File size: 8,565 Bytes
62f3c0d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
"""Parse SP 21 PDF into structured per-standard records.

The PDF is laid out so that each standard is bounded by a 'SUMMARY OF' header
followed by an IS code line. We scan the full text page-by-page, segment by
those headers, and extract metadata + body for each standard.
"""
from __future__ import annotations

import json
import re
from dataclasses import dataclass, asdict
from pathlib import Path

import fitz

IS_CODE_RE = re.compile(
    r"IS\s+(\d+)\s*(\(Part\s*[IVX0-9]+\))?\s*:\s*(\d{4})",
    re.IGNORECASE,
)
SUMMARY_HDR_RE = re.compile(r"SUMMARY\s+OF\s*\n\s*(IS\s+[^\n]+)", re.IGNORECASE)
REVISION_TAIL_RE = re.compile(r"\((First|Second|Third|Fourth|Fifth)\s+Revision\)\s*$", re.IGNORECASE)


@dataclass
class Standard:
    is_code: str           # canonical: "IS 3466: 1988" or "IS 2185 (Part 2): 1983"
    is_code_norm: str      # normalized: "is3466:1988" — matches eval_script normalization
    title: str             # e.g. "MASONRY CEMENT"
    revision: str | None   # e.g. "Second Revision" or None
    page_start: int        # 1-indexed
    page_end: int
    scope: str             # extracted "Scope" body if found
    full_text: str         # entire body text (post header)


def normalize_is_code(s: str) -> str:
    return re.sub(r"\s+", "", s).lower()


def canonicalize_is_code(part_num: str | None, number: str, year: str) -> str:
    """Canonical form mirrors the public_test_set.json style: 'IS 2185 (Part 2): 1983'."""
    if part_num:
        # Normalize whitespace and title-case the word 'Part'
        part_clean = re.sub(r"\s+", " ", part_num.strip())
        part_clean = re.sub(r"^\(\s*part\s*", "(Part ", part_clean, flags=re.IGNORECASE)
        return f"IS {number} {part_clean}: {year}"
    return f"IS {number}: {year}"


def extract_standards(pdf_path: Path) -> list[Standard]:
    """Walk every page, detect 'SUMMARY OF\\nIS ...' anchors, slice the doc accordingly."""
    doc = fitz.open(pdf_path)

    # Build a list of (page_idx, page_text). Append page markers so we can map back.
    pages_text: list[str] = []
    for page in doc:
        pages_text.append(page.get_text("text"))
    doc.close()

    # Find anchors per page
    anchors: list[tuple[int, str, str, str | None, str, str, str | None]] = []
    # (page_idx_0based, raw_header_line, is_number, part, year, title, revision)

    for page_idx, text in enumerate(pages_text):
        for m in SUMMARY_HDR_RE.finditer(text):
            header_line = m.group(1).strip()
            code_match = IS_CODE_RE.search(header_line)
            if not code_match:
                continue
            number = code_match.group(1)
            part = code_match.group(2)
            year = code_match.group(3)

            # Title = everything in header_line after the IS code.
            # The "(X Revision)" tag may appear inline OR on one of the next
            # 1–3 lines (some entries break the title onto multiple lines).
            after_code = header_line[code_match.end():].strip()
            tail_after_anchor = text[m.end():m.end() + 400]
            tail_lines = [ln.strip() for ln in tail_after_anchor.lstrip().split("\n") if ln.strip()]

            # If the title is empty after the IS-code line, take the next
            # non-empty line as the title (some standards put it there).
            if not after_code and tail_lines:
                after_code = tail_lines[0]
                tail_lines = tail_lines[1:]

            # Search for a revision marker in the title or in any of the next
            # ~3 short lines (gives multi-line titles a chance).
            revision = None
            rev_match = REVISION_TAIL_RE.search(after_code)
            if rev_match:
                revision = rev_match.group(0).strip("() ")
                after_code = REVISION_TAIL_RE.sub("", after_code).strip()
            else:
                _REV_LINE_RE = re.compile(
                    r"^\(?\s*(First|Second|Third|Fourth|Fifth|Sixth|Seventh|Eighth|Ninth|Tenth)"
                    r"\s+Revision\s*\)?\s*$",
                    re.IGNORECASE,
                )
                for cand in tail_lines[:3]:
                    if _REV_LINE_RE.match(cand):
                        revision = re.sub(r"[()]", "", cand).strip()
                        break

            title = after_code.strip()

            anchors.append((page_idx, header_line, number, part, year, title, revision))

    # Build standards by slicing pages_text from anchor i to anchor i+1
    standards: list[Standard] = []
    for i, (page_idx, header_line, number, part, year, title, revision) in enumerate(anchors):
        start_page = page_idx
        end_page = (anchors[i + 1][0] - 1) if i + 1 < len(anchors) else len(pages_text) - 1
        # If the next anchor is on the same page, the standard ends on this page
        if i + 1 < len(anchors) and anchors[i + 1][0] == page_idx:
            end_page = page_idx

        body_chunks: list[str] = []
        for p in range(start_page, end_page + 1):
            body_chunks.append(pages_text[p])
        full_text = "\n".join(body_chunks).strip()

        # Extract Scope: text after "Scope" / "1. Scope" until next numbered heading
        scope = _extract_scope(full_text)

        canonical = canonicalize_is_code(part, number, year)
        standards.append(Standard(
            is_code=canonical,
            is_code_norm=normalize_is_code(canonical),
            title=title,
            revision=revision,
            page_start=start_page + 1,
            page_end=end_page + 1,
            scope=scope,
            full_text=full_text,
        ))
    return standards


# Note: NO re.IGNORECASE here — we rely on uppercase-letter classes acting as
# case-sensitive markers. The lookahead terminates on next numbered section
# (e.g. "2. ..."), an uppercase-only heading line, or "SUMMARY OF" (start of
# next standard).
_SCOPE_RE = re.compile(
    r"(?:^|\n)\s*1[\.\s]+(?:Scope|SCOPE)[\s\-\u2014\u2013:]*"
    r"([\s\S]+?)"
    r"(?=\n\s*(?:\d+\.\s|[A-Z][A-Z][A-Z]+\b|SUMMARY\s+OF))"
)


def _extract_scope(full_text: str) -> str:
    m = _SCOPE_RE.search(full_text)
    if not m:
        return ""
    scope = m.group(1).strip()
    scope = re.sub(r"\s+", " ", scope)
    return scope[:1500]


def main():
    import os as _os
    pdf_path = Path("datasets/dataset.pdf")
    out_standards = Path("data/parsed_standards.json")
    out_whitelist = Path("data/is_code_whitelist.json")

    # Idempotency: skip parse if both outputs already exist (they're committed
    # in the repo so a fresh `git clone` already has them). Override with
    # PARSER_FORCE=1.
    if (
        not _os.getenv("PARSER_FORCE")
        and out_standards.exists()
        and out_whitelist.exists()
        and out_standards.stat().st_size > 1024
        and out_whitelist.stat().st_size > 100
    ):
        print(f"Parsed standards already exist at {out_standards} — skipping parse.")
        print("(Set PARSER_FORCE=1 to re-parse.)")
        return

    standards = extract_standards(pdf_path)
    print(f"Extracted {len(standards)} standards from {pdf_path}")

    # Save full structured output
    out_standards.parent.mkdir(parents=True, exist_ok=True)
    with out_standards.open("w", encoding="utf-8") as f:
        json.dump([asdict(s) for s in standards], f, ensure_ascii=False, indent=2)
    print(f"  -> {out_standards}")

    # Save whitelist (canonical + normalized) for hallucination guard
    whitelist = {
        "canonical": sorted({s.is_code for s in standards}),
        "normalized": sorted({s.is_code_norm for s in standards}),
    }
    with out_whitelist.open("w", encoding="utf-8") as f:
        json.dump(whitelist, f, ensure_ascii=False, indent=2)
    print(f"  -> {out_whitelist} ({len(whitelist['canonical'])} unique IS codes)")

    # Print a quick sanity check vs public test set expectations
    print("\nSanity check vs public_test_set.json expected_standards:")
    with open("datasets/public_test_set.json", encoding="utf-8") as f:
        pub = json.load(f)
    norm_set = set(whitelist["normalized"])
    miss = []
    for q in pub:
        for exp in q["expected_standards"]:
            n = normalize_is_code(exp)
            mark = "OK " if n in norm_set else "MISS"
            print(f"  {mark} {exp}")
            if n not in norm_set:
                miss.append(exp)
    if miss:
        print(f"\nMissing {len(miss)}/10 expected codes — investigate parser.")
    else:
        print("\nAll 10 expected codes present in whitelist.")


if __name__ == "__main__":
    main()