Character.isLetter() es mucho más rápido que string.matches(), porque string.matches() compila un nuevo patrón cada vez. Incluso guardando en caché el patrón, creo que Letter() todavía lo superaría.
EDIT: acaba de ejecutar a través de este nuevo y pensé que iba a tratar de llegar a algunos números reales. Aquí está mi intento de un punto de referencia, comprobando los tres métodos (matches()
con y sin almacenar en caché el Pattern
y Character.isLetter()
). También me aseguré de que se verificaran los caracteres válidos y no válidos, para no sesgar las cosas.
import java.util.regex.*;
class TestLetter {
private static final Pattern ONE_CHAR_PATTERN = Pattern.compile("\\p{L}");
private static final int NUM_TESTS = 10000000;
public static void main(String[] args) {
long start = System.nanoTime();
int counter = 0;
for (int i = 0; i < NUM_TESTS; i++) {
if (testMatches(Character.toString((char) (i % 128))))
counter++;
}
System.out.println(NUM_TESTS + " tests of Pattern.matches() took " +
(System.nanoTime()-start) + " ns.");
System.out.println("There were " + counter + "/" + NUM_TESTS +
" valid characters");
/*********************************/
start = System.nanoTime();
counter = 0;
for (int i = 0; i < NUM_TESTS; i++) {
if (testCharacter(Character.toString((char) (i % 128))))
counter++;
}
System.out.println(NUM_TESTS + " tests of isLetter() took " +
(System.nanoTime()-start) + " ns.");
System.out.println("There were " + counter + "/" + NUM_TESTS +
" valid characters");
/*********************************/
start = System.nanoTime();
counter = 0;
for (int i = 0; i < NUM_TESTS; i++) {
if (testMatchesNoCache(Character.toString((char) (i % 128))))
counter++;
}
System.out.println(NUM_TESTS + " tests of String.matches() took " +
(System.nanoTime()-start) + " ns.");
System.out.println("There were " + counter + "/" + NUM_TESTS +
" valid characters");
}
private static boolean testMatches(final String c) {
return ONE_CHAR_PATTERN.matcher(c).matches();
}
private static boolean testMatchesNoCache(final String c) {
return c.matches("\\p{L}");
}
private static boolean testCharacter(final String c) {
return Character.isLetter(c.charAt(0));
}
}
Y mi salida:
10000000 tests of Pattern.matches() took 4325146672 ns.
There were 4062500/10000000 valid characters
10000000 tests of isLetter() took 546031201 ns.
There were 4062500/10000000 valid characters
10000000 tests of String.matches() took 11900205444 ns.
There were 4062500/10000000 valid characters
Así que eso es casi 8 veces mejor, incluso con un caché Pattern
. (Y sin caché es casi 3 veces peor que en caché.)
Debe usar 'c.codePointAt (0)' en lugar de 'c.charAt (0)' en 'testCharacter()'; de lo contrario, fallará para los caracteres fuera del BMP. –