search_area[html_start..];
    let lang_start = find_short(rest, b"lang=")?;
    extract_quoted_or_unquoted(&rest[lang_start + 5..])
}

/// Detect the encoding used in an HTML file.
#[inline]
pub fn detect_encoding(html_content: &[u8]) -> Option<String> {
    let search_area = &html_content[..html_content.len().min(1024)];
    let mut pos = 0;

    while pos < search_area.len() {
        let remaining = &search_area[pos..];
        let meta_start = match find_short(remaining, b"<meta") {
            Some(s) => s,
            None => break,
        };
        let meta_content = &remaining[meta_start..];
        pos += meta_start + 5;

        // Case 1: <meta charset="...">
        if let Some(charset_start) = find_short(meta_content, b"charset=") {
            if let Some(result) = extract_charset_quoted(&meta_content[charset_start + 8..]) {
                return Some(result);
            }
        }

        // Case 2: <meta http-equiv="Content-Type" content="...; charset=...">
        if let Some(he_start) = find_short(meta_content, b"http-equiv=\"Content-Type\"") {
            let after_he = &meta_content[he_start + 25..];
            if let Some(cs) = find_short(after_he, b"content=") {
                let after_content = &after_he[cs + 8..];
                if let Some((&quote, rest)) = after_content.split_first() {
                    if quote == b'"' || quote == b'\'' {
                        if let Some(end) = find_byte(rest, quote) {
                            let full = &rest[..end];
                            if let Some(cp) = find_short(full, b"charset=") {
                                let after_cs = &full[cp + 8..];
                                let cs_end = after_cs
                                    .iter()
                                    .position(|&c| c == b';' || c.is_ascii_whitespace())
                                    .unwrap_or(after_cs.len());
                                if let Ok(charset) = std::str::from_utf8(&after_cs[..cs_end]) {
                                    return Some(charset.to_string());
                                }
                            }
                        }
                    }
                }
            }
        }
    }

    None
}

/// Detect the html metadata to process the element based on the encoding or language found.
#[inline]
pub fn detect_html_metadata(html_content: &[u8]) -> Option<HtmlMetadata> {
    if html_content.is_empty() {
        return Some(HtmlMetadata {
            lang: None,
            encoding: None,
        });
    }

    let search_area = &html_content[..html_content.len().min(1024)];

    // Detect language
    let lang = find_short(search_area, b"<html").and_then(|html_start| {
        let rest = &search_area[html_start..];
        find_short(rest, b"lang=")
            .and_then(|lang_start| extract_quoted_or_unquoted(&rest[lang_start + 5..]))
    });

    // Detect encoding
    let mut encoding: Option<String> = None;
    let mut pos = 0;
    while pos < search_area.len() {
        let remaining = &search_area[pos..];
        let meta_start = match find_short(remaining, b"<meta") {
            Some(s) => s,
            None => break,
        };
        let meta_content = &remaining[meta_start..];
        pos += meta_start + 5;

        if let Some(charset_start) = find_short(meta_content, b"charset=") {
            encoding = extract_charset_quoted(&meta_content[charset_start + 8..]);
            if encoding.is_some() {
                break;
            }
        }

        if let Some(he_start) = find_short(meta_content, b"http-equiv=\"Content-Type\"") {
            let after_he = &meta_content[he_start + 25..];
            if let Some(cs) = find_short(after_he, b"content=") {
                let after_content = &after_he[cs + 8..];
                if let Some((&quote, rest)) = after_content.split_first() {
                    if quote == b'"' || quote == b'\'' {
                        if let Some(end) = find_byte(rest, quote) {
                            let full = &rest[..end];
                            if let Some(cp) = find_short(full, b"charset=") {
                                let after_cs = &full[cp + 8..];
                                let cs_end = after_cs
                                    .iter()
                                    .position(|&c| c == b';' || c.is_ascii_whitespace())
                                    .unwrap_or(after_cs.len());
                                encoding = std::str::from_utf8(&after_cs[..cs_end])
                                    .ok()
                                    .map(String::from);
                                if encoding.is_some() {
                                    break;
                                }
                            }
                        }
                    }
                }
            }
        }
    }

    Some(HtmlMetadata { lang, encoding })
}

/// Helper function to find a subsequence in a slice.
/// Uses memchr for first-byte SIMD scan + manual verify for the rest.
#[inline(always)]
pub fn find_subsequence(haystack: &[u8], needle: &[u8]) -> Option<usize> {
    find_short(haystack, needle)
}