Performance-wise parseInt
and such are much worser than other solutions, because at least require exception handling.
I've run jmh tests and have found that iterating over String using charAt
and comparing chars with boundary chars is the fastest way to test if string contains only digits.
JMH testing
Tests compare performance of Character.isDigit
vs Pattern.matcher().matches
vs Long.parseLong
vs checking char values.
These ways can produce different result for non-ascii strings and strings containing +/- signs.
Tests run in Throughput mode (greater is better) with 5 warmup iterations and 5 test iterations.
Results
Note that parseLong
is almost 100 times slower than isDigit
for first test load.
Benchmark Mode Cnt Score Error Units
testIsDigit thrpt 5 9.275 ± 2.348 ops/s
testPattern thrpt 5 2.135 ± 0.697 ops/s
testParseLong thrpt 5 0.166 ± 0.021 ops/s
Benchmark Mode Cnt Score Error Units
testCharBetween thrpt 5 16.773 ± 0.401 ops/s
testCharAtIsDigit thrpt 5 8.917 ± 0.767 ops/s
testCharArrayIsDigit thrpt 5 6.553 ± 0.425 ops/s
testPattern thrpt 5 1.287 ± 0.057 ops/s
testIntStreamCodes thrpt 5 0.966 ± 0.051 ops/s
testParseLong thrpt 5 0.174 ± 0.013 ops/s
testParseInt thrpt 5 0.078 ± 0.001 ops/s
Test suite
@State(Scope.Benchmark)
public class StringIsNumberBenchmark {
private static final long CYCLES = 1_000_000L;
private static final String[] STRINGS = {"12345678901","98765432177","58745896328","35741596328", "123456789a1", "1a345678901", "1234567890 "};
private static final Pattern PATTERN = Pattern.compile("\\d+");
@Benchmark
public void testPattern() {
for (int i = 0; i < CYCLES; i++) {
for (String s : STRINGS) {
boolean b = false;
b = PATTERN.matcher(s).matches();
}
}
}
@Benchmark
public void testParseLong() {
for (int i = 0; i < CYCLES; i++) {
for (String s : STRINGS) {
boolean b = false;
try {
Long.parseLong(s);
b = true;
} catch (NumberFormatException e) {
}
}
}
}
@Benchmark
public void testCharArrayIsDigit() {
for (int i = 0; i < CYCLES; i++) {
for (String s : STRINGS) {
boolean b = false;
for (char c : s.toCharArray()) {
b = Character.isDigit(c);
if (!b) {
break;
}
}
}
}
}
@Benchmark
public void testCharAtIsDigit() {
for (int i = 0; i < CYCLES; i++) {
for (String s : STRINGS) {
boolean b = false;
for (int j = 0; j < s.length(); j++) {
b = Character.isDigit(s.charAt(j));
if (!b) {
break;
}
}
}
}
}
@Benchmark
public void testIntStreamCodes() {
for (int i = 0; i < CYCLES; i++) {
for (String s : STRINGS) {
boolean b = false;
b = s.chars().allMatch(c -> c > 47 && c < 58);
}
}
}
@Benchmark
public void testCharBetween() {
for (int i = 0; i < CYCLES; i++) {
for (String s : STRINGS) {
boolean b = false;
for (int j = 0; j < s.length(); j++) {
char charr = s.charAt(j);
b = '0' <= charr && charr <= '9';
if (!b) {
break;
}
}
}
}
}
}
Updated on Feb 23, 2018
- Add two more cases - one using
charAt
instead of creating extra array and another using IntStream
of char codes
- Add immediate break if non-digit found for looped test cases
- Return false for empty string for looped test cases
Updated on Feb 23, 2018
- Add one more test case (the fastest!) that compares char value without using stream
matches("\\d{2,}")
or try with aPattern
andMatcher
– Guillaume PoletPattern.matches("[a-zA-Z]+", text) == false
can be simplified to!Pattern.matches("[a-zA-Z]+", text)
– SARoseboolean isNumeric = someString.chars().allMatch(x -> Character.isDigit(x));
formMax Malysh
Post. – Yash