From 7601edf9b6027c8b87304f455f4906728f4a368f Mon Sep 17 00:00:00 2001
From: Phillip Cloud <417981+cpcloud@users.noreply.github.com>
Date: Mon, 13 May 2024 07:51:18 -0400
Subject: [PATCH] fix(snowflake): ensure that timestamp conversion from parquet
 files is correct

---
 ibis/backends/snowflake/__init__.py | 41 +++++++++++++++++++++--------
 1 file changed, 30 insertions(+), 11 deletions(-)

diff --git a/ibis/backends/snowflake/__init__.py b/ibis/backends/snowflake/__init__.py
index 6627600337626..d8ed2814f7687 100644
--- a/ibis/backends/snowflake/__init__.py
+++ b/ibis/backends/snowflake/__init__.py
@@ -1057,22 +1057,41 @@ def read_parquet(
         #
         # see
         # https://community.snowflake.com/s/article/How-to-load-logical-type-TIMESTAMP-data-from-Parquet-files-into-Snowflake
+        type_mapper = self.compiler.type_mapper
         names_types = [
-            (
-                name,
-                self.compiler.type_mapper.to_string(typ),
-                typ.nullable,
-                typ.is_timestamp(),
-            )
-            for name, typ in schema.items()
+            (name, type_mapper.to_string(typ), typ) for name, typ in schema.items()
         ]
+
         snowflake_schema = ", ".join(
-            f"{sg.to_identifier(col, quoted=quoted)} {typ}{' NOT NULL' * (not nullable)}"
-            for col, typ, nullable, _ in names_types
+            f"{sg.to_identifier(col, quoted=quoted)} {typ}{' NOT NULL' * (not dtype.nullable)}"
+            for col, typ, dtype in names_types
         )
+
+        def make_column(value, typ, dtype):
+            # don't do anything special if we're not handling a timestamp
+            if not dtype.is_timestamp():
+                return f"{value}::{typ}"
+
+            comp = self.compiler
+            f = comp.f
+
+            # apparently Snowflake sometimes treats Parquet timestamps as
+            # variant strings or variant numbers, so check for which one we
+            # have using is_integer
+            value = sge.Var(this=value)
+            return comp.if_(
+                sg.or_(f.is_integer(value), f.is_double(value), f.is_decimal(value)),
+                f.to_timestamp(
+                    comp.cast(value, dt.int),
+                    # seconds (dtype.scale == 0) is the snowflake default
+                    dtype.scale or 0,
+                ),
+                f.try_to_timestamp(comp.cast(value, dt.string)),
+            ).sql(self.dialect)
+
         cols = ", ".join(
-            f"$1:{col}{'::VARCHAR' * is_timestamp}::{typ}"
-            for col, typ, _, is_timestamp in names_types
+            make_column(f"$1:{col}", typ=typ, dtype=dtype)
+            for col, typ, dtype in names_types
         )
 
         stmts = [