# extractor.py - Canadian Building Code GPT Helper
# Upload this file to GPT Knowledge for faster PDF extraction
import re
def extract_section(pdf_path, page_num, section_id):
"""
Extract section text from PDF by Section ID.
Args:
pdf_path: Path to uploaded PDF file
page_num: Page number (1-indexed)
section_id: Section ID like "9.8.8.2", "3.2.2.55"
Returns:
Extracted text or error message
"""
# Get page text
text = None
try:
import fitz
doc = fitz.open(pdf_path)
page = doc[page_num - 1]
text = page.get_text("text")
doc.close()
except:
try:
import pdfplumber
with pdfplumber.open(pdf_path) as pdf:
page = pdf.pages[page_num - 1]
text = page.extract_text() or ""
except Exception as e:
return f"Error reading PDF: {e}"
if not text:
return f"Could not extract text from page {page_num}"
# Find section ID in text
if section_id not in text:
# Try without leading zeros or with variations
variations = [
section_id,
section_id.replace(".", ". "),
section_id + ".",
section_id + " "
]
found = False
for var in variations:
if var in text:
section_id = var
found = True
break
if not found:
return f"Section {section_id} not found on page {page_num}.\n\nPage preview:\n{text[:1000]}"
# Extract from section ID to next section
start_idx = text.find(section_id)
remaining = text[start_idx:]
# Find next section (pattern: newline + number.number.number)
next_match = re.search(r'\n\d+\.\d+\.\d+\.?\d*[\s\.]', remaining[len(section_id):])
if next_match:
end_idx = len(section_id) + next_match.start()
result = remaining[:end_idx].strip()
else:
# No next section, take up to 2500 chars
result = remaining[:2500].strip()
return result
def search_json(json_data, query, code_filter=None):
"""
Search JSON for sections matching query.
Args:
json_data: Loaded JSON data (dict)
query: Search keywords
code_filter: Optional code name filter (e.g., "OBC", "NBC")
Returns:
List of matching sections
"""
results = []
query_lower = query.lower()
query_words = query_lower.split()
# Get codes to search
if "codes" in json_data:
codes = json_data["codes"]
else:
codes = {json_data.get("code", "unknown"): json_data}
for code_name, code_data in codes.items():
# Apply code filter
if code_filter and code_filter.upper() not in code_name.upper():
continue
sections = code_data.get("sections", [])
for section in sections:
score = 0
# Check title
title = section.get("title", "").lower()
for word in query_words:
if word in title:
score += 10
# Check keywords
keywords = section.get("keywords", [])
for word in query_words:
if word in keywords:
score += 5
for kw in keywords:
if word in kw:
score += 2
if score > 0:
results.append({
"id": section.get("id"),
"title": section.get("title"),
"page": section.get("page"),
"code": code_name,
"score": score
})
# Sort by score
results.sort(key=lambda x: x["score"], reverse=True)
return results[:10] # Top 10
def extract_sections_batch(pdf_path, sections):
"""
Extract multiple sections in ONE PDF open (faster!).
Args:
pdf_path: Path to uploaded PDF file
sections: List of {"page": int, "id": str}
e.g., [{"page": 452, "id": "4.1.3.2"}, ...]
Returns:
Dict of {section_id: extracted_text}
"""
results = {}
# Try PyMuPDF first
try:
import fitz
doc = fitz.open(pdf_path)
for s in sections:
page_num = s["page"]
section_id = s["id"]
try:
page = doc[page_num - 1]
text = page.get_text("text")
if section_id in text:
start_idx = text.find(section_id)
remaining = text[start_idx:]
# Find next section
next_match = re.search(
r'\n\d+\.\d+\.\d+\.?\d*[\s\.]',
remaining[len(section_id):]
)
if next_match:
end_idx = len(section_id) + next_match.start()
results[section_id] = remaining[:end_idx].strip()
else:
results[section_id] = remaining[:2000].strip()
else:
results[section_id] = f"[Section {section_id} not found on page {page_num}]"
except Exception as e:
results[section_id] = f"[Error extracting {section_id}: {e}]"
doc.close()
return results
except ImportError:
pass
# Fallback: pdfplumber
try:
import pdfplumber
with pdfplumber.open(pdf_path) as pdf:
for s in sections:
page_num = s["page"]
section_id = s["id"]
try:
page = pdf.pages[page_num - 1]
text = page.extract_text() or ""
if section_id in text:
start_idx = text.find(section_id)
remaining = text[start_idx:]
next_match = re.search(
r'\n\d+\.\d+\\.d+\.?\d*[\s\.]',
remaining[len(section_id):]
)
if next_match:
end_idx = len(section_id) + next_match.start()
results[section_id] = remaining[:end_idx].strip()
else:
results[section_id] = remaining[:2000].strip()
else:
results[section_id] = f"[Section {section_id} not found on page {page_num}]"
except Exception as e:
results[section_id] = f"[Error: {e}]"
return results
except Exception as e:
return {"error": f"Could not open PDF: {e}"}
def extract_table(json_data, table_id):
"""
Extract table markdown from JSON maps (fast, primary method).
Args:
json_data: Loaded JSON map data (single code or multi-code dict)
table_id: Table ID (e.g., "4.1.5.3" or "Table-4.1.5.3")
Returns:
Dict with table info or error message
{
"id": "Table-4.1.5.3",
"title": "Live Loads on Floors...",
"page": 452,
"markdown": "| Col1 | Col2 |\n|---|---|\n...",
"table_info": {"rows": 15, "cols": 4},
"source": "JSON"
}
"""
# Normalize ID - add "Table-" prefix if missing
if not table_id.startswith("Table-"):
table_id = f"Table-{table_id}"
# Handle both single-code and multi-code JSON structures
codes = json_data if isinstance(json_data, dict) and "tables" not in json_data else {"default": json_data}
# Search in tables array
for code_name, code_data in codes.items():
tables = code_data.get("tables", [])
for table in tables:
if table.get("id") == table_id:
return {
"id": table["id"],
"title": table.get("title", ""),
"page": table.get("page", 0),
"markdown": table.get("markdown", ""),
"table_info": table.get("table_info", {}),
"source": "JSON"
}
return {"error": f"Table {table_id} not found in JSON"}
def extract_table_from_pdf(pdf_path, page, table_id):
"""
Attempt to extract table from PDF by searching for table ID.
Note: No bbox available, uses text search + heuristics.
This is a FALLBACK method - JSON markdown is preferred.
Args:
pdf_path: Path to PDF file
page: Page number
table_id: Table ID to search for (e.g., "Table-4.1.5.3")
Returns:
Dict with extracted text or error
"""
# Normalize table ID
if not table_id.startswith("Table-"):
table_id = f"Table-{table_id}"
# Try PyMuPDF first
try:
import fitz
doc = fitz.open(pdf_path)
page_obj = doc[page - 1]
text = page_obj.get_text("text")
doc.close()
# Search for table ID (with or without "Table " prefix)
search_patterns = [
table_id,
table_id.replace("Table-", "Table "),
table_id.replace("Table-", "")
]
start_idx = -1
for pattern in search_patterns:
if pattern in text:
start_idx = text.find(pattern)
break
if start_idx >= 0:
# Extract ~2000 chars after table ID
excerpt = text[start_idx:start_idx + 2000]
return {
"id": table_id,
"page": page,
"text": excerpt,
"source": "PDF (heuristic)",
"note": "⚠️ Table extracted from PDF text. Format may not match original."
}
else:
return {"error": f"Table {table_id} not found on page {page}"}
except ImportError:
pass
# Fallback: pdfplumber
try:
import pdfplumber
with pdfplumber.open(pdf_path) as pdf:
page_obj = pdf.pages[page - 1]
text = page_obj.extract_text() or ""
search_patterns = [
table_id,
table_id.replace("Table-", "Table "),
table_id.replace("Table-", "")
]
start_idx = -1
for pattern in search_patterns:
if pattern in text:
start_idx = text.find(pattern)
break
if start_idx >= 0:
excerpt = text[start_idx:start_idx + 2000]
return {
"id": table_id,
"page": page,
"text": excerpt,
"source": "PDF (heuristic)",
"note": "⚠️ Table extracted from PDF text. Format may not match original."
}
else:
return {"error": f"Table {table_id} not found on page {page}"}
except Exception as e:
return {"error": f"PDF extraction failed: {e}"}
def extract_tables_batch(json_data, table_ids):
"""
Extract multiple tables from JSON in one call (fast).
Args:
json_data: Loaded JSON map
table_ids: List of table IDs
e.g., ["4.1.5.3", "9.10.14.4"] or ["Table-4.1.5.3", ...]
Returns:
Dict of {table_id: table_data}
e.g., {
"Table-4.1.5.3": {"id": "...", "markdown": "...", ...},
"Table-9.10.14.4": {"id": "...", "markdown": "...", ...}
}
"""
results = {}
for tid in table_ids:
# Normalize ID
normalized_id = tid if tid.startswith("Table-") else f"Table-{tid}"
# Extract table
result = extract_table(json_data, tid)
results[normalized_id] = result
return results
# Quick usage examples:
#
# 1. Extract single section from PDF:
# text = extract_section("/path/to/obc.pdf", 127, "9.8.8.2")
#
# 2. Extract multiple sections (FASTER!):
# texts = extract_sections_batch("/path/to/nbc.pdf", [
# {"page": 452, "id": "4.1.3.2"},
# {"page": 538, "id": "4.3.3.1"},
# {"page": 537, "id": "4.2.7.2"}
# ])
#
# 3. Search JSON:
# import json
# with open("OBC_Vol1.json") as f:
# data = json.load(f)
# results = search_json(data, "guard height stairs")
#
# 4. Extract table from JSON (NEW!):
# table = extract_table(data, "4.1.5.3")
# print(table["markdown"]) # Display formatted table
#
# 5. Extract multiple tables (BATCH!):
# tables = extract_tables_batch(data, ["4.1.5.3", "9.10.14.4"])
# for tid, table in tables.items():
# print(f"## {table['title']}")
# print(table['markdown'])
#
# 6. Extract table from PDF (FALLBACK!):
# table = extract_table_from_pdf("/path/to/nbc.pdf", 452, "Table-4.1.5.3")
# print(table["text"])
