Ballerina-lang: Some issues with floating point literal parsing

[SupunS]

Case I:

function foo() {
    float f = 5.;
}

Above is parsed as a valid code.

Case II:

function foo() {
    float f = 5.4e;
}

Above gives multiple error messages for a single error. Below are the error messages and the resulting tree.

// Error messages
xxx.bal:0:0:missing digit
xxx.bal:0:0:invalid token '5.4e'
xxx.bal:0:0:missing  MISSING[]

// Resulting tree
function foo() {
    float f =  MISSING[]5.4e;
}

Case III

function foo() {
    float f = 5.e1;
}

Above also get parsed as a valid code.

[lochana-chathura]

@SupunS I allowed case I and III intentionally. Test cases for them also added.

floating-point-literal := DecimalFloatingPointNumber
DecimalFloatingPointNumber := DottedDecimalNumber [Exponent] [FloatingPointTypeSuffix]
DottedDecimalNumber := DecimalNumber . Digit*

IIUC , DecimalNumber . Digit* means it is allowed have no digits after dot token right?

[lochana-chathura]

I will have a look on case II. :)

[SupunS]

@SupunS I allowed case I and III intentionally. Test cases for them also added.

floating-point-literal := DecimalFloatingPointNumber
DecimalFloatingPointNumber := DottedDecimalNumber [Exponent] [FloatingPointTypeSuffix]
DottedDecimalNumber := DecimalNumber . Digit*

IIUC , DecimalNumber . Digit* means it is allowed have no digits after dot token right?

I see. Yes you are correct. I didn't notice that part.

[SupunS]

I think for case-III, we can log an error in the lexer, but still return a valid floating-point-literal token, so that the rest of the parser will work normally.. WDYT?

[lochana-chathura]

I think for case-III, we can log an error in the lexer, but still return a valid floating-point-literal token, so that the rest of the parser will work normally.. WDYT?

Case III should be fine. I think you are referring to case II here. Yeah we can do that change. +1

[SupunS]

Yes, I meant Case-II

[lochana-chathura]

Similar to case II, we have another two scenarios where numerical literals are processed as invalid tokens.

1. When an int or a float starts with a zero. e.g. float f = 02.3
2. When hex-digit starts with a dot token and there's no at least one hex-digit after that. e.g. float f = 0x.p12

Are we going to allow them to parse as well while logging an error at lexer level?

[SupunS]

hmm.. yeah we should do that.

Anyway, I'm lowering the priority of the issue since this doesn't break any code, and only an improvement.