Add helpful error message for invalid number literal like '.42'

[makenowjust]

Dot-started floating-point number like '.42' is not supported on Crystal. However compiler does not tell us this, so this fix adds to detect such a literal and to show helpful error message.

For example, old compiler shows an error message:

$ crystal run foo.cr
Syntax error in foo.cr:1: unexpected token: .

foo = .42
      ^

And this PR compiler shows:

$ ./bin/crystal run foo.cr
Using compiled compiler at `.build/crystal'
Syntax error in foo.cr:1: .1 style number literal is not supported, put 0 before dot

foo = .42
      ^

If you have more helpful error message text, please tell me. You know, my English is not good.

[straight-shoota]

I'd suggest a wording like float literals cannot omit a leading zero

[ysbaddaden]

Or just follow the octal literal message (octal constants should be prefixed with 0o):

float constants should be prefixed with 0

[akzhan]

float literals should be started with decimal digit

[straight-shoota]

@ysbaddaden I think the wording should differentiate between constant and literal. So we should probably change the octal message to literal as well. Plus, prefixing float literals with 0 in general would actually be wrong (like 03.141).

What about:

float literals always require an integer part, omitting `0` is not supported

[asterite]

What happens if you write foo.42?

[david50407]

Why Crystal not supporting .123 type floats? Any reason?

[makenowjust]

@david50407 .123 number conflicts ... operator: 1...2 means (1)...(2) or (1)..(.2)?

[makenowjust]

@asterite .42 is detected by the lexer, so an error against foo.42 is also .1 style number literal is not supported.... Direct foo.42 is rare case, but when we have a macro foo.{{name}} and name is given as number by carelessness, then we get same error. I think it is not helpful. Should I let the parser detect .42 inside parse_atomic_without_location?

[asterite]

@makenowjust It's hard.

I don't know if we should give an error. I wouldn't expect this:

MOV EAX, 1

to give an error:

MOV EAX, 1
^
this is not assembler, it's Crystal

(this is not C, it's Crystal)

I mean, we can't give an error for every idiom or piece of syntax that appears in other languages. If you get an error in .42 saying "unexpected '.'" then it's probably clear then this style of numbers if not supported.

[makenowjust]

@asterite Nevertheless I believe such an error message is helpful and useful. For example ruby shows warning message for .42:

$ ruby -e .42
-e:1: no .<digit> floating literal anymore; put 0 before dot
-e:1: syntax error, unexpected '.'

[straight-shoota]

There is a difference between those cases: writing .42 instead of 0.42 is typically a mistake by a newcomer to Crystal coming from a language where you can omit leading zeros. For them a more detailed error message would be helpful and encouraging.

[david50407]

@makenowjust oops I missed that possible

How about just give a nicer tip like:

Syntax error in foo.cr:1: unexpected token: .
Tip: If you are trying to declaring a float constant, put 0 before dot.

[makenowjust]

@david50407 unexpected_token method takes msg optional argument to show additional information. For example:

$ echo 'def foo bar; end' > foo.cr
$ crystal run foo.cr
Syntax error in foo.cr:1: unexpected token: bar (parentheses are mandatory for def arguments)

def foo bar; end
        ^

I'll try to implement by using it.

[sdogruyol]

Even though it's a small thing, I believe this kind of small things makes a language more friendly to newcomers 👍

[straight-shoota]

I'd like to see this merged. It's not a strict necessity, but as @sdogruyol puts it, a small think to make the language more friendly to its users (in this case newcomers specifically).

My preferred error message: float literals require an integer part, a `0` integer part can't be ommited

[beta-ziliani]

@makenowjust can you rebase this? Let us know otherwise. Thanks!

[makenowjust]

rebase is done.