Vito Sansevero
1d4446082f
The Library stores text.strip() while a run keeps its exact raw text. The history window, run detail scope id, newest pair run and prompt hints now compare the raw text's stored form (one rule, utils/hashing.py, mirrored in SQLite with the same code points). Hints also keep matching the exact raw text, so older rows that kept outer whitespace stay hinted and protected.
131 lines
4.2 KiB
Python
131 lines
4.2 KiB
Python
"""Hashing utilities for PromptManager prompt deduplication.
|
|
|
|
This module provides cryptographic hashing functions to enable efficient
|
|
deduplication of prompts and content in the PromptManager system. It uses
|
|
SHA256 hashing with text normalization to ensure consistent hash generation
|
|
regardless of minor formatting differences.
|
|
|
|
Key features:
|
|
- SHA256-based prompt text hashing with normalization
|
|
- Content hashing for complex prompt metadata structures
|
|
- Duplicate detection utilities
|
|
- Consistent normalization (lowercase, trimmed whitespace)
|
|
|
|
Typical usage:
|
|
from utils.hashing import generate_prompt_hash, is_duplicate_prompt
|
|
|
|
hash1 = generate_prompt_hash("Beautiful landscape")
|
|
hash2 = generate_prompt_hash(" beautiful landscape ")
|
|
# hash1 == hash2 (normalization makes them identical)
|
|
|
|
if is_duplicate_prompt(text1, text2):
|
|
print("Duplicate prompts detected")
|
|
|
|
The hashing is designed for:
|
|
- Database deduplication (preventing duplicate prompt storage)
|
|
- Quick similarity checking
|
|
- Content integrity verification
|
|
- Efficient database indexing
|
|
"""
|
|
|
|
import hashlib
|
|
|
|
|
|
def generate_prompt_hash(text: str) -> str:
|
|
"""
|
|
Generate a SHA256 hash for prompt text to enable deduplication.
|
|
|
|
Normalizes the input text (strips whitespace, converts to lowercase)
|
|
before hashing to ensure consistent results for functionally identical
|
|
prompts with minor formatting differences.
|
|
|
|
Args:
|
|
text: The prompt text to hash
|
|
|
|
Returns:
|
|
SHA256 hexadecimal digest of the normalized text
|
|
|
|
Raises:
|
|
TypeError: If the input is not a string
|
|
|
|
Example:
|
|
>>> generate_prompt_hash("Beautiful Landscape")
|
|
'c3d5f8a9b2e1...'
|
|
>>> generate_prompt_hash(" beautiful landscape ")
|
|
'c3d5f8a9b2e1...'
|
|
# Both produce the same hash due to normalization
|
|
"""
|
|
if not isinstance(text, str):
|
|
raise TypeError("Text must be a string")
|
|
|
|
# Normalize the text by stripping whitespace and converting to lowercase
|
|
# for consistent hashing regardless of minor formatting differences
|
|
normalized_text = text.strip().lower()
|
|
|
|
return hashlib.sha256(normalized_text.encode("utf-8")).hexdigest()
|
|
|
|
|
|
def stored_prompt_text(text: str) -> str:
|
|
"""The text the Library stores for raw node text: outer whitespace removed.
|
|
|
|
Every prompt writer stores ``text.strip()`` and the hash above strips too,
|
|
so this is the form recorded raw run text is compared in (#438). Case and
|
|
inner whitespace stay significant.
|
|
"""
|
|
return text.strip()
|
|
|
|
|
|
# The code points str.strip() removes (every str.isspace() character), for
|
|
# SQLite's char(); tests/test_hashing.py pins it to the running Python.
|
|
STRIPPED_CODE_POINTS = (
|
|
*range(0x09, 0x0E),
|
|
*range(0x1C, 0x21),
|
|
0x85,
|
|
0xA0,
|
|
0x1680,
|
|
*range(0x2000, 0x200B),
|
|
0x2028,
|
|
0x2029,
|
|
0x202F,
|
|
0x205F,
|
|
0x3000,
|
|
)
|
|
_STRIP_CODES = ",".join(map(str, STRIPPED_CODE_POINTS))
|
|
|
|
|
|
def stored_prompt_text_sql(expression: str) -> str:
|
|
"""``stored_prompt_text`` as an SQLite expression over ``expression``.
|
|
|
|
SQLite's one-argument trim() removes spaces only; trim(X, Y) removes every
|
|
character of Y, multi-byte ones included, so it strips what Python does.
|
|
"""
|
|
return f"trim({expression}, char({_STRIP_CODES}))"
|
|
|
|
|
|
def is_duplicate_prompt(text1: str, text2: str, threshold: float = 0.95) -> bool:
|
|
"""
|
|
Check if two prompts are likely duplicates using hash comparison.
|
|
|
|
Compares the normalized hashes of two prompt texts to determine if
|
|
they are functionally identical. This is an exact match after normalization,
|
|
not a similarity measure.
|
|
|
|
Args:
|
|
text1: First prompt text to compare
|
|
text2: Second prompt text to compare
|
|
threshold: Similarity threshold (not used - kept for API compatibility)
|
|
|
|
Returns:
|
|
True if the prompts have identical normalized hashes (i.e., are duplicates),
|
|
False otherwise
|
|
|
|
Note:
|
|
The threshold parameter is not used in the current implementation as
|
|
this performs exact hash matching rather than similarity scoring.
|
|
It's maintained for potential future fuzzy matching capabilities.
|
|
"""
|
|
hash1 = generate_prompt_hash(text1)
|
|
hash2 = generate_prompt_hash(text2)
|
|
|
|
return hash1 == hash2
|