Convex Ents: how to express optional reference in 1 to many relationship?
From this documentation https://labs.convex.dev/convex-ents/schema, I'm trying to figure out a common use case, how to express zero-or-many?
Looking at the documentation's example, how can I express that a user can have zero or many messages?
14 Replies
Adding , { optional: true } works for edge() but not edges(). Why?
I think I'm mostly done with my migration to Ents, but I am stuck on this issue.
This is the schema validation error message I get since I cannot figure out how to express a zero-or-many.
✖ Schema validation failed
Document with ID "kn7591x3ngxfc7gyw8zgx9c1f96w4abc" in table "analyzeSessions" does not match the schema: Value does not match validator.
Path: .userId
Value: ""
Validator: v.id("users")
If I add a optional property,
Then I will get this error in npx convex dev.
users table has
.edges('analyzeSessions', { ref: true })
I have found that although this error message appears in the npx convex dev terminal, there is no runtime error or log event when running the application and inserting documents.
In other words, the document inserts successfully, with a "" value in userId, despite this npx convex dev error message:
So this setup should work, its by default optional because the remote object stores the list of ids in this case the messages table has the list, so I dont ever specify optional
And i just always clear the table after changing shape, i don't think ive been able to insert something with schema validation on that would cause a schema error.
Thanks, but after I clear the table, and insert a new document, I get that error again
It's a blocker because my
npm run build will error
I think the problem is that my insert() function is inserting a "" value.
Is it possible to insert a null value at all? How about with with Convex Ents? If so, how about for a userId column that is supposed to be a foreign key to users table?I've had to be careful in designing it so that i only insert the downstream records after the upstream has been created.
Inserting blank would be bad for sure. We spent at least a week working on schema and i still have refactored so many times not catching some of these things. Or wanting to use ents in more cases then i should.
So, is it expected behavior that convex ents cannot store null? A zero-to-many relationship is a common need
So if you have a zero to many relationship, are you forced to exempt the table out of Ents (and so use vanilla convex for that table)?
By the way, I’ve tried to insert null by skipping the property in the ctx.table insert(), but “” will still be in the database
Looking at the documentation's example, how can I express that a user can have zero or many messages?This is already what the code does. Are you instead looking for "message can have an author or not"? In that case you'd be running into: https://github.com/xixixao/convex-ents/issues/5 And the workaround is to use a many:many edge, and enforce the author uniqueness in your code.
GitHub
1:many edges require single cardinality end to be required · Issue ...
This will require more configuration, especially for cascading deletes. We'll hold off on adding this until we get people asking for it.
Yes, I think i am looking for "message can have an author or not". In other words, I'm looking for a zero-to-many relationship, or a optional one-to-many relationship.
I think the many:many edge can be a workaround I can do for now. Or limit my app's functionality to require user sign-in on feature that insert to a table with a userId column.
Thanks Michal, It's an unfortunate limitation to not have optional one-to-many, which is a very common relationship
Another workaround I considered is to make two separate tables, one with a userId column and one without a userId column. That way, I can have a normal looking one-to-many edge.
This approach can be pretty complicated since I want a non-logged-in user to be able to see a significant amount of my app before landing on a login wall
This is what anonymous users are for, that way you already have a user ID. They have been built into Convex Auth but not entirely finished yet.
null wouldnt' probably work at all for you, if that's the scenario?
Can you describe in more detail what you are trying to implement? (from end-user perspective)Yeah, right now, an end user can go to https://www.claritytext.com/translate right now without hitting a login wall.
When that translation form is submitted, Convex will essentially insert documents into a table with a userId column
ClarityText
Foreign language learning and content analysis.
Your suggestion about an anonymous user account is interesting.
Maybe in Clerk I can create a single "guest" user account that holds all this anonymous activity
But anyway, I think I'm going to reach this zero-to-many limitation again and again. It's a very common pattern
I haven't found to be very common, so far I know of one other user running into this.
Looks like Clerk doesn't support Anonymous users yet: https://github.com/orgs/clerk/discussions/3542 😦
GitHub
Does it support anonymous login? · clerk · Discussion #3542
Hello. I'm wondering if there is support for anonymous login, similar to what is offered in Supabase, or if there are plans to support it in the future, or perhaps methods to implement it mysel...
Hmm okay let me walk back that statement and think through how common it is. I suppose I would need a situation where a one-to-many is neccesary and the reference does not (yet) exist
I suppose I didn't feel the pain of this in other situations in my career because I could always readily insert a null value
I'm a bit confused, we've been building something with complexities that rival most platforms. We have users that can have many orgs, with many members within that org. Within the org the member has roles, those roles have many permissions and the member has many roles or zero roles. Each member can control what any other member can see. Plus we plan to allow a lot of outside visibility. I have yet to run into a inescapable many to zero situation. I guess since our users are many to many with members now, i could get away with linking to zero users. But in principle our members own control of records, through object that have many owner types. So i can create those objects without members needing to exist. But we designed it like this for many other reasons.
its built so many members or other entities can own a object at a near infinite level of granularity, it just happens to fix this problem. but i deal with all the issues around having unions of ids and a table type identifier. which i could in theory go without? 😅