Insert Select With Conflict for postgres #155
Conversation
src/Database/Esqueleto/PostgreSQL.hs
Outdated
| entCurrent = EEntity $ I (tableName proxy) | ||
| -- there must be a better way to get the constrain name from a unique, make this not a list search | ||
| uniqueDef = head . filter ((==) (persistUniqueToFieldNames unique) . uniqueFields) . entityUniques . entityDef $ proxy |
There was a problem hiding this comment.
head . filter p === find p
It would be nice to have a function toUniqueDef :: PersistEntity record => Unique record -> UniqueDef, but this is basically how I'd implement it 🤷♂️
There was a problem hiding this comment.
smart guy, I always forget find exists, but it keeps being a list search, but I think it should not be too difficult to add toUniqueDef on persist, though I will have to look at the code to be sure.
Also, the thing with the undefined, I would also like it to be automatic, but the Unique creation function may have any amount of parameters, when defining a unique in persist we can:
...
Foo
num1 Int
num2 Int
num3 Int
UniqueFooNum1 num1
UniqueFooNum23 num2 num3
deriving Eq Show
...
so I can't just change the firm to (a -> Unique val) -> SqlQuery (SqlExpr (Insertion val)) -> (SqlExpr (Entity val) -> SqlExpr (Entity val) -> [SqlExpr (Update val)]) -> SqlWriteT m Int64 because I would not be able to use UniqueFooNum23 on this new function, if there is a way receive a variable parameter function and then fill all parameters with undefined I will add it.
There was a problem hiding this comment.
I linked to KnowResult in another comment. You can implement a type class that will "fill in" the unused values with undefined also. It's a bit of a trick of type class programming so if you'd like help on that please let me know :)
src/Database/Esqueleto/PostgreSQL.hs
Outdated
| insertSelectWithConflict :: (MonadIO m, PersistEntity val) => | ||
| Unique val | ||
| -- ^ Uniqueness constraint, this is used just to get the name of the postgres constraint, the value(s) is(are) never used, so if you have a unique "MyUnique 0", "MyUnique undefined" will work as well. |
There was a problem hiding this comment.
I've used a trick like this before that matches on the final result type of a value - see KnowResult here which allows you to only provide the data constructor. With another type class, you can even pass in the required undefineds automatically so that the user is not concerned with this.
There was a problem hiding this comment.
Thanks! nice recommendation, I have not use type families much, but seeing KnowResult did help me avoid the undefined, though I think I should move KnowResult to another place (I copy KnowResult to Database.Esqueleto.PostgreSQL), any recommendations on where to place it and if it is possible to make the firm for insertSelectWithConflict prettier?
| let test = map (OneUnique "test" . personFavNum) [p1,p2,p3] | ||
| test2 = map (OneUnique "test" . (+4) . (*2) . personFavNum) [p1,p2,p3] | ||
| liftIO $ map entityVal uniques1 `shouldBe` test | ||
| liftIO $ map entityVal uniques2 `shouldBe` test2 |
There was a problem hiding this comment.
Very nice, thanks for the tests!
There was a problem hiding this comment.
Awesome job! I don't mind things being where they are.
@bitemyapp thoughts on this? I'm in favor of inclusion.
src/Database/Esqueleto/PostgreSQL.hs
Outdated
| finalR = id | ||
|
|
||
| instance (FinalResult b) => FinalResult (a -> b) where | ||
| finalR f = finalR (f undefined) |
|
Is this still WIP? If it's ready to merge, please add a Changelog entry and update the version in the .cabal file. |
Co-Authored-By: Matt Parsons <parsonsmatt@gmail.com>
Co-Authored-By: Matt Parsons <parsonsmatt@gmail.com>
Co-Authored-By: Matt Parsons <parsonsmatt@gmail.com>
Co-Authored-By: Matt Parsons <parsonsmatt@gmail.com>
JoseD92
changed the title
|
Once CI is green I'll merge. Thanks @JoseD92 !! |
|
3.1.3 is released. Thanks again! |
Based on #86 I came with insertSelectWithConflict, it is like insertSelect but allows to perform an update if some row conflicts with the new addition on a given constraint. There is no way in postgres (that I could find) of responding to any possible conflict, so a specific one needs to be indicated with a unique.