I think it is also Unicode aware in 5.8+ so that introduces all sorts of variation - if Unicode is supported, \w usually refers to all alphanumberics - in fact you might see \w start to die out and things like unicode properties appear

$word = m/([\p{Letter}\p{Number}]+)/;
[download]

I think that is correct...

I know that in Java's regex package, \w == \A-Za-z0-9_\ only

I got most of this info from Friedl's Mastering Regular Expressions - if you dont own it - buy it and read it from chapters 1-7 - it is truely excellent, a standout book.