Looks clever to use just one loop for toRoman and bitwise XOR, as a hack get an integer numeral part, it could be replaced by | as well. Also the second argument for repeat method allows to avoid using return statement

It's pretty well step to use regexp match into fromRoman, it may be simplified a bit to rom.match(/CM|CD|XC|XL|IX|IV|\w/gi)

The perfect solution with fromRoman regexp, it's really clever to guess for it, as a good splitter for target roman numeral pairs

It seems that performance leaks are not issue for this purpose as any roman number can't extremelly big in reality. Cool approach, especially for fromRoman function. Unfortunatelly substr is already considered as deprecated, so it may be replaced

m is the part of a string that's matched with regexp provided, the callback replaces this part with incremented value. For instance if there is 'somestring0' it returns ++0(1) and if it's 'somestring499' ++499(500) is returned. Then it replaces regexp in the way as joekotvas explained above

Say what?