Repository navigation
Pass existing [EntityDef] into mkPersist #1241
Description
mkPersist :: MkPersistSettings -> [EntityDef] -> Q [Dec]A problem with this is that the generated declarations don't know about any of the EntityDefs that are not provided as part of this invocation. So if you have two blocks, one of which defines User and another defining an Organization with a user UserId field, the foreign key won't get properly invoked, since the liftAndFIxKeys machinery won't find the User in the existing map of tables.
This would fix #1073 as well. That issue has some good exploration on why this particular problem impacts foreign key generation. persistent figures that a BlahBlahId type is a reference to a BlahBlah model, and it tries to find the BlahBlah model in a Map HaskellName EntityDef that it constructs... from the [EntityDef] that is passed in.
A work-around is to... well, pass them in!
module Foo where
share [mkPersist sqlSettings, mkEntityDefList "fooDefs"] [persistLowerCase|
Foo
name Text
|]
module Bar where
import Foo
share
[ mkPersist sqlSettings . mappend fooDefs
, mkEntityDefList "barAndFooDefs"
] [persistLowerCase|
Bar
foo FooId
name Text
|]This requires that you do that . mappend dependentDefs for each one. But it should fix the issue.
It's a bit annoying to tease these relationships together manually. And it's very easy to forget, especially since, well, everyone pretty much has. Can we do better?
Activity
parsonsmatt commentedon Apr 21, 2021
I think we can!
reifyInstances, called likereifyInstances [''PersistEntity] [VarT (mkName "a")], should find all instances ofPersistEntitythat are in scope. We get back an[InstanceDef]. This is [guaranteed to be anInstanceDwith an empty[Dec](https://www.stackage.org/haddock/lts-17.9/template-haskell-2.16.0.0/Language-Haskell-TH.html#t:InstanceDec).InstanceDcarries theTypefor the instance - in this case, it'd look something likeAppT (ConT ''PersistEntity) theType.Given
theType :: Type, we can write:which produces an
EntityDefin the spliced code. ThisEntityDefis going to be the one defined on thePersistEntityclass, and so is the canonical one, not the one produced by the QuasiQuoter (which may be left unspecified).Now, we don't yet have those
EntityDefs in scope. It's still aQ Exprepresenting the list of all definitions. But we have a[EntityDef]that we're working with. How can we bridge this gap?parsonsmatt commentedon Apr 21, 2021
Oh, duh, same way we get around
persistLowerCase :: QuasiQuoterreturning aString -> Q Exp- we splice it in.It's still annoying to need to mappend them manually.
It's also obnoxious to have to do
entityDefsInScopefor every module - that's going to be a linear instance lookup hit on each module that compiles, recollecting the same instances over and over again. Will definitely slow things down. Is there a way to keep track of some kind of global state? A compile-timeIORef [Q Exp]that carries all theentityDefs would solve this nicely...parsonsmatt commentedon Apr 21, 2021
Hmm,
reifyInstancesdoesn't have to grab all the instances. We can also callreifyInstanceswhen we're trying to look up a foreign key, specifically for a table that matches.Eg:
When working with
Bar, first we check theEntityMap. If it's not there, we callreifyInstances ''PersistEntity (ConT (mkName "Foo")). If this returns anything, then we can splice in the resulting stuff. I'm not sure if this will operate at the phase level we care about, though.That'll end up doing a
reifyInstancessearch for every key use, which will (I think) end up being slower than just grabbing all the instances up front and stuffing them into the map.parsonsmatt commentedon Apr 22, 2021
wait, maybe we could do the splicing in
persistLowerCasethe quoter exp is basically
pure . parseReferences ps, so we coudl make itbut then we need some check or lookup in
mkPersistto differentiate - shouldn't be hard to verify ifAis an instance ofPersistEntityalready or not, but that's just another lookup.(really the quoter should not return
[EntityDef]it should return something else...)then
mkPersistwould have no trouble differentiating the two.