spectre_pdf 1.0.0

Native Rust PDF extraction engine: text, markdown for RAG, AcroForm widgets, image decoding, and encrypted PDFs. Lazy parser, persistent Document handle, no C dependencies.
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

"""
Benchmark Document.open_with_password() against pymupdf authenticate.

Tests every PDF Standard Security Handler variant on a fixed corpus
generated via qpdf, comparing wall-clock decrypt-and-open + text-extract
time and verifying the extracted text exactly matches the unencrypted
baseline.

Usage:
    pip install pymupdf spectre_rs
    python scripts/bench_encrypted_vs_pymupdf.py [--corpus DIR]

The corpus is generated on first run via qpdf if not present.
"""

from __future__ import annotations

import argparse
import os
import statistics
import subprocess
import time
import urllib.request
from pathlib import Path

import fitz
import spectre_rs


SOURCE_URL = "https://www.irs.gov/pub/irs-pdf/fw9.pdf"

VARIANTS = [
    ("v1_rc4_40.pdf", b"secret", "V=1 R=2", "RC4 40-bit"),
    ("v2_rc4_128.pdf", b"secret", "V=2 R=3", "RC4 128-bit"),
    ("v4_aes_128.pdf", b"secret", "V=4 R=4", "AES-128 CBC"),
    ("v4_aes_128_empty_user.pdf", b"", "V=4 R=4", "AES-128, empty user"),
    ("v5_aes_256_r5.pdf", b"secret", "V=5 R=5", "AES-256 (Acrobat 9, withdrawn)"),
    ("v5_aes_256.pdf", b"secret", "V=5 R=6", "AES-256 (Alg 2.B)"),
    ("v5_aes_256_empty_user.pdf", b"", "V=5 R=6", "AES-256, empty user"),
]


def ensure_corpus(corpus_dir: Path) -> None:
    corpus_dir.mkdir(parents=True, exist_ok=True)
    source = corpus_dir / "source.pdf"
    if not source.exists():
        print("fetch IRS W-9 source PDF")
        urllib.request.urlretrieve(SOURCE_URL, source)
    if not (corpus_dir / "v5_aes_256.pdf").exists():
        if subprocess.run(["which", "qpdf"], capture_output=True).returncode != 0:
            raise SystemExit(
                "qpdf is required to generate the encrypted corpus. "
                "Install via `apt install qpdf` and re-run."
            )
        cmds = [
            ("v1_rc4_40.pdf",
             ["qpdf", "--allow-weak-crypto", "--encrypt", "secret", "secret",
              "40", "--", str(source), str(corpus_dir / "v1_rc4_40.pdf")]),
            ("v2_rc4_128.pdf",
             ["qpdf", "--allow-weak-crypto", "--encrypt", "secret", "secret",
              "128", "--use-aes=n", "--", str(source),
              str(corpus_dir / "v2_rc4_128.pdf")]),
            ("v4_aes_128.pdf",
             ["qpdf", "--encrypt", "secret", "secret", "128",
              "--use-aes=y", "--", str(source),
              str(corpus_dir / "v4_aes_128.pdf")]),
            ("v4_aes_128_empty_user.pdf",
             ["qpdf", "--encrypt", "", "secret", "128", "--use-aes=y",
              "--", str(source), str(corpus_dir / "v4_aes_128_empty_user.pdf")]),
            ("v5_aes_256_r5.pdf",
             ["qpdf", "--encrypt", "secret", "secret", "256", "--force-R5",
              "--", str(source), str(corpus_dir / "v5_aes_256_r5.pdf")]),
            ("v5_aes_256.pdf",
             ["qpdf", "--encrypt", "secret", "secret", "256", "--",
              str(source), str(corpus_dir / "v5_aes_256.pdf")]),
            ("v5_aes_256_empty_user.pdf",
             ["qpdf", "--encrypt", "", "secret", "256", "--",
              str(source), str(corpus_dir / "v5_aes_256_empty_user.pdf")]),
        ]
        for name, cmd in cmds:
            print(f"generate {name}")
            subprocess.run(cmd, check=True)


def time_decrypt_and_extract_spectre(b: bytes, pwd: bytes, n: int = 3) -> tuple[float, str]:
    text_out = ""
    runs = []
    for _ in range(n):
        t0 = time.perf_counter()
        d = spectre_rs.Document(b, password=pwd)
        text_out = d.text()
        runs.append((time.perf_counter() - t0) * 1000)
    return statistics.median(runs), text_out


def time_decrypt_and_extract_pymupdf(b: bytes, pwd: bytes, n: int = 3) -> tuple[float, str]:
    text_out = ""
    runs = []
    for _ in range(n):
        t0 = time.perf_counter()
        d = fitz.open(stream=b)
        d.authenticate(pwd.decode("latin-1"))
        text_out = "".join(p.get_text() for p in d).strip()
        runs.append((time.perf_counter() - t0) * 1000)
        d.close()
    return statistics.median(runs), text_out


def bench(corpus_dir: Path) -> None:
    source_bytes = (corpus_dir / "source.pdf").read_bytes()
    baseline = spectre_rs.Document(source_bytes).text().strip()
    print(f"plaintext baseline: {len(baseline)} chars")
    print("=" * 110)
    print(f"{'variant':<22} {'cipher':<25} {'pmu ms':>8} {'spec ms':>9} "
          f"{'speedup':>9} {'text match':>11}")
    print("-" * 110)
    total_s = total_p = 0.0
    matches_spec = matches_pmu = 0
    for name, pwd, version, label in VARIANTS:
        path = corpus_dir / name
        if not path.exists() or path.stat().st_size == 0:
            print(f"  {name}: SKIPPED (file missing/empty)")
            continue
        b = path.read_bytes()
        spec_ms, spec_text = time_decrypt_and_extract_spectre(b, pwd)
        pmu_ms, pmu_text = time_decrypt_and_extract_pymupdf(b, pwd)
        total_s += spec_ms
        total_p += pmu_ms
        spec_ok = spec_text.strip() == baseline
        pmu_ok = pmu_text.strip() == baseline
        matches_spec += int(spec_ok)
        matches_pmu += int(pmu_ok)
        speedup = pmu_ms / spec_ms if spec_ms else float("inf")
        tag_spec = "OK" if spec_ok else "DIFF"
        print(f"{version:<22} {label:<25} {pmu_ms:>8.2f} {spec_ms:>9.2f} "
              f"{speedup:>8.2f}x {tag_spec:>11}")
    print("-" * 110)
    print(f"TOTAL  speedup {total_p / total_s:.2f}x "
          f"({total_p:.1f}ms / {total_s:.1f}ms)")
    print(f"  spectre baseline-text match:  {matches_spec}/{len(VARIANTS)}")
    print(f"  pymupdf baseline-text match:  {matches_pmu}/{len(VARIANTS)}")

    # Negative test — wrong password should error cleanly on both.
    print()
    print("Wrong-password test (expect both libraries to reject):")
    for name in ["v4_aes_128.pdf", "v5_aes_256.pdf"]:
        path = corpus_dir / name
        if not path.exists():
            continue
        b = path.read_bytes()
        try:
            spectre_rs.Document(b, password=b"wrong-password")
            print(f"  {name}: spectre INCORRECTLY ACCEPTED wrong password")
        except Exception as e:
            print(f"  {name}: spectre rejected ({type(e).__name__})")
        d = fitz.open(stream=b)
        rc = d.authenticate("wrong-password")
        print(f"  {name}: pymupdf rc={rc}")
        d.close()


def main() -> None:
    p = argparse.ArgumentParser()
    p.add_argument("--corpus", default="bench-corpus/encrypted")
    args = p.parse_args()
    corpus = Path(args.corpus)
    ensure_corpus(corpus)
    bench(corpus)


if __name__ == "__main__":
    main()