Initial SFERA platform baseline

This commit is contained in:
2026-05-16 19:03:49 +03:00
commit 3b845c8fce
282 changed files with 55045 additions and 0 deletions
+9
View File
@@ -0,0 +1,9 @@
# sfera-semantic-search
Deterministic semantic search over SIR snapshots.
Search covers:
- node name, qualified name, kind, and source path;
- metadata attributes;
- indexed module source text where present.
+11
View File
@@ -0,0 +1,11 @@
[project]
name = "sfera-semantic-search"
version = "0.1.0"
requires-python = ">=3.12"
dependencies = [
"pydantic>=2.0",
"sfera-sir",
]
[tool.uv]
package = true
@@ -0,0 +1,85 @@
from __future__ import annotations
from collections.abc import Iterable
from pydantic import BaseModel
from sir import SemanticNode, SirSnapshot
class SearchResult(BaseModel):
node: SemanticNode
score: float
matched_fields: list[str]
def search_snapshot(
snapshot: SirSnapshot,
query: str,
*,
kinds: set[str] | None = None,
limit: int = 20,
) -> list[SearchResult]:
normalized_query = query.casefold().strip()
if not normalized_query:
return []
results: list[SearchResult] = []
for node in snapshot.nodes:
if kinds is not None and node.kind.value not in kinds:
continue
score, fields = _score_node(node, normalized_query)
if score > 0:
results.append(SearchResult(node=node, score=score, matched_fields=fields))
results.sort(key=lambda result: (-result.score, result.node.qualified_name))
return results[:limit]
def _score_node(node: SemanticNode, query: str) -> tuple[float, list[str]]:
fields = {
"name": node.name,
"qualified_name": node.qualified_name,
"kind": node.kind.value,
"source_path": node.source_ref.source_path,
}
score = 0.0
matched: list[str] = []
for field, value in fields.items():
field_score = _score_text(value, query)
if field_score:
score += field_score
matched.append(field)
for field, value in _attribute_search_fields(node.attributes):
field_score = _score_text(value, query)
if field_score:
score += max(field_score - 1.0, 1.0)
matched.append(field)
return score, matched
def _score_text(value: object, query: str) -> float:
normalized = str(value).casefold()
if normalized == query:
return 10.0
if normalized.startswith(query):
return 5.0
if query in normalized:
return 2.0
return 0.0
def _attribute_search_fields(attributes: dict) -> Iterable[tuple[str, object]]:
for key, value in sorted(attributes.items()):
field = f"attributes.{key}"
if isinstance(value, dict):
for nested_key, nested_value in _attribute_search_fields(value):
yield f"{field}.{nested_key.removeprefix('attributes.')}", nested_value
elif isinstance(value, list):
for index, item in enumerate(value):
yield f"{field}[{index}]", item
else:
yield field, value
__all__ = ["SearchResult", "search_snapshot"]
@@ -0,0 +1,59 @@
from pathlib import Path
from semantic_kernel import index_project
from semantic_search import search_snapshot
def test_search_snapshot_finds_routine_by_partial_name(tmp_path: Path):
module = tmp_path / "demo_module.bsl"
module.write_text(
"""
Процедура Проведение()
КонецПроцедуры
""",
encoding="utf-8",
)
snapshot = index_project(tmp_path, project_id="demo")
results = search_snapshot(snapshot, "Пров", kinds={"PROCEDURE"})
assert results
assert results[0].node.name == "Проведение"
def test_search_snapshot_finds_node_by_source_text(tmp_path: Path):
module = tmp_path / "integration.bsl"
module.write_text(
"""
Процедура Отправить()
Адрес = "https://api.example.local/orders";
КонецПроцедуры
""",
encoding="utf-8",
)
snapshot = index_project(tmp_path, project_id="search-source-text")
results = search_snapshot(snapshot, "api.example.local")
assert results
module_result = next(result for result in results if result.node.name == "integration")
assert "attributes.source_text" in module_result.matched_fields
def test_search_snapshot_finds_node_by_metadata_attribute(tmp_path: Path):
xml = tmp_path / "metadata.xml"
xml.write_text(
"""
<Configuration>
<Document name="ЗаказПокупателя" qualifiedName="Документ.ЗаказПокупателя" synonym="Customer Order" />
</Configuration>
""",
encoding="utf-8",
)
snapshot = index_project(tmp_path, project_id="search-attributes")
results = search_snapshot(snapshot, "Customer")
assert results
assert results[0].node.qualified_name == "Документ.ЗаказПокупателя"
assert "attributes.synonym" in results[0].matched_fields