Fix handling of semicolons inside BEGIN...END blocks (fixes #809).

Even though the logic in StatementSplitter gets more and more fragile, this approach is a bit simpler than the proposal in #810 as it focusses only on that part.

AI helped here.

Author: andialbrecht

CHANGELOG

@@ -27,6 +27,7 @@ Bug Fixes
   comments (issue801, pr803 by stropysh).
 * Fix splitting statements with IF EXISTS/IF NOT EXISTS inside BEGIN...END
   blocks (issue812).
+* Fix splitting on semicolons inside BEGIN...END blocks (issue809).
 
 
 Release 0.5.3 (Dez 10, 2024)

sqlparse/engine/statement_splitter.py

@@ -20,6 +20,7 @@ def _reset(self):
         self._in_case = False
         self._is_create = False
         self._begin_depth = 0
+        self._seen_begin = False
 
         self.consume_ws = False
         self.tokens = []
@@ -54,6 +55,7 @@ def _change_splitlevel(self, ttype, value):
 
         if unified == 'BEGIN':
             self._begin_depth += 1
+            self._seen_begin = True
             if self._is_create:
                 # FIXME(andi): This makes no sense.  ## this comment neither
                 return 1
@@ -106,9 +108,24 @@ def process(self, stream):
             # When implementing a language toggle, it's not only to add
             # keywords it's also to change some rules, like this splitting
             # rule.
-            if (self.level <= 0 and ttype is T.Punctuation and value == ';') \
-                    or (ttype is T.Keyword and value.split()[0] == 'GO'):
+            # Issue809: Ignore semicolons inside BEGIN...END blocks, but handle
+            # standalone BEGIN; as a transaction statement
+            if ttype is T.Punctuation and value == ';':
+                # If we just saw BEGIN; then this is a transaction BEGIN,
+                # not a BEGIN...END block, so decrement depth
+                if self._seen_begin:
+                    self._begin_depth = max(0, self._begin_depth - 1)
+                self._seen_begin = False
+                # Split on semicolon if not inside a BEGIN...END block
+                if self.level <= 0 and self._begin_depth == 0:
+                    self.consume_ws = True
+            elif ttype is T.Keyword and value.split()[0] == 'GO':
                 self.consume_ws = True
+            elif (ttype not in (T.Whitespace, T.Comment.Single, T.Comment.Multiline)
+                  and not (ttype is T.Keyword and value.upper() == 'BEGIN')):
+                # Reset _seen_begin if we see a non-whitespace, non-comment token
+                # but not for BEGIN itself (which just set the flag)
+                self._seen_begin = False
 
         # Yield pending statement (if any)
         if self.tokens and not all(t.is_whitespace for t in self.tokens):

tests/test_split.py

@@ -222,3 +222,35 @@ def test_split_if_exists_in_begin_end():  # issue812
     assert len(stmts) == 2
     assert 'CREATE TASK' in stmts[0]
     assert 'EXECUTE TASK' in stmts[1]
+
+
+def test_split_begin_end_semicolons():  # issue809
+    # Semicolons inside BEGIN...END blocks should not split statements
+    sql = """WITH
+FUNCTION meaning_of_life()
+  RETURNS tinyint
+  BEGIN
+    DECLARE a tinyint DEFAULT CAST(6 as tinyint);
+    DECLARE b tinyint DEFAULT CAST(7 as tinyint);
+    RETURN a * b;
+  END
+SELECT meaning_of_life();"""
+    stmts = sqlparse.split(sql)
+    assert len(stmts) == 1
+    assert 'WITH' in stmts[0]
+    assert 'SELECT meaning_of_life()' in stmts[0]
+
+
+def test_split_begin_end_procedure():  # issue809
+    # Test with CREATE PROCEDURE (BigQuery style)
+    sql = """CREATE OR REPLACE PROCEDURE mydataset.create_customer()
+BEGIN
+  DECLARE id STRING;
+  SET id = GENERATE_UUID();
+  INSERT INTO mydataset.customers (customer_id)
+    VALUES(id);
+  SELECT FORMAT("Created customer %s", id);
+END;"""
+    stmts = sqlparse.split(sql)
+    assert len(stmts) == 1
+    assert 'CREATE OR REPLACE PROCEDURE' in stmts[0]