import re
from pathlib import Path
from src.memory.models import ExtractedLink
LINK_PATTERN = re.compile(r'\[\[([^\]]+)\]\]')
EDGE_TYPE_KEYWORDS: dict[str, list[str]] = {
"refactors": ["refactor", "rewrite", "restructure", "reorganize", "cleanup"],
"plans": ["plan", "todo", "will", "should", "need to", "going to"],
"debugs": ["bug", "fix", "issue", "error", "problem", "broken"],
"mentions": ["note", "remember", "mention", "see also", "refer"],
}
MEMORY_RELATIONSHIP_KEYWORDS: dict[str, list[str]] = {
"SUPERSEDES": ["supersedes", "replaces", "obsoletes", "deprecates"],
"EXTENDS": ["extends", "builds on", "elaborates", "expands on"],
"CONTRADICTS": ["contradicts", "conflicts with", "disagrees with"],
"DEPENDS_ON": ["depends on", "requires", "needs", "builds on"],
}
def normalize_tag(tag: str) -> str:
return tag.strip().lower()
def infer_edge_type(context: str) -> str:
context_lower = context.lower()
for edge_type, keywords in EDGE_TYPE_KEYWORDS.items():
for keyword in keywords:
if keyword in context_lower:
return edge_type
return "related_to"
def infer_memory_relationship(context: str) -> str:
context_lower = context.lower()
for edge_type, keywords in MEMORY_RELATIONSHIP_KEYWORDS.items():
for keyword in keywords:
if keyword in context_lower:
return edge_type
return "RELATED_TO"
def parse_link_target(target: str) -> tuple[str, bool]:
if target.startswith("memory:"):
memory_name = target[7:].strip()
memory_id = f"memory:{Path(memory_name).stem}"
return memory_id, True
return target, False
def extract_links(content: str, context_chars: int = 100) -> list[ExtractedLink]:
links: list[ExtractedLink] = []
for match in LINK_PATTERN.finditer(content):
target = match.group(1).strip()
if not target:
continue
start_pos = match.start()
end_pos = match.end()
context_start = max(0, start_pos - context_chars)
context_end = min(len(content), end_pos + context_chars)
before_text = content[context_start:start_pos]
after_text = content[end_pos:context_end]
anchor_context = f"{before_text}[[{target}]]{after_text}"
resolved_target, is_memory = parse_link_target(target)
if is_memory:
edge_type = infer_memory_relationship(anchor_context)
else:
edge_type = infer_edge_type(anchor_context)
links.append(ExtractedLink(
target=resolved_target,
edge_type=edge_type,
anchor_context=anchor_context.strip(),
position=start_pos,
is_memory_link=is_memory,
))
return links
