I want to compare an URI String over different patterns in java and I want fastest code possible.
Should I use :
if(uri.contains("/br/fab") || uri.contains("/br/err") || uri.contains("/br/sts")
Or something like :
if(uri.matches(".*/br/(fab|err|sts).*"))
Note that I can have a lot more uri and this method is called very often.
What is the best answer between my choices ?
If you're going to use a regular expression, create it up-front and reuse the same Pattern object:
private static final Pattern pattern = Pattern.compile(".*/br/(fab|err|sts).*");
Do you actually need the ".*" at each end? I wouldn't expect it to be required, if you use Matcher.find().
Which is faster? The easiest way to find out is to measure it against some sample data - with as realistic samples as possible. (The fastest solution may very well depend on
Are you already sure this is a bottleneck though? If you've already measured the code enough to find out that it's a bottleneck, I'm surprised you haven't just tried both already. If you haven't verified that it's a problem, that's the first thing to do before worrying about the "fastest code possible".
If it's not a bottleneck, I would personally opt for the non-regex version unless you're a regex junkie. Regular expressions are very powerful, but also very easy to get wrong.
I've done a test and it is faster to use contains. As Ewan Todd said, they both fast enough to don't really bother with that.
Both are fast enough, but contains is faster. Facts: ~20mil ops vs ~1mil ops
Using the following jmh code to test
@State(Scope.Benchmark)
public class Main {
private String uri = "https://google.com/asdfasdf/ptyojty/aeryethtr";
@Benchmark
@Warmup(iterations = 5)
@Measurement(iterations = 5)
@Fork(value = 1, warmups = 0)
public void initContains() throws InterruptedException {
if (uri.contains("/br/fab") || uri.contains("/br/err") || uri.contains("/br/sts")) {}
}
@Benchmark
@Warmup(iterations = 5)
@Measurement(iterations = 5)
@Fork(value = 1, warmups = 0)
public void initMatches() throws InterruptedException {
if (uri.matches(".*/br/(fab|err|sts).*")) {}
}
public static void main(String[] args) throws Exception {
org.openjdk.jmh.Main.main(args);
}
}
The results
# Run complete. Total time: 00:00:37
Benchmark Mode Cnt Score Error Units
Main.initContains thrpt 5 21004897.968 ± 1987176.746 ops/s
Main.initMatches thrpt 5 1177562.581 ± 248488.092 ops/s
I would expect contains() to be faster since it won't have to compile and iterate through a (relatively) complex regular expression, but rather simply look for a sequence of characters.
But (as with all optimisations) you should measure this. Your particular situation may impact results, to a greater or lesser degree.
Furthermore, is this known to be causing you grief (wrt. performance) ? If not, I wouldn't worry about it too much, and choose the most appropriate solution for your requirements regardless of performance issues. Premature optimisation will cause you an inordinate amount of grief if you let it!
UPDATE: I know this is not the best benchmark code and for each case there are several ways to optimize it.
What I wanted to achieve was, for a regular developer that will use the simpler ways of doing things and it's not a JVM expert, that's the "common" way to use it, so here it goes.
ORIGINAL:
The below code produced the following output
contains took: 70
matches took: 113
matches with pre pattern took: 419
The test class
public class MatchesTester {
public static void main(String[] args) {
String typeStr = "Nunc rhoncus odio ac tellus pulvinar, et volutpat sapien aliquet. Nam sed libero nec ex laoreet pretium sed id mi. Aliquam erat volutpat. Aenean at erat vitae massa iaculis mattis. Quisque sagittis massa orci, sit amet vestibulum turpis tempor a. Etiam eget venenatis arcu. Nunc enim augue, pulvinar at nulla ut, pellentesque porta sapien. Maecenas ut erat id nisi tincidunt faucibus eget vel erat. Morbi quis magna et massa pharetra venenatis ut a lacus. Vivamus egestas vitae nulla eget tristique. Praesent consectetur, tellus quis bibendum suscipit, nisl turpis mattis sapien, ultrices mollis leo quam eu eros.application/binaryNunc rhoncus odio ac tellus pulvinar, et volutpat sapien aliquet. Nam sed libero nec ex laoreet pretium sed id mi. Aliquam erat volutpat. Aenean at erat vitae massa iaculis mattis. Quisque sagittis massa orci, sit amet vestibulum turpis tempor a. Etiam eget venenatis arcu. Nunc enim augue, pulvinar at nulla ut, pellentesque porta sapien. Maecenas ut erat id nisi tincidunt faucibus eget vel erat. Morbi quis magna et massa pharetra venenatis ut a lacus. Vivamus egestas vitae nulla eget tristique. Praesent consectetur, tellus quis bibendum suscipit, nisl turpis mattis sapien, ultrices mollis leo quam eu eros.";
int timesToTest = 10000;
long start = System.currentTimeMillis();
int count = 0;
//test contains
while(count < timesToTest){
if (typeStr.contains("image") || typeStr.contains("audio") || typeStr.contains("video") || typeStr.contains("application")) {
//do something non expensive like creating a simple native var
int a = 10;
}
count++;
}
long end = System.currentTimeMillis();
System.out.println("contains took: "+ (end - start));
long start2 = System.currentTimeMillis();
count = 0;
while(count < timesToTest){
if (typeStr.matches("(image|audio|video|application)")) {
//do something non expensive like creating a simple native var
int a = 10;
}
count++;
}
long end2 = System.currentTimeMillis(); //new var to have the same cost as contains
System.out.println("matches took: "+ (end2 - start2));
long start3 = System.currentTimeMillis();
count = 0;
Pattern pattern = Pattern.compile("(image|audio|video|application)");
while(count < timesToTest){
if (pattern.matcher(typeStr).find()) {
//do something non expensive like creating a simple native var
int a = 10;
}
count++;
}
long end3 = System.currentTimeMillis(); //new var to have the same cost as contains
System.out.println("matches with pre pattern took: "+ (end3 - start3));
}
If the bit you are trying to match against is always at the beginning, or end, or is in some other way predictable then: neither!
For example, if urls are like http://example.com/br/fab or http://example.com/br/err all the time, then you could store "br/fab" and "br/err" etc in a HashSet or similar, and then given an incoming URL, chop off the last part of it and query the Set to see if it contains it. This will scale better than either method you gave (with a HashSet it should get no slower to lookup entries no matter how many there are).
If you do need to match against substrings appearing in arbitrary locations... it depends what you mean by "a lot more". One thing you should do regardless of the specifics of the problem is try things out and benchmark them!
its much faster if you use indexOf().
if(uri.indexOf("/br/fab")>-1 || uri.indexOf("/br/err")>-1 || uri.indexOf("/br/sts") >-1 )
{
your code.
}
and problem with contains() is internally it creates a Matcher(java.util.regex.Matcher) object and evalates the expression.
Matcher is a very costly thing if processing large amount of data.
They're both fast enough to be over before you know it. I'd go for the one that you can read more easily.
