feat(bigquery): implement CountDistinctStar (#9470)

Co-authored-by: Phillip Cloud <417981+cpcloud@users.noreply.github.com>

ibis/backends/bigquery/compiler.py

@@ -36,7 +36,6 @@ class BigQueryCompiler(SQLGlotCompiler):
     )
 
     UNSUPPORTED_OPS = (
-        ops.CountDistinctStar,
         ops.DateDiff,
         ops.ExtractAuthority,
         ops.ExtractUserInfo,
@@ -645,6 +644,25 @@ def visit_CountStar(self, op, *, arg, where):
             return self.f.countif(where)
         return self.f.count(STAR)
 
+    def visit_CountDistinctStar(self, op, *, where, arg):
+        # Bigquery does not support count(distinct a,b,c) or count(distinct (a, b, c))
+        # as expressions must be "groupable":
+        # https://cloud.google.com/bigquery/docs/reference/standard-sql/query-syntax#group_by_grouping_item
+        #
+        # Instead, convert the entire expression to a string
+        # SELECT COUNT(DISTINCT concat(to_json_string(a), to_json_string(b)))
+        # This works with an array of datatypes which generates a unique string
+        # https://cloud.google.com/bigquery/docs/reference/standard-sql/json_functions#json_encodings
+        row = sge.Concat(
+            expressions=[
+                self.f.to_json_string(sg.column(x, quoted=self.quoted))
+                for x in op.arg.schema.keys()
+            ]
+        )
+        if where is not None:
+            row = self.if_(where, row, NULL)
+        return self.f.count(sge.Distinct(expressions=[row]))
+
     def visit_Degrees(self, op, *, arg):
         return self._pudf("degrees", arg)
 

ibis/backends/tests/test_aggregation.py

@@ -713,7 +713,7 @@ def test_arbitrary(backend, alltypes, df, filtered):
     ],
 )
 @pytest.mark.notyet(
-    ["bigquery", "druid", "mssql", "oracle", "sqlite", "flink"],
+    ["druid", "mssql", "oracle", "sqlite", "flink"],
     raises=(
         OracleDatabaseError,
         com.UnsupportedOperationError,