feat: Expose function to read serialized schema from parquet metadata (#82)

Author: borchero

dataframely/__init__.py

@@ -51,7 +51,7 @@
     filter_relationship_one_to_at_least_one,
     filter_relationship_one_to_one,
 )
-from .schema import Schema, deserialize_schema
+from .schema import Schema, deserialize_schema, read_parquet_metadata_schema
 
 __all__ = [
     "random",
@@ -69,6 +69,7 @@
     "filter_relationship_one_to_one",
     "Schema",
     "deserialize_schema",
+    "read_parquet_metadata_schema",
     "Any",
     "Bool",
     "Column",

dataframely/schema.py

@@ -860,21 +860,20 @@ def _requires_validation_for_reading_parquet(
         # First, we check whether the source provides the dataframely schema. If it
         # does, we check whether it matches this schema. If it does, we assume that the
         # data adheres to the schema and we do not need to run validation.
-        metadata = (
-            pl.read_parquet_metadata(source).get(SCHEMA_METADATA_KEY)
+        serialized_schema = (
+            read_parquet_metadata_schema(source)
             if not isinstance(source, list)
             else None
         )
-        if metadata is not None:
-            serialized_schema = deserialize_schema(metadata)
+        if serialized_schema is not None:
             if cls.matches(serialized_schema):
                 return False
 
         # Otherwise, we definitely need to run validation. However, we emit different
         # information to the user depending on the value of `validate`.
         msg = (
             "current schema does not match stored schema"
-            if metadata is not None
+            if serialized_schema is not None
             else "no schema to check validity can be read from the source"
         )
         if validation == "forbid":
@@ -956,6 +955,24 @@ def _rules_match(lhs: dict[str, Rule], rhs: dict[str, Rule]) -> bool:
         )
 
 
+def read_parquet_metadata_schema(
+    source: str | Path | IO[bytes] | bytes,
+) -> type[Schema] | None:
+    """Read a dataframely schema from the metadata of a parquet file.
+
+    Args:
+        source: Path to a parquet file or a file-like object that contains the metadata.
+
+    Returns:
+        The schema that was serialized to the metadata or ``None`` if no schema metadata
+        is found.
+    """
+    metadata = pl.read_parquet_metadata(source)
+    if (schema_metadata := metadata.get(SCHEMA_METADATA_KEY)) is not None:
+        return deserialize_schema(schema_metadata)
+    return None
+
+
 def deserialize_schema(data: str) -> type[Schema]:
     """Deserialize a schema from a JSON string.