Utilize protocol to get defaults (set-builder-fn, table-fn, label-fn and etc)

[kirillsalykin]

Is your feature request related to a problem? Please describe.
It is pretty annoying to pass custom set-builder/table-fn/label-fn to every query or statement.

Describe the solution you'd like
It would help a lot if there would be a way to specify it only once.
But having the possibility to pass override via opts still important.

Describe the solution you'd like
Maybe having a protocol to get defaults would help with this?
And if one wants to override defaults - it can be achieved with extend-protocol.

Similar to ReadableColumn approach.

[seancorfield]

Given that protocols are based on types I don't see how this would work -- could you elaborate on what you have in mind?

[kirillsalykin]

Sorry for not being clear.

I propose to create a new protocol:

(defprotocol NamingConvention
  (builder-fn [this])
  ...)

Then extend-protocol to all types which extended by Executable:

next-jdbc/src/next/jdbc/result_set.clj

Line 407 in 6bc7164

(extend-protocol p/Executable

• java.sql.Connection
• javax.sql.DataSource
• java.sql.PreparedStatement

(extend-protocol p/NamingConvention
  java.sql.Connection
  (builder-fn [_] 
     as-maps)

   javax.sql.DataSource
   (builder-fn [_] 
      as-maps)

   java.sql.PreparedStatement
  (builder-fn [_] 
     as-maps))

And get default builder-fn using protocol

next-jdbc/src/next/jdbc/result_set.clj

Line 424 in 6bc7164

(let [builder-fn (get opts :builder-fn as-maps)

builder-fn (get opts :builder-fn (p/builder-fn this)

If one wants to override it - he should extend-protocol to the java(x).sql types.

A bit annoying, but doable.

What do you think?

[kirillsalykin]

Hm, not sure it works...

[seancorfield]

I think you could base it on ResultSet -- I think everywhere that "gets" the :builder-fn option has access to a rs object...?

[kirillsalykin]

I'll play more with to ensure it actually works.
Will get back to you.

Thanks!

[seancorfield]

So you're thinking that the library would provide a default implementation for Object and then users could override it for ResultSet with their defaults instead?

I still feel this is really hacky...

[kirillsalykin]

I agree, feels hacky, but cant think of any other approach to override default behavior w/o global variable.

[seancorfield]

I canvassed some of Clojure contributors and everyone who had an opinion felt this was a bad approach. Protocols should really only be extended either by the code that owns the protocol (next.jdbc in this case) or owns the type -- and the connection to ResultSet was felt to be tenuous at best.

So I'm closing this out. My original suggestion of creating a wrapper API still stands -- but I think that should be in user code.

[kirillsalykin]

I agree.

Thanks for investigating it.