Being able to initialise a fully populated ROM::Struct

[mereghost]

ROM::Structs have albeit having a schema (defined by query returns if I got it right) don't allow you to use them for initialisation.

# table authors (name:string, id:integer, created_at:datetime, updated_at:datetime)
class Author < ROM::Struct ; end
Author.new(name: 'Bob') # => <Author {}>

This makes it hard to test pieces of your code without touching the DB at all and therefore slowing the testing down. Is it possible to offer some similar functionality (maybe as part of rom-sql :/ )?

Author.new(name: 'Bob') # => <Author {name: 'Bob'}>

[flash-gordon]

@mereghost rom-factory does that, have you seen it?

[solnic]

@flash-gordon the idea is to be able to do it via class constant. Thoughts?

[flash-gordon]

@solnic type metadata will be missing, it can be a problem in some cases. For instance, you won't be able to notice column renaming. Otherwise, it's possible with method_missing and overriding of #to_s

[flash-gordon]

though I think I would go with a different name, Struct.build or something

[solnic]

@solnic type metadata will be missing, it can be a problem in some cases. For instance, you won't be able to notice column renaming

@flash-gordon not sure if I follow - we'd use actual struct classes generated by the struct compiler, they have meta-data. I'd imagine that if you do User.build(name: "Jane") then the factory would search for the canonical representation of that struct class (basically going to users relation and grabbing its users.mapper.model) and using that. So, essentially a little bit of magic behind the scenes.

I like Struct.build btw.

[flash-gordon]

@solnic linking User to a container would require some sort of a global state, this concerns me somewhat :)

[solnic]

@flash-gordon but it's already "linked" through struct_namespace and class name inference, isn't it?

[flash-gordon]

@solnic no, it's the other way around, container references classes. Currently, User is an empty class that can be easily shared by multiple containers with (possibly) different schemas

[solnic]

@flash-gordon you're right. I guess we should stick to factories after all. @mereghost would you be OK with that? Please remember that User is only a placeholder class, the actual struct classes used at runtime are subclasses of that User class and they are created on-the-fly, based on relation schemas.

[mereghost]

I took a look at rom-factory but I'm under the impression that it ends up writing to the DB, right?

[solnic]

@mereghost no, you can ask for an in-memory struct via ie Factory#structs[:user]

[mereghost]

I'm looking at it using my hanami/model's glasses:

If someone is using their own structs (struct_namespace) there (even with a couple of methods inside), they will probably be caught off-guard by not being able to initialize them.

What model used to do was to look at the relations (and associations) and craft a mapper for that relation and a schema for the entity, registered as :entity, that we would use by default. For that we did a couple of workarounds that looked like 9and were) an ugly hack but that created a nice developer experience.

That being said: this is OK as it seems to be the preferred rom way to do it and if we want the best for both projects we both need to let go of previous assumptions. =)

But I'd love to see this piece of dev UX to improve, either as plugin or as part of rom itself. =)

[solnic]

@mereghost every rom-rb user must learn how auto-structs work though, so pretending that an abstract entity class is something that it's not would be confusing. We can make it nice and easy though, but unfortunately we shouldn't rely on abstract class constructors for that because that would give you the wrong impression and cause confusion. I'm totally for adding support for this to rom core btw. I cringe every time I do relation.mapper.model.new 😆