Expected End of File

[ejfasting]

Hello,

I've been testing things with the playground today, just adding some basic variableDeclarations and if/while/for statements etc.
Can anyone spot why the editor says it "Expects end of file" here ? Its a little annoying since I cant seem to figure out what/why it's there, and maybe i'm just doing something wrong (?)

I also have some expressions defined, to be able to handle setting/calculations on my variables, but for now I'm just 'hardcoding it' to be whatever i i want it to be (STRING or INT).
The lox-example shows how you can use Expressions to handle setting/calculcating of the variables more dynamically, but I'm unsure if this is the correct approach in my case..

In my case i have variables of a specific type (String, Integer, Number, etc), so i need to have specific expressions for each type of variable (for instance multiplication dont make sense for a String type).
My question is how to best solve this. Would having multiple 'base expressions' (StringExpression, IntExpression.. based on this), and would it then be possible to have for example 'Assigment' infer multiple BaseExpressions?

Last question is if its possible to include from other files? Just to split the .langium file into multiple subfiles which is a little easier to navigate through.

Any other example-projects are also much appreciated, preferably a language that has variables, methods and some classes (like mine does). :)

//Eivind

[msujew]

Hey @ejfasting,

Can anyone spot why the editor says it "Expects end of file"? Its a little annoying since I cant seem to figure out what/why it's there, and maybe i'm just doing something wrong (?)

It might be a little unintuitive, but it's the expected behavior with your grammar. An input such as:

Integer blabla = 123;
print("x");

Works as expected. A short explanation on the parser error: Since there's no parser rule that is reachable from the Model rule can parse an ID token on its own, the parser simply assumes that the file is to end, since ending the parsing at such a position is "a valid move" by the parser. Note that parser error recovery will always be a heuristic at best and is heavily influenced by the actual design of the grammar. I.e. there might be a different "shape" of your grammar that parses the same input but returns better parser errors/deals better with errors in the input.

A general statement rule that allows to parse an ID (and then maybe follow up with a () function call invocation) would fix this issue by itself, as the parser would recognize that this is the "intended" path to take.

The lox-example shows how you can use Expressions to handle setting/calculcating of the variables more dynamically, but I'm unsure if this is the correct approach in my case..The lox-example shows how you can use Expressions to handle setting/calculcating of the variables more dynamically, but I'm unsure if this is the correct approach in my case..

It's generally the best solution for almost all (imperative) programming languages, yes. Yours especially, since you mentioned that your language also has classes. You should also take a look at the class member scoping guide that basically explains what the Lox code is doing under the hood.

In my case i have variables of a specific type (String, Integer, Number, etc), so i need to have specific expressions for each type of variable (for instance multiplication dont make sense for a String type). My question is how to best solve this. Would having multiple 'base expressions' (StringExpression, IntExpression.. based on this), and would it then be possible to have for example 'Assigment' infer multiple BaseExpressions?My question is how to best solve this. Would having multiple 'base expressions' (StringExpression, IntExpression.. based on this), and would it then be possible to have for example 'Assigment' infer multiple BaseExpressions?

I wouldn't solve this in the grammar, as that's either (1) a ton of work or (2) impossible given the context-awareness of most programming languages. I.e. you cannot embed a type system in a programming language grammar. Lox solves this by performing validation on the expressions and checking whether a used operator is valid under the given operands.

Last question is if its possible to include from other files? Just to split the .langium file into multiple subfiles which is a little easier to navigate through.

Yes, see our requirements example. Note that I would really only recommend splitting up the grammar file in case you have multiple languages.

[ejfasting]

Since String is a direct reference to your String class, it will automatically display the documentation of the class on hover

Unfortunately this is what doesnt work, the hover from the class doesnt show on the variable: Class , Variable

Since String is a variableName, its not a valid name for a Class. So i have to add the keyword as a valid identifier.. I guess this breaks the connection between the variable String and class String, since its not a TypeReference?

Edit:
The same can be seen in the lox example. Define a class named 'var', add the keyword 'var' as a valid name for a class, and add some comments to the class. Every 'var' in the document doesnt get the hover-comment. So something else needs to be done as well

[msujew]

Since String is a variableName, its not a valid name for a Class. So i have to add the keyword as a valid identifier.. I guess this breaks the connection between the variable String and class String, since its not a TypeReference?

Yes, having a real cross reference to the class String here would be preferable. But I'm not sure I can follow the "Since String is a variableName, its not a valid name for a Class" part. Why is String not a valid name for a class?

[ejfasting]

Because it's a keyword, so i have to define it like this on the class:
Class: 'class' name=(ID | 'String')
Then it becomes valid: https://langium.org/docs/recipes/keywords-as-identifiers/

Got it working, somewhat, in the playground
This is pretty much a copy-paste of the lox-example, with some minor edits to make the 'connection' work:

VariableDeclaration returns NamedElement:
    {infer VariableDeclaration} (type=[Class:StringType])? (type=[Class:IntType])? name=ID (assignment?='=' value=Expression)?
;

StringType returns string: ID | 'String';
IntType returns string: ID | 'Integer';

and

Class: 'class' name=(ID | StringType | IntType) ('<' superClass=[Class:ID])? '{'
    members+=ClassMember*
'}';

It blows up the whole valiation and assignments in my local project, though, but that is to be expected i guess.
The VariableDeclaration no longer has a specific type, it has a Reference instead:

Argument of type 'Reference<Class>' is not assignable to parameter of type 'AstNode'.
  Property '$type' is missing in type 'Reference<Class>' but required in type 'AstNode'.

Might be there is an easier way to achieve it, but it looks like i have to add some logic to handle the case where you assign something to a Class which has the same name as my BaseType (String, Integer, Bool etc).
Maybe there also exists an easy way to resolve the reference(?)

[msujew]

I'm a bit confused. Why are String and Integer keywords in the first place? It should be enough for them to only be defined as classes.

Also, a construct like (type=[Class:StringType])? (type=[Class:IntType])? would allow for <type-reference> <type-reference> <variable-name>. Multiple assignments to the same property is a bad idea in general.

[ejfasting]

.. you are of course correct. Doing it that way was much easier.

Again, thank you so much! :)