feat: add working middleware to dbt proxy to sync alter table modify …

…column comment stmts to manifest

src/dbt_osmosis/sql/proxy.py

@@ -1,22 +1,25 @@
+# pyright: reportMissingTypeStubs=false, reportAny=false, reportImplicitOverride=false, reportUnknownMemberType=false, reportUnusedImport=false, reportUnknownParameterType=false
 """Proxy server experiment that any MySQL client (including BI tools) can connect to."""
 
-# pyright: reportMissingTypeStubs=false, reportAny=false, reportImplicitOverride=false, reportUnknownMemberType=false, reportUnusedImport=false, reportUnknownParameterType=false
 import asyncio
 import functools
 import re
 import typing as t
 from collections import defaultdict
 from collections.abc import Iterator
+from itertools import chain
 
 from dbt.adapters.contracts.connection import AdapterResponse
 from mysql_mimic import MysqlServer, Session
 from mysql_mimic.errors import MysqlError
+from mysql_mimic.results import AllowedResult
 from mysql_mimic.schema import (
     Column,
     InfoSchema,
     dict_depth,  # pyright: ignore[reportUnknownVariableType, reportPrivateLocalImportUsage]
     info_schema_tables,
 )
+from mysql_mimic.session import Query
 from sqlglot import exp
 
 import dbt_osmosis.core.logger as logger
@@ -29,38 +32,8 @@
     execute_sql_code,
 )
 
-# TODO: this doesn't capture comment body consistently
 ALTER_MODIFY_COL_PATTERN = re.compile(
-    r"""
-    ^\s*                              # start, allow leading whitespace
-    ALTER\s+TABLE                     # "ALTER TABLE" (case-insensitive via flags=)
-    \s+
-    (?:                               # optional schema part:
-       (?:"(?P<schema>[^"]+)"         # double-quoted schema
-         |(?P<schema_unquoted>\w+)    # or unquoted schema
-       )
-       \s*\.
-    )?
-    (?:"(?P<table>[^"]+)"             # table in double-quotes
-      |(?P<table_unquoted>\w+)        # or unquoted table
-    )
-    \s+
-    MODIFY\s+COLUMN\s+                # "MODIFY COLUMN" (case-insensitive via flags=)
-    (?:"(?P<column>[^"]+)"            # column in double-quotes
-      |(?P<column_unquoted>\w+)       # or unquoted column
-    )
-    .*?                               # lazily consume anything until we might see COMMENT
-    (?:                                # optional comment group
-       COMMENT\s+                     # must have "COMMENT" then space(s)
-       (["'])                         # capture the quote symbol in group 1
-       (?P<comment>
-         (?:.|[^"'])*              # any escaped char or anything that isn't ' or "
-       )
-       \1                             # match the same quote symbol (group 1)
-    )?
-    \s*;?\s*$                         # optional whitespace, optional semicolon
-    """,
-    flags=re.IGNORECASE | re.DOTALL | re.VERBOSE,
+    r"(?i)(?:/\*.*?\*/\s*)?ALTER TABLE\s+(?:(?P<schema>[^\s\.]+)\.)?(?P<table>[^\s\.]+)\s+MODIFY COLUMN\s+(?P<column>[^\s]+)\s+.*?COMMENT\s+'(?P<comment>[^']*)';?"
 )
 
 
@@ -70,70 +43,79 @@ def parse_alter_modify_column(sql: str) -> dict[str, str] | None:
         ALTER TABLE schema.table MODIFY COLUMN col TYPE ... COMMENT 'some text';
 
     Returns None if the pattern does not match, otherwise a dict with:
-       {
-         "schema": ... or None,
-         "table":  ...,
-         "column": ...,
-         "comment": ... or None
-       }
+       {"schema": ..., "table":  ..., "column": ..., "comment": ...}
     """
     match = ALTER_MODIFY_COL_PATTERN.match(sql)
     if not match:
         return None
-
-    # Because we have both quoted and unquoted named groups, pick whichever matched:
-    schema = match.group("schema") or match.group("schema_unquoted")
-    table = match.group("table") or match.group("table_unquoted")
-    column = match.group("column") or match.group("column_unquoted")
-    comment = match.group("comment")  # can be None if COMMENT was not present
-
+    schema = match.group("schema")
+    table = match.group("table")
+    column = match.group("column")
+    comment = match.group("comment")
+    if not all((schema, table, column, comment)):
+        return None
     return {"schema": schema, "table": table, "column": column, "comment": comment}
 
 
 class QueryException(MysqlError):
     def __init__(self, response: AdapterResponse) -> None:
-        self.response: AdapterResponse = response
         super().__init__(response._message)  # pyright: ignore[reportPrivateUsage]
+        self.response: AdapterResponse = response
 
 
 class DbtSession(Session):
     def __init__(self, project: DbtProjectContext, *args: t.Any, **kwargs: t.Any) -> None:
-        self.project: DbtProjectContext = project
         super().__init__(*args, **kwargs)
+        self.project: DbtProjectContext = project
+        self.middlewares.append(self._alter_column_middleware)
 
     def _parse(self, sql: str) -> list[exp.Expression]:
         if _has_jinja(sql):
             node = compile_sql_code(self.project, sql)
             sql = node.compiled_code or node.raw_code
         return [e for e in self.dialect().parse(sql) if e]
 
-    async def query(self, expression: exp.Expression, sql: str, attrs: dict[str, t.Any]):
+    async def _alter_column_middleware(self, q: Query) -> AllowedResult:
+        """Intercept ALTER TABLE ... MODIFY COLUMN ... COMMENT statements
+
+        This middleware will update the column description in the dbt project manifest. Eventually
+        it could use our Yaml context class to actually write the changes to disk.
+        """
+        if isinstance(q.expression, exp.Command):
+            lower_sql = q.sql.lower()
+            likely_alter_column = all(
+                k in lower_sql for k in ("alter", "modify", "column", "comment")
+            )
+            if doc_update_req := (likely_alter_column and parse_alter_modify_column(q.sql)):
+                rel = (doc_update_req["schema"], doc_update_req["table"])
+                for node in chain(
+                    self.project.manifest.sources.values(), self.project.manifest.nodes.values()
+                ):
+                    if rel == (node.schema, node.name):
+                        for column in node.columns.values():
+                            if column.name == doc_update_req["column"]:
+                                column.description = doc_update_req["comment"]
+                                break
+            return [], []
+        return await q.next()
+
+    async def query(
+        self, expression: exp.Expression, sql: str, attrs: dict[str, t.Any]
+    ) -> AllowedResult:
         logger.info("Query: %s", sql)
-        if isinstance(expression, exp.Command):
-            cmd = f"{expression.this} {expression.expression}"
-            doc_update = "alter" in sql.lower() and parse_alter_modify_column(cmd)
-            if doc_update:
-                logger.info("Will update doc: %s", doc_update)
-            else:
-                logger.info("Ignoring command %s", sql)
-            return (), []  # pyright: ignore[reportUnknownVariableType]
         resp, table = await asyncio.to_thread(
             execute_sql_code, self.project, expression.sql(dialect=self.project.adapter.type())
         )
         if resp.code:
             raise QueryException(resp)
-        logger.info(table)
-        return [
-            t.cast(tuple[t.Any], row.values()) for row in t.cast(tuple[t.Any], table.rows.values())
-        ], t.cast(tuple[str], table.column_names)
+        rows = t.cast(tuple[t.Any], table.rows.values())
+        return [row.values() for row in rows], t.cast(tuple[str], table.column_names)
 
     async def schema(self):
         schema: defaultdict[str, dict[str, dict[str, tuple[str, str | None]]]] = defaultdict(dict)
-        for source in self.project.manifest.sources.values():
-            schema[source.schema][source.name] = {
-                c.name: (c.data_type or "UNKOWN", c.description) for c in source.columns.values()
-            }
-        for node in self.project.manifest.nodes.values():
+        for node in chain(
+            self.project.manifest.sources.values(), self.project.manifest.nodes.values()
+        ):
             schema[node.schema][node.name] = {
                 c.name: (c.data_type or "UNKOWN", c.description) for c in node.columns.values()
             }