feat: mimic dbt nuanced on_schema_change behavior

Author: eakmanrq

sqlmesh/core/config/connection.py

@@ -100,6 +100,7 @@ class ConnectionConfig(abc.ABC, BaseConfig):
     register_comments: bool
     pre_ping: bool
     pretty_sql: bool = False
+    schema_differ_overrides: t.Optional[t.Dict[str, t.Any]] = None
 
     # Whether to share a  single connection across threads or create a new connection per thread.
     shared_connection: t.ClassVar[bool] = False
@@ -174,6 +175,7 @@ def create_engine_adapter(
             pre_ping=self.pre_ping,
             pretty_sql=self.pretty_sql,
             shared_connection=self.shared_connection,
+            schema_differ_overrides=self.schema_differ_overrides,
             **self._extra_engine_config,
         )
 

sqlmesh/core/engine_adapter/athena.py

@@ -39,7 +39,7 @@ class AthenaEngineAdapter(PandasNativeFetchDFSupportMixin, RowDiffMixin):
     # CTAS, Views: No comment support at all
     COMMENT_CREATION_TABLE = CommentCreationTable.UNSUPPORTED
     COMMENT_CREATION_VIEW = CommentCreationView.UNSUPPORTED
-    SCHEMA_DIFFER = TrinoEngineAdapter.SCHEMA_DIFFER
+    SCHEMA_DIFFER_KWARGS = TrinoEngineAdapter.SCHEMA_DIFFER_KWARGS
     MAX_TIMESTAMP_PRECISION = 3  # copied from Trino
     # Athena does not deal with comments well, e.g:
     # >>> self._execute('/* test */ DESCRIBE foo')

sqlmesh/core/engine_adapter/base.py

@@ -14,7 +14,7 @@
 import logging
 import sys
 import typing as t
-from functools import partial
+from functools import cached_property, partial
 
 from sqlglot import Dialect, exp
 from sqlglot.errors import ErrorLevel
@@ -108,7 +108,7 @@ class EngineAdapter:
     SUPPORTS_CLONING = False
     SUPPORTS_MANAGED_MODELS = False
     SUPPORTS_CREATE_DROP_CATALOG = False
-    SCHEMA_DIFFER = SchemaDiffer()
+    SCHEMA_DIFFER_KWARGS: t.Dict[str, t.Any] = {}
     SUPPORTS_TUPLE_IN = True
     HAS_VIEW_BINDING = False
     SUPPORTS_REPLACE_TABLE = True
@@ -131,6 +131,7 @@ def __init__(
         pretty_sql: bool = False,
         shared_connection: bool = False,
         correlation_id: t.Optional[CorrelationId] = None,
+        schema_differ_overrides: t.Optional[t.Dict[str, t.Any]] = None,
         **kwargs: t.Any,
     ):
         self.dialect = dialect.lower() or self.DIALECT
@@ -153,6 +154,7 @@ def __init__(
         self._pretty_sql = pretty_sql
         self._multithreaded = multithreaded
         self.correlation_id = correlation_id
+        self._schema_differ_overrides = schema_differ_overrides
 
     def with_settings(self, **kwargs: t.Any) -> EngineAdapter:
         extra_kwargs = {
@@ -203,6 +205,15 @@ def comments_enabled(self) -> bool:
     def catalog_support(self) -> CatalogSupport:
         return CatalogSupport.UNSUPPORTED
 
+    @cached_property
+    def schema_differ(self) -> SchemaDiffer:
+        return SchemaDiffer(
+            **{
+                **self.SCHEMA_DIFFER_KWARGS,
+                **(self._schema_differ_overrides or {}),
+            }
+        )
+
     @classmethod
     def _casted_columns(
         cls,
@@ -1094,7 +1105,7 @@ def get_alter_operations(
         """
         return t.cast(
             t.List[TableAlterOperation],
-            self.SCHEMA_DIFFER.compare_columns(
+            self.schema_differ.compare_columns(
                 current_table_name,
                 self.columns(current_table_name),
                 self.columns(target_table_name),

sqlmesh/core/engine_adapter/bigquery.py

@@ -22,7 +22,7 @@
     set_catalog,
 )
 from sqlmesh.core.node import IntervalUnit
-from sqlmesh.core.schema_diff import SchemaDiffer, TableAlterOperation
+from sqlmesh.core.schema_diff import TableAlterOperation, NestedSupport
 from sqlmesh.utils import optional_import, get_source_columns_to_types
 from sqlmesh.utils.date import to_datetime
 from sqlmesh.utils.errors import SQLMeshError
@@ -67,8 +67,8 @@ class BigQueryEngineAdapter(InsertOverwriteWithMergeMixin, ClusteredByMixin, Row
     MAX_TABLE_COMMENT_LENGTH = 1024
     MAX_COLUMN_COMMENT_LENGTH = 1024
 
-    SCHEMA_DIFFER = SchemaDiffer(
-        compatible_types={
+    SCHEMA_DIFFER_KWARGS = {
+        "compatible_types": {
             exp.DataType.build("INT64", dialect=DIALECT): {
                 exp.DataType.build("NUMERIC", dialect=DIALECT),
                 exp.DataType.build("FLOAT64", dialect=DIALECT),
@@ -82,17 +82,17 @@ class BigQueryEngineAdapter(InsertOverwriteWithMergeMixin, ClusteredByMixin, Row
                 exp.DataType.build("DATETIME", dialect=DIALECT),
             },
         },
-        coerceable_types={
+        "coerceable_types": {
             exp.DataType.build("FLOAT64", dialect=DIALECT): {
                 exp.DataType.build("BIGNUMERIC", dialect=DIALECT),
             },
         },
-        support_coercing_compatible_types=True,
-        parameterized_type_defaults={
+        "support_coercing_compatible_types": True,
+        "parameterized_type_defaults": {
             exp.DataType.build("DECIMAL", dialect=DIALECT).this: [(38, 9), (0,)],
             exp.DataType.build("BIGDECIMAL", dialect=DIALECT).this: [(76.76, 38), (0,)],
         },
-        types_with_unlimited_length={
+        "types_with_unlimited_length": {
             # parameterized `STRING(n)` can ALTER to unparameterized `STRING`
             exp.DataType.build("STRING", dialect=DIALECT).this: {
                 exp.DataType.build("STRING", dialect=DIALECT).this,
@@ -102,9 +102,8 @@ class BigQueryEngineAdapter(InsertOverwriteWithMergeMixin, ClusteredByMixin, Row
                 exp.DataType.build("BYTES", dialect=DIALECT).this,
             },
         },
-        support_nested_operations=True,
-        support_nested_drop=False,
-    )
+        "nested_support": NestedSupport.ALL_BUT_DROP,
+    }
 
     @property
     def client(self) -> BigQueryClient:

sqlmesh/core/engine_adapter/clickhouse.py

@@ -15,7 +15,7 @@
     CommentCreationView,
     InsertOverwriteStrategy,
 )
-from sqlmesh.core.schema_diff import SchemaDiffer, TableAlterOperation
+from sqlmesh.core.schema_diff import TableAlterOperation
 from sqlmesh.utils import get_source_columns_to_types
 
 if t.TYPE_CHECKING:
@@ -37,7 +37,7 @@ class ClickhouseEngineAdapter(EngineAdapterWithIndexSupport, LogicalMergeMixin):
     SUPPORTS_REPLACE_TABLE = False
     COMMENT_CREATION_VIEW = CommentCreationView.COMMENT_COMMAND_ONLY
 
-    SCHEMA_DIFFER = SchemaDiffer()
+    SCHEMA_DIFFER_KWARGS = {}
 
     DEFAULT_TABLE_ENGINE = "MergeTree"
     ORDER_BY_TABLE_ENGINE_REGEX = "^.*?MergeTree.*$"

sqlmesh/core/engine_adapter/databricks.py

@@ -15,7 +15,7 @@
 )
 from sqlmesh.core.engine_adapter.spark import SparkEngineAdapter
 from sqlmesh.core.node import IntervalUnit
-from sqlmesh.core.schema_diff import SchemaDiffer
+from sqlmesh.core.schema_diff import NestedSupport
 from sqlmesh.engines.spark.db_api.spark_session import connection, SparkSessionConnection
 from sqlmesh.utils.errors import SQLMeshError, MissingDefaultCatalogError
 
@@ -34,15 +34,14 @@ class DatabricksEngineAdapter(SparkEngineAdapter):
     SUPPORTS_CLONING = True
     SUPPORTS_MATERIALIZED_VIEWS = True
     SUPPORTS_MATERIALIZED_VIEW_SCHEMA = True
-    SCHEMA_DIFFER = SchemaDiffer(
-        support_positional_add=True,
-        support_nested_operations=True,
-        support_nested_drop=True,
-        array_element_selector="element",
-        parameterized_type_defaults={
+    SCHEMA_DIFFER_KWARGS = {
+        "support_positional_add": True,
+        "nested_support": NestedSupport.ALL,
+        "array_element_selector": "element",
+        "parameterized_type_defaults": {
             exp.DataType.build("DECIMAL", dialect=DIALECT).this: [(10, 0), (0,)],
         },
-    )
+    }
 
     def __init__(self, *args: t.Any, **kwargs: t.Any) -> None:
         super().__init__(*args, **kwargs)

sqlmesh/core/engine_adapter/duckdb.py

@@ -18,7 +18,6 @@
     SourceQuery,
     set_catalog,
 )
-from sqlmesh.core.schema_diff import SchemaDiffer
 
 if t.TYPE_CHECKING:
     from sqlmesh.core._typing import SchemaName, TableName
@@ -29,11 +28,11 @@
 class DuckDBEngineAdapter(LogicalMergeMixin, GetCurrentCatalogFromFunctionMixin, RowDiffMixin):
     DIALECT = "duckdb"
     SUPPORTS_TRANSACTIONS = False
-    SCHEMA_DIFFER = SchemaDiffer(
-        parameterized_type_defaults={
+    SCHEMA_DIFFER_KWARGS = {
+        "parameterized_type_defaults": {
             exp.DataType.build("DECIMAL", dialect=DIALECT).this: [(18, 3), (0,)],
         },
-    )
+    }
     COMMENT_CREATION_TABLE = CommentCreationTable.COMMENT_COMMAND_ONLY
     COMMENT_CREATION_VIEW = CommentCreationView.COMMENT_COMMAND_ONLY
     SUPPORTS_CREATE_DROP_CATALOG = True

sqlmesh/core/engine_adapter/mixins.py

@@ -259,9 +259,9 @@ def _default_precision_to_max(
     ) -> t.Dict[str, exp.DataType]:
         # get default lengths for types that support "max" length
         types_with_max_default_param = {
-            k: [self.SCHEMA_DIFFER.parameterized_type_defaults[k][0][0]]
-            for k in self.SCHEMA_DIFFER.max_parameter_length
-            if k in self.SCHEMA_DIFFER.parameterized_type_defaults
+            k: [self.schema_differ.parameterized_type_defaults[k][0][0]]
+            for k in self.schema_differ.max_parameter_length
+            if k in self.schema_differ.parameterized_type_defaults
         }
 
         # Redshift and MSSQL have a bug where CTAS statements have non-deterministic types. If a LIMIT
@@ -270,7 +270,7 @@ def _default_precision_to_max(
         # and supports "max" length, we convert it to "max" length to prevent inadvertent data truncation.
         for col_name, col_type in columns_to_types.items():
             if col_type.this in types_with_max_default_param and col_type.expressions:
-                parameter = self.SCHEMA_DIFFER.get_type_parameters(col_type)
+                parameter = self.schema_differ.get_type_parameters(col_type)
                 type_default = types_with_max_default_param[col_type.this]
                 if parameter == type_default:
                     col_type.set("expressions", [exp.DataTypeParam(this=exp.var("max"))])

sqlmesh/core/engine_adapter/mssql.py

@@ -30,7 +30,6 @@
     SourceQuery,
     set_catalog,
 )
-from sqlmesh.core.schema_diff import SchemaDiffer
 from sqlmesh.utils import get_source_columns_to_types
 
 if t.TYPE_CHECKING:
@@ -54,8 +53,8 @@ class MSSQLEngineAdapter(
     COMMENT_CREATION_TABLE = CommentCreationTable.UNSUPPORTED
     COMMENT_CREATION_VIEW = CommentCreationView.UNSUPPORTED
     SUPPORTS_REPLACE_TABLE = False
-    SCHEMA_DIFFER = SchemaDiffer(
-        parameterized_type_defaults={
+    SCHEMA_DIFFER_KWARGS = {
+        "parameterized_type_defaults": {
             exp.DataType.build("DECIMAL", dialect=DIALECT).this: [(18, 0), (0,)],
             exp.DataType.build("BINARY", dialect=DIALECT).this: [(1,)],
             exp.DataType.build("VARBINARY", dialect=DIALECT).this: [(1,)],
@@ -67,12 +66,12 @@ class MSSQLEngineAdapter(
             exp.DataType.build("DATETIME2", dialect=DIALECT).this: [(7,)],
             exp.DataType.build("DATETIMEOFFSET", dialect=DIALECT).this: [(7,)],
         },
-        max_parameter_length={
+        "max_parameter_length": {
             exp.DataType.build("VARBINARY", dialect=DIALECT).this: 2147483647,  # 2 GB
             exp.DataType.build("VARCHAR", dialect=DIALECT).this: 2147483647,
             exp.DataType.build("NVARCHAR", dialect=DIALECT).this: 2147483647,
         },
-    )
+    }
     VARIABLE_LENGTH_DATA_TYPES = {"binary", "varbinary", "char", "varchar", "nchar", "nvarchar"}
 
     @property

sqlmesh/core/engine_adapter/mysql.py

@@ -19,7 +19,6 @@
     DataObjectType,
     set_catalog,
 )
-from sqlmesh.core.schema_diff import SchemaDiffer
 
 if t.TYPE_CHECKING:
     from sqlmesh.core._typing import SchemaName, TableName
@@ -40,8 +39,8 @@ class MySQLEngineAdapter(
     MAX_COLUMN_COMMENT_LENGTH = 1024
     SUPPORTS_REPLACE_TABLE = False
     MAX_IDENTIFIER_LENGTH = 64
-    SCHEMA_DIFFER = SchemaDiffer(
-        parameterized_type_defaults={
+    SCHEMA_DIFFER_KWARGS = {
+        "parameterized_type_defaults": {
             exp.DataType.build("BIT", dialect=DIALECT).this: [(1,)],
             exp.DataType.build("BINARY", dialect=DIALECT).this: [(1,)],
             exp.DataType.build("DECIMAL", dialect=DIALECT).this: [(10, 0), (0,)],
@@ -52,7 +51,7 @@ class MySQLEngineAdapter(
             exp.DataType.build("DATETIME", dialect=DIALECT).this: [(0,)],
             exp.DataType.build("TIMESTAMP", dialect=DIALECT).this: [(0,)],
         },
-    )
+    }
 
     def get_current_catalog(self) -> t.Optional[str]:
         """Returns the catalog name of the current connection."""