Why newExpr requires new keyword?

[taichi-ishitani]

I have a question about newExpr.

I'm using YAML as the input format of my app like this and tring to rewrite it by using Pkl.
https://github.com/rggen/rggen-sample/blob/master/uart_csr.yml

This is the result.

import "rggen.pkl"

register_block = (rggen.register_block) {
  name = "uart_csr"
  byte_size = 32

  (rggen.register) {
    name            = "rbr"
    offset_address  = 0x00
    type            = new { "indirect"; new { "lcr.dlab"; 0 } }
    comment         = "Receiver Buffer Register"

    (rggen.bit_field) {
      bit_assignment  = new { lsb = 0; width = 8 }
      type            = "rotrg"
    }
  }
}

This trial led me to think that {} blocks without new keyword looks better.
Is there any reason why the newExpr requires the new keyword?

[odenix]

Most of the new keywords in your code shouldn't be required. Instead of foo = new { ... }, you can write foo { ... }.

[taichi-ishitani]

Yes, I know but I'd like to use the same assignment notation like below.

foo = 1
bar = { baz = 1 }

In addition, an newExpr within a {} block requires the new keyword. I'd like to ommit this new keyword too.

foo { 0 new { bar = 1 } } // Need new keyword
baz { 2 { qux = 3 } } // I'd like to write like this

[holzensp]

You can indeed do this, but as bioball said, it's not preferred. That is to say; it may well still go away. The problem is that it muddies the namespaces of types and values (which is bad for autocomplete). The idea that parentheses are an operator that takes a class and produces an empty instance to amend is confusion waiting to happen, if you ask me.

[bioball]

Another issue with this syntax is that it's not possible to instantiate more complex types, like new Listing<Person>.

[taichi-ishitani]

I think it would be nice if Pkl has object literal like other languages.
#419 (reply in thread)
What do you think about this?

[odenix]

Another issue with this syntax is that it's not possible to instantiate more complex types

This may be a concern for Pkl, but a typed config/data format only needs to know the type of each individual value. However, new looks strange in a data format.

I think it would be nice if Pkl has object literal like other languages.

Pkl’s object literal syntax is new Type {…}. Only in some cases this can be abbreviated to new {…}.

If I could, I'd be tempted to phase out new Type {...} in favor of (Type) {...} and to only allow simple types in object literals.
Not sure what to do about new {...} though.

[holzensp]

The (Type) { ... } syntax is suboptimal in that it conflates types and values (bad for autocomplete) and it makes the meaning of parentheses significantly more loaded than what most people would assume ((<expr>) { ... } now means "evaluate the object from <expr> and amend it", whereas the (Type) { ... } instantiation means "evaluate the object from <expr> and, if it's a type, instantiate it, otherwise use it as-is, and then amend it").

[bioball]

In Pkl, object bodies always amend something. new {} amends the inferred default value. For example, consider listing defaults:

class Person {
  name: String
  birthday: String
}

people = new Listing<Person> {
  default { name = "Ellis" }
  new { birthday = "2020/01/01" }
}

Here, the new is derived from the default value defined right above it (you end up with new Person { name = "Ellis"; birthday = "2020/01/01" })

Simply using {} for notation would perhaps make this behavior a little too hidden. This feels too surprising to me:

people = new Listing<Person> {
  default { name = "Ellis" };
  { birthday = "2020/01/01" }
}

There's also a grammar ambiguity here; this is parsed as a chained amends expression (foo) {} {} {} {}. If {} can also mean "standalone object", then we have now introduced an ambiguity.

myObj {
  (foo) {}
  {}
  {}
  {}
}

[taichi-ishitani]

I understood it's difficult to omit new keyword simply and new notation is needed for the newExpr without new keyword.
How about using a quotation mark instead of new keyword like this?

foo = '{ bar = 0 }

BTW, how can I highlight Pkl code in Markdown?

[bioball]

It's certainly possible to change the grammar. This is subjective, but I think this is more legible:

foo = new { bar = 0 }

This seems less intuitive to me:

foo = '{ bar = 0 }

I'm using the Groovy syntax highlighter right now for Pkl code. That seems to be the best option for now, until github-linguist/linguist#6730 gets merged

```groovy
foo = new {}
```

[taichi-ishitani]

Most langurages have array literal/hash literal. This is Ruby example.

a = [0, 1, 2]
h = { foo: 0, bar: 1, baz: 2 }

I would like a similar syntax for Pkl.

I'm using the Groovy syntax highlighter right now for Pkl code.

Thanks! I will use this technique.