In D8426#219120, @vlorentz wrote:

Here is a way to do it:

diff --git a/swh/graphql/resolvers/resolvers.py b/swh/graphql/resolvers/resolvers.py
index 60992bb..1b0e93f 100644
--- a/swh/graphql/resolvers/resolvers.py
+++ b/swh/graphql/resolvers/resolvers.py
@@ -93,6 +93,8 @@ def snapshot_resolver(
 def visit_snapshot_resolver(
     obj, info: GraphQLResolveInfo, **kw
 ) -> rs.snapshot.VisitSnapshotNode:
+    if obj.snapshotSWHID is None:
+        return None
     resolver = get_node_resolver("visit-snapshot")
     return resolver(obj, info, **kw)
 
diff --git a/swh/graphql/tests/functional/test_visit_status.py b/swh/graphql/tests/functional/test_visit_status.py
index a250a9a..d83861a 100644
--- a/swh/graphql/tests/functional/test_visit_status.py
+++ b/swh/graphql/tests/functional/test_visit_status.py
@@ -28,7 +28,7 @@ def test_get_visit_status(client):
         visit.origin,
         visit.visit,
     )
-    data, _ = get_query_response(client, query_str)
+    data, err = get_query_response(client, query_str)
     result_status = get_visit_status()[3]
     assert data["visit"]["status"]["nodes"][0] == {
         "date": result_status.date.isoformat(),
@@ -36,6 +36,7 @@ def test_get_visit_status(client):
         "status": result_status.status,
         "type": result_status.type,
     }
+    assert err is None
 
 
 def test_get_visit_missing_snapshot(client):
@@ -67,7 +68,4 @@ def test_get_visit_missing_snapshot(client):
         "status": result_status.status,
         "type": result_status.type,
     }
-    assert len(err) == 1
-    assert (
-        "Cannot return null for non-nullable field Snapshot.swhid" in err[0]["message"]
-    )
+    assert err is None

update error to raise

In D8426#219093, @vlorentz wrote:

Sure, but in that example, the error is because for some reason, they could not get the value of name for an object that exists. The error in the example even reflects this, as it mentions "character with ID 1002".

In our case, that object does not exist at all.

In D8426#219084, @vlorentz wrote:

The section you linked mentions null values in Non-Null fields, but not of null values for objects with a Non-Null field.

Are you sure this is disallowed by the standard, not "only" a bug in the GraphQL library we use?

In D8426#219084, @vlorentz wrote:

The section you linked mentions null values in Non-Null fields, but not of null values for objects with a Non-Null field.

Are you sure this is disallowed by the standard, not "only" a bug in the GraphQL library we use?

In D8426#219070, @vlorentz wrote:

In D8426#219051, @jayeshv wrote:

The issue is snapshot has non nullable fields (like SWHID) and GraphQL standard will not allow a
null for an object with non nullable field without an error.

I can't find what part of the standard disallows it, nor any mention one way or another outside the standard. Any pointer?

In D8426#219077, @anlambert wrote:

In D8426#219076, @jayeshv wrote:

> I do not get the issue here, using the following changes on top of your diff, it works as expected:
>

It will work as expected in the data part. But there will be an error associated, that we think is unnecessary.
With your changes, it will be a "NoneType' object has no attribute 'object_id" instead of an ObjectError

Could you write a snippet query to reproduce the error ? I still do not get what's wrong here.

> I do not get the issue here, using the following changes on top of your diff, it works as expected:
>

It will work as expected in the data part. But there will be an error associated, that we think is unnecessary.
With your changes, it will be a "NoneType' object has no attribute 'object_id" instead of an ObjectError

In D8426#219043, @vlorentz wrote:

Why have two different types, instead of making the snapshot field nullable, like this?

type VisitStatus {
  status: String!
  date: DateTime!
  type: String
  snapshot: Snapshot
}

@vlorentz The right fix is to add a new visit status type to the schema like below.

remove ununsed test

use parametrize in directory entry tests

remove wrong code

address review comments

rebase

In D8426#218850, @vlorentz wrote:

Should this really be an error? Isn't it enough to set snapshot to null?

In D8426#218850, @vlorentz wrote:

Should this really be an error? Isn't it enough to set snapshot to null?

This is a bit tricky to fix.
Snapshot is available only in a visit status with 'full' status. But there is only one visit-status object and it has a field snapshot.
Did a fix that returns a sensible error instead of a creepy one.

update the commit message

In D8396#218275, @vlorentz wrote:

As this allows nested paths, then the cost calculator needs to take into account the number of / in the path, as each of them requires two joins in the database

add a custom validator

address review comments

Unit tests

post to sentry

doc improvements

doc string update

fix tests

Using multipliers

rebase

97aa757412b725b08cc123620de9cd121f44778b

Changed dev to asgi in the the doc and commit message

fix typo