(java) Scientific notation numbers are only legal for floating points

[0xflotus]

Describe the issue

HighlightJS highlights illegal number literals

Which language seems to have the issue?

java

Are you using highlight or highlightAuto?

highlight

Sample Code to Reproduce

Case a and b should be illegal, case c and d are fine.

https://jsfiddle.net/e1xnqskb/

a = 2e4L;
b = 2e3l;
c = 2e3f;
d = 1.2e4D;

Expected behavior

I would expect that illegal number literals are not highlighted

Additional context

It's a similar issue to #2507

[joshgoebel]

You want to make an attempt at a patch? Fixing this will likely require splitting the huge regex we currently have up into variants to match the different types...

[0xflotus]

I tried the following code locally but it doesn't work.

var JAVA_NUMBER_RE = '\\b' +
    '(' +
    '(' +
    '0[bB]([01]+[01_]+[01]+|[01]+)' + // 0b...
    '|' +
    '0[xX]([a-fA-F0-9]+[a-fA-F0-9_]+[a-fA-F0-9]+|[a-fA-F0-9]+)' + // 0x...
    '|' +
    '(' +
    '([\\d]+[\\d_]+[\\d]+|[\\d]+)(\\.([\\d]+[\\d_]+[\\d]+|[\\d]+))?' +
    '|' +
    '\\.([\\d]+[\\d_]+[\\d]+|[\\d]+)' +
    ')' +
    ')' +
    '[lLfF]?' + '|' + '(' +
    '([\\d]+[\\d_]+[\\d]+|[\\d]+)(\\.([\\d]+[\\d_]+[\\d]+|[\\d]+))?' +
    '|' +
    '\\.([\\d]+[\\d_]+[\\d]+|[\\d]+)' +
    ')' +
    '([eE][-+]?\\d+)?' + '[fF]?' + ')';

Regexes are really hard. :D

Sorry for bad indentation :(

[joshgoebel]

Yeah I wouldn't accept that anyways, we need to switch to variants. Look at how some of the other syntaxes use variant for numbers. :-)

And we'll need some markup tests for as many number combos as you want to come up with. It's possible we already have some.

[joshgoebel]

There are some nice apps that can make testing them easier... on Mac: Patterns, Expressions, etc

[0xflotus]

Yeah ok, I will try again with variants :-)

For sure. I used https://regex101.com/ to test manually the regex with some cases

[joshgoebel]

eah ok, I will try again with variants :-)

It should be easier when you only have to get one type working at a time vs a huge regex for them all, etc...

[0xflotus]

btw shouldn't it be the same like hljs.C_NUMBER_RE ?

[joshgoebel]

That might work for some of the variants... See arcade for an example:

   variants: [
      { begin: '\\b(0[bB][01]+)' },
      { begin: '\\b(0[oO][0-7]+)' },
      { begin: hljs.C_NUMBER_RE }
    ],

[joshgoebel]

We already have tests so you'd just be exepanding those: test/markup/java/numbers.txt

[0xflotus]

That might work for some of the variants... See arcade for an example:

   variants: [
      { begin: '\\b(0[bB][01]+)' },
      { begin: '\\b(0[oO][0-7]+)' },
      { begin: hljs.C_NUMBER_RE }
    ],

I will try it, it's similar to javascript.

[0xflotus]

It's almost hard, because of some other facts:

0779 is a valid octal number
0778 not
octal numbers may have a suffix of [fFdDlL]
binary and hexadecimal numbers are integers and may only have the suffix [lL]
but hexadecimal floating point numbers could have only suffix [dDfF]