Replacing static HashSet<String> with Java 7 string switch?
Has anyone investigated whether Java 7 switch is more memory efficient than HashSet<String> or ImmutableSet<String>? I think it might be. Read on....
I ask because HashSet is one of the more bloated java collection classes. It uses 32 bytes for each used entry + an array for the capacity. For example the follow
ing HashSet uses 128 bytes (2*32 + 16 * 4):
private static final HashSet<String> FOO = new HashSet<>();
foo.add("a")
shpub note --category=googplus --syndication=https://plus.google.com/107786897865850743842/posts/VPYQiMunRRY --published="Mon Dec 29 12:22:44 PM PST 2014" --name="" - <<EOF
Replacing static HashSet<String> with Java 7 string switch?
Has anyone investigated whether Java 7 switch is more memory efficient than HashSet<String> or ImmutableSet<String>? I think it might be. Read on....
I ask because HashSet is one of the more bloated java collection classes. It uses 32 bytes for each used entry + an array for the capacity. For example the follow
ing HashSet uses 128 bytes (2*32 + 16 * 4):
private static final HashSet<String> FOO = new HashSet<>();
foo.add("a")
foo.add("b");
boolean isFoo(String s) { return FOO.contains(s);}
We can replace that with:
boolean isFoo(String s) {
switch (s) {
case "a":
case "b":
return true;
default:
return false;
}
}
Java 7 transforms this code into something like this:
switch (s.hashCode()) {
case -1234:
return s.equals("a");
case -55999:
return s.equals("b");
default:
return false;
}
Which maps to a tableswitch or lookupswitch. Squinting it appears that you'd need the following for each entry:
tableswitch:
per-entry: hashcode + result value (8 bytes)
equals test: 4 opcodes (16 bytes)
So it would appear that you have 24bytes/entry.
Performance would be limited by the JVM/Dalvik/Art implementation of tableswitch which can be O(n)
Also obligatory reference to:
http://blog.jamesdbloom.com/JavaCodeToByteCode_PartOne.html
http://blog.jamesdbloom.com/JavaCodeToByteCode_PartOne.html
#googplus
#googplus
