[kdl v2] .1 is currently an ident string, should it be a number?

[tabatkins]

Currently, the decimal grammar is sign? integer ('.' integer)? exponent?. In other words, you must have one integer digit, and all the other parts are optional. This means that .1 is not a decimal number; instead, it's a valid ident.

I don't think this is very good. I know JSON doesn't allow a number to start with a decimal point, but writing .1 is a syntax error in JSON and you'll catch it. KDL's flexibility works against it here, making it accidentally masquerade as a valid value of a different type. Meanwhile, YAML allows .1, as do JS and CSS (and every other programming language I've ever used, I think?)

I suggest we change the grammar of decimal to:

decimal := sign? ( integer ('.' integer)? | '.' integer ) exponent?

and then modify the "bare-identifier" grammar to be able to look three characters forward, so it can successfully exclude -.1x from being an ident.

bare-identifier := ((identifier-char - digit - sign - '.') identifier-char* | sign ((identifier-char - digit - '.') identifier-char*)? | sign '.' ((identifier-char - digit) identifier-char*)? | '.' ((identifier-char - digit) identifier-char*)?) - keyword

[larsgw]

We can also disallow .0 and -.0 (and the like) entirely, from both decimals and identifiers/strings.

[tabatkins]

Yes, that would be my second preference. I'd much prefer they be numbers, since that's widely accepted as a number in many contexts, but if we think that shouldn't be done I'd at least like to make it a syntax error, like JSON.

[zkat]

I prefer -.0 and .0 both be illegal tbh.

[tabatkins]

That's acceptable. Would you like me to integrate that into the v2 branch?

[zkat]

Yes, please.

[tabatkins]

Done!