parseInt vs unary plus, when to use which?

Question

The ultimate whatever-to-number conversion table:

EXPRS = [
    'parseInt(x)',
    'parseFloat(x)',
    'Number(x)',
    '+x',
    '~~x',
    'x>>>0',
    'isNaN(x)'

];

VALUES = [
    '"123"',
    '"+123"',
    '"-123"',
    '"123.45"',
    '"-123.45"',
    '"12e5"',
    '"12e-5"',
    
    '"0123"',
    '"0000123"',
    '"0b111"',
    '"0o10"',
    '"0xBABE"',
    
    '"4294967295"',
    '"123456789012345678"',
    '"12e999"',

    '""',
    '"123foo"',
    '"123.45foo"',
    '"  123   "',
    '"foo"',
    '"12e"',
    '"0b567"',
    '"0o999"',
    '"0xFUZZ"',

    '"+0"',
    '"-0"',
    '"Infinity"',
    '"+Infinity"',
    '"-Infinity"',

    'null',
    'undefined',
    'true',
    'false',
    'Infinity',
    'NaN',

    '{}',
    '{valueOf: function(){return 42}}',
    '{toString: function(){return "56"}}',

];

//////

function wrap(tag, s) {
    if (s && s.join)
        s = s.join('');
    return '<' + tag + '>' + String(s) + '</' + tag + '>';
}

function table(head, rows) {
    return wrap('table', [
        wrap('thead', tr(head)),
        wrap('tbody', rows.map(tr))
    ]);
}

function tr(row) {
    return wrap('tr', row.map(function (s) {
        return wrap('td', s)
    }));
}

function val(n) {
    return n === true || Number.isNaN(n) ? wrap('b', n) : String(n);
}

var rows = VALUES.map(function (v) {
    var x = eval('(' + v + ')');
    return [v].concat(EXPRS.map(function (e) {
        return val(eval(e))
    }));
});

document.body.innerHTML = table(["x"].concat(EXPRS), rows);

table { border-collapse: collapse }
tr:nth-child(odd) { background: #fafafa }
td { border: 1px solid #e0e0e0; padding: 5px; font: 12px monospace }
td:not(:first-child) { text-align: right }
thead td { background: #3663AE; color: white }
b { color: red }

Well, here are a few differences I know of:

• An empty string "" evaluates to a 0, while parseInt evaluates it to NaN. IMO, a blank string should be a NaN.

    +'' === 0;              //true
    isNaN(parseInt('',10)); //true
  

• The unary + acts more like parseFloat since it also accepts decimals.

  parseInt on the other hand stops parsing when it sees a non-numerical character, like the period that is intended to be a decimal point ..

    +'2.3' === 2.3;           //true
    parseInt('2.3',10) === 2; //true
  

• parseInt and parseFloat parses and builds the string left to right. If they see an invalid character, it returns what has been parsed (if any) as a number, and NaN if none was parsed as a number.

  The unary + on the other hand will return NaN if the entire string is non-convertible to a number.

    parseInt('2a',10) === 2; //true
    parseFloat('2a') === 2;  //true
    isNaN(+'2a');            //true
  

• As seen in the comment of @Alex K., parseInt and parseFloat will parse by character. This means hex and exponent notations will fail since the x and e are treated as non-numerical components (at least on base10).

  The unary + will convert them properly though.

    parseInt('2e3',10) === 2;  //true. This is supposed to be 2000
    +'2e3' === 2000;           //true. This one's correct.
  
    parseInt("0xf", 10) === 0; //true. This is supposed to be 15
    +'0xf' === 15;             //true. This one's correct.
  

The table in thg435's answer I believe is comprehensive, however we can summarize with the following patterns:

• Unary plus does not treat all falsy values the same, but they all come out falsy.
• Unary plus sends true to 1, but "true" to NaN.
• On the other hand, parseInt is more liberal for strings that are not pure digits. parseInt('123abc') === 123, whereas + reports NaN.
• Number will accept valid decimal numbers, whereas parseInt merely drops everything past the decimal. Thus parseInt mimics C behavior, but is perhaps not ideal for evaluating user input.
• Both trim whitespace in strings.
• parseInt, being a badly designed parser, accepts octal and hexadecimal input. Unary plus only takes hexademical.

Falsy values convert to Number following what would make sense in C: null and false are both zero. "" going to 0 doesn't quite follow this convention but makes enough sense to me.

Therefore I think if you are validating user input, unary plus has correct behavior for everything except it accepts decimals (but in my real life cases I'm more interested in catching email input instead of userId, value omitted entirely, etc.), whereas parseInt is too liberal.

I recommend use Math.floor (or ~~ if you know numbers are positive) instead of parseString. +(expression) is out of scope, because +(expression) is more like parseFloat. Look this little benchmark:

// 1000000 iterations each one
node test_speed
Testing ~~, time: 5 ms
Testing parseInt with number, time: 25 ms
Testing parseInt with string, time: 386 ms
Testing Math.floor, time: 18 ms

The source code of benchmark:

/* el propósito de este script es evaluar
que expresiones se ejecutan más rápido para así 
decidir cuál usar */

main()
async function main(){
    let time, x 
    let number = 23456.23457
    
    let test1 = ()=>{
        x = 0
        time = Date.now() 
        for(let i=0;i<1000000;i++){
            let op = Math.floor(number / 3600)
            x = op
        }
        console.info("Testing Math.floor, time:", Date.now() - time, "ms")
    }

    let test2 = ()=>{
        x = 0
        time = Date.now() 
        for(let i=0;i<1000000;i++){
            let op = parseInt(number / 3600)
            x = op
        }
        console.info("Testing parseInt with number, time:", Date.now() - time, "ms")
    }

    let test3 = ()=>{
        x = 0
        time = Date.now() 
        for(let i=0;i<1000000;i++){
            let op = parseInt((number / 3600).toString())
            x = op
        }
        console.info("Testing parseInt with string, time:", Date.now() - time, "ms")
    }

    let test4 = ()=>{
        x = 0
        time = Date.now() 
        for(let i=0;i<1000000;i++){
            let op = ~~(number / 3600)
            x = op
        }
        console.info("Testing ~~, time:", Date.now() - time, "ms")
    }
    
    test4()
    test2()
    test3()
    test1()
    
}