Hey @a631807682 thank you so much for your review. My responses to your questions are as below.

Why is Array special, what about Slice? I think we shouldn't provide unique compatibility for external packages.

The *uuid.UUID resolves to a reflect.Ptr to a reflect.Array. This is because the uuid.UUID from the Google UUID package is essentially a 128 bit (16 byte) array. Yes, I agree with you that providing unique compatibility for external packages is not a good idea and can bloat up the project, but considering the widespread use of the UUID package in the Go community, can we support it? It will be really helpful for several developers.

Actually the issue is whether we should update nil values ​​to the original object when performing Update.

Yes, we should surely update nil values in the original object when performing Update, since original object needs to be consistent with the underlying database changes performed in the update operation. It's usually a best practice in ORMs to keep the model entity object (i.e. the original object) consistent with the underlying database, as developers may continue using the original object in code for further operations after the update, assuming that it's already in sync with the database.

If so, how should the Embedded object handle it? Actually this PR still doesn't work in Embedded objects.

This PR specifically fixes only the issue with nil assignment for *uuid.UUID. So it doesn't have any code for handling of nil assignment in embedded objects.

For this issue, my first instinct is to use Customize Data Types:

package datatypes

import (
	"database/sql/driver"

	"github.com/google/uuid"
	"gorm.io/gorm/schema"
)

type UUID uuid.UUID

func (UUID) GormDataType() string {
	return string(schema.String)
}

func (u *UUID) Scan(value interface{}) error {
	var result uuid.UUID
	if err := result.Scan(value); err != nil {
		return err
	}
	*u = UUID(result)
	return nil
}

func (u UUID) Value() (driver.Value, error) {
	return uuid.UUID(u).Value()
}

func (u UUID) String() string {
	return uuid.UUID(u).String()
}

then use datatypes.UUID instead of uuid.UUID:

var uuidPtr *datatypes.UUID = nil
DB.Model(&p).Updates(map[string]interface{}{"unique_id": uuidPtr}).Error
// UPDATE "p" SET "unique_id"=NULL WHERE "id" = 1

but considering the widespread use of the UUID package in the Go community, can we support it? It will be really helpful for several developers.

Yes, we should try to support it, but I would like it to be in a more general way.

This PR specifically fixes only the issue with nil assignment for *uuid.UUID. So it doesn't have any code for handling of nil assignment in embedded objects.

This test describes the problem with updating the original object in an embedded model, and the behavior that this PR breaks.

The problem here is that to set the value of a field of an embedded object, you need to instantiate the object, even if it is a nil value.

In my opinion, a more general solution may be to use Scanner and Valuer as @iTanken said (if feasible, it can be included in the https://github.com/go-gorm/datatypes project), or to check whether the embedded object of pointer type is zero value after setting all the fields of the object.

	type Author struct {
		ID          string
		Name        string
		Email       string
		Age         int
		Content     Content
		ContentPtr  *Content
		Birthday    time.Time
		BirthdayPtr *time.Time
+		UUID        *uuid.UUID
	}

Hey @a631807682 @iTanken, thank you both for your comments.

Yes I totally agree that we should keep it more general, so as per both your suggestions above, I've raised this PR on the gorm/datatypes repo to add UUID as a datatype for gorm.

@a631807682 I've also added handling for embedded structs and have included the test case that you provided above (refer here). That test case is passing now.

Also have replaced uuid.UUID with datatypes.UUID in this PR, but I haven't committed it since it will break until the datatypes PR has been merged. You can review the diff below. I suppose we can keep this PR open until the datatypes PR is reviewed and merged, after which I can add the below diff to this PR.

Note: PR check reviewdog/runner/golangci-lint has been failing due to an issue with filtering, refer this issue for more information.

diff --git a/tests/update_test.go b/tests/update_test.go
index a4f902b..84d41c9 100644
--- a/tests/update_test.go
+++ b/tests/update_test.go
@@ -9,6 +9,7 @@ import (
 	"time"
 
 	"github.com/google/uuid"
+	"gorm.io/datatypes"
 	"gorm.io/gorm"
 	"gorm.io/gorm/clause"
 	"gorm.io/gorm/utils"
@@ -185,7 +186,7 @@ func TestUpdates(t *testing.T) {
 	user3.Age += 100
 	AssertObjEqual(t, user4, user3, "UpdatedAt", "Age")
 
-	// Updates() with map and uuid.UUID - Case 1 - Update with UUID value
+	// Updates() with map and datatypes.UUID - Case 1 - Update with UUID value
 	uuidVal, uuidErr := uuid.NewUUID()
 	if uuidErr != nil {
 		t.Errorf("No error should occur while generating UUID, but got %v", uuidErr)
@@ -198,8 +199,8 @@ func TestUpdates(t *testing.T) {
 	// Expecting the model object (user4) to reflect the UUID value assignment.
 	AssertEqual(t, user4.UserUUID, uuidVal)
 
-	// Updates() with map and uuid.UUID - Case 2 - Update with UUID nil pointer
-	var nilUUIDPtr *uuid.UUID = nil
+	// Updates() with map and datatypes.UUID - Case 2 - Update with UUID nil pointer
+	var nilUUIDPtr *datatypes.UUID = nil
 	uuidErr = tx.Updates(map[string]interface{}{"user_uuid": nilUUIDPtr}).Error
 	if uuidErr != nil {
 		t.Errorf("No error should occur while updating with nil UUID pointer, but got %v", uuidErr)
@@ -207,12 +208,13 @@ func TestUpdates(t *testing.T) {
 	// Expecting the model object (user4) to reflect the UUID nil pointer assignment.
 	AssertEqual(t, user4.UserUUID, nilUUIDPtr)
 
-	// Updates() with map and uuid.UUID - Case 3 - Update with a non-nil UUID pointer
+	// Updates() with map and datatypes.UUID - Case 3 - Update with a non-nil UUID pointer
 	uuidVal2, uuidErr := uuid.NewUUID()
 	if uuidErr != nil {
 		t.Errorf("No error should occur while generating UUID, but got %v", uuidErr)
 	}
-	var nonNilUUIDPtr *uuid.UUID = &uuidVal2
+	castedUUIDVal2 := datatypes.UUID(uuidVal2)
+	var nonNilUUIDPtr *datatypes.UUID = &castedUUIDVal2
 	uuidErr = tx.Updates(map[string]interface{}{"user_uuid": nonNilUUIDPtr}).Error
 	if uuidErr != nil {
 		t.Errorf("No error should occur while updating with non-nil UUID pointer, but got %v", uuidErr)
diff --git a/utils/tests/models.go b/utils/tests/models.go
index 7690641..060467c 100644
--- a/utils/tests/models.go
+++ b/utils/tests/models.go
@@ -4,7 +4,7 @@ import (
 	"database/sql"
 	"time"
 
-	"github.com/google/uuid"
+	"gorm.io/datatypes"
 	"gorm.io/gorm"
 )
 
@@ -31,7 +31,7 @@ type User struct {
 	Languages []Language `gorm:"many2many:UserSpeak;"`
 	Friends   []*User    `gorm:"many2many:user_friends;"`
 	Active    bool
-	UserUUID  *uuid.UUID
+	UserUUID  *datatypes.UUID
 }
 
 type Account struct {

Hi! As per our discussion in the above comment, I've added a datatype datatypes.UUID (as well as datatypes.BinUUID), and both have been released in datatypes v1.2.2.

I've updated this PR to use datatypes v1.2.2 and also replaced uuid.UUID with datatypes.UUID as per discussion. Kindly review and let me know if any further changes are required, thank you! 🙂

sqlserver tests have been failing because the GitHub Actions test containers are failing to start. I tried with increased timeouts and retries, and it still fails after trying for around 10 mins.

Will dig deeper on why this is happening, as I see similar failures in #7198 as well. If anyone has any context on this issue, please share. Thanks!