feat(snowflake): make literal maps and params work

ibis/backends/snowflake/registry.py

@@ -1,10 +1,11 @@
 from __future__ import annotations
 
 import itertools
+import json
 
 import numpy as np
 import sqlalchemy as sa
-from snowflake.sqlalchemy.custom_types import VARIANT
+from snowflake.sqlalchemy import VARIANT
 
 import ibis.expr.operations as ops
 from ibis.backends.base.sql.alchemy.registry import (
@@ -52,7 +53,7 @@ def _literal(t, op):
         return sa.func.array_construct(*value)
     elif dtype.is_map():
         return sa.func.object_construct_keep_null(
-            *zip(itertools.chain.from_iterable(value.items()))
+            *itertools.chain.from_iterable(value.items())
         )
     return _postgres_literal(t, op)
 
@@ -112,6 +113,17 @@ def _array_slice(t, op):
     return sa.func.array_slice(t.translate(op.arg), start, stop)
 
 
+def _map(_, op):
+    if not (
+        isinstance(keys := op.keys, ops.Literal)
+        and isinstance(values := op.values, ops.Literal)
+    ):
+        raise TypeError("Both keys and values of an `ibis.map` call must be literals")
+
+    obj = dict(zip(keys.value, values.value))
+    return sa.func.to_object(sa.func.parse_json(json.dumps(obj, separators=",:")))
+
+
 _SF_POS_INF = sa.cast(sa.literal("Inf"), sa.FLOAT)
 _SF_NEG_INF = -_SF_POS_INF
 _SF_NAN = sa.cast(sa.literal("NaN"), sa.FLOAT)
@@ -201,7 +213,7 @@ def _array_slice(t, op):
         ops.ArrayColumn: lambda t, op: sa.func.array_construct(
             *map(t.translate, op.cols)
         ),
-        ops.ArraySlice: _array_slice,
+        ops.Map: _map,
     }
 )
 

ibis/backends/tests/test_map.py

@@ -124,12 +124,16 @@ def test_literal_map_get_broadcast(backend, alltypes, df):
     backend.assert_series_equal(result, expected)
 
 
-@pytest.mark.notyet(["snowflake"])
-def test_map_construction(con, alltypes, df):
+def test_map_construct_dict(con):
     expr = ibis.map(['a', 'b'], [1, 2])
     result = con.execute(expr.name('tmp'))
     assert result == {'a': 1, 'b': 2}
 
+
+@pytest.mark.notimpl(
+    ["snowflake"], reason="unclear how to implement two arrays -> object construction"
+)
+def test_map_construct_array_column(con, alltypes, df):
     expr = ibis.map(ibis.array([alltypes.string_col]), ibis.array([alltypes.int_col]))
     result = con.execute(expr)
     expected = df.apply(lambda row: {row['string_col']: row['int_col']}, axis=1)

ibis/backends/tests/test_param.py

@@ -58,7 +58,7 @@ def test_timestamp_accepts_date_literals(alltypes):
     assert expr.compile(params=params) is not None
 
 
-@pytest.mark.notimpl(["dask", "datafusion", "impala", "pandas", "pyspark", "snowflake"])
+@pytest.mark.notimpl(["dask", "datafusion", "impala", "pandas", "pyspark"])
 @pytest.mark.never(
     ["mysql", "sqlite", "mssql"], reason="backend will never implement array types"
 )
@@ -84,17 +84,7 @@ def test_scalar_param_struct(con):
 
 
 @pytest.mark.notimpl(
-    [
-        "clickhouse",
-        "datafusion",
-        # TODO: duckdb maps are tricky because they are multimaps
-        "duckdb",
-        "impala",
-        "pyspark",
-        "snowflake",
-        "polars",
-        "trino",
-    ]
+    ["clickhouse", "datafusion", "duckdb", "impala", "pyspark", "polars", "trino"]
 )
 @pytest.mark.never(
     ["mysql", "sqlite", "mssql"],