Internal type split

[msujew]

This change splits the type system into a "plain" (string-based) type system and a concrete (reference-based) type system. Splitting the types allows for easier handling of purely generated information (i.e. container types, union/interface hierarchies and type names). It also has an impact on how we deal with property types and union types in general, which are now way more flexible.

Aims to resolve multiple issues:

• Closes Add the ability to use parentheses when declaring TypeAlternatives #511 (The grammar has been slightly modified)
• Closes Merge union type definitions #476
• Closes Improve type inferrence of data type rules #752
• Closes Type with a property should become an interface #534
• Closes Incorrect inference of property optionality for a rule with an action #868

[spoenemann]

The result looks mostly really good! But it doesn't properly solve #534: with the rule

entry X: id=ID ({infer Y} 'a' | {infer Z} 'b');

I get the types

export type X = Y | Z;

export interface Y extends AstNode {
    readonly $type: 'Y';
}

export interface Z extends AstNode {
    readonly $type: 'Z';
}

so the id property is forgotten. Could you add a test for this?

[msujew]

@spoenemann Thanks for the hint 👍 It should as expected now.

[spoenemann]

How does this relate to #554?

[pluralia]

• Closes also Incorrect inference of property optionality for a rule with an action #868.
• After generating the code, the empty line at the end of these files is deleted:
  • examples/requirements/syntaxes/requirements.tmLanguage.json
  • examples/requirements/syntaxes/tests.tmLanguage.json

[msujew]

How does this relate to #554?

It's closely related, but not 100% there yet. My code can be used as a basis to build the graph mentioned in #554.

[msujew]

After generating the code, the empty line at the end of these files is deleted

Unrelated to the changes in my PR. It might be due to some recent changes in the generator API.

[pluralia]

Unrelated to the changes in my PR. It might be due to some recent changes in the generator API.

True. Could you fix it in this PR?

[msujew]

True. Could you fix it in this PR?

I would rather see this fixed in a separate PR.

[spoenemann]

After generating the code, the empty line at the end of these files is deleted

Unrelated to the changes in my PR. It might be due to some recent changes in the generator API.

It's my fault: I had to manually fix the generated files (#888 (comment)), then @msujew fixed the generator, but we didn't regenerate the files. The file ending is different because the text editor automatically changes it on save.

[montymxb]

General suggestions & comments. In particular I'm curious if we can still retain some (or all) of the previous inferred type tests structure. and then there's the point about the subtle change from undefined to false for values when parsing fails.

[montymxb]

Tests look good after a double check. There was one that was slightly changed (inferred types using chained actions), where G, H, and I are changed around. But this looks okay based on testing locally. Also checked this against some other Langium projects as well, and nothing appears out of the ordinary from what I can tell.

[msujew]

@montymxb Thanks for the thorough review!

@pluralia Do you have time this week to look into this? pinging @sailingKieler as well. Otherwise I would merge this soonish.