Dart String operation much slower than Java

[jumperchen]

Scenario

We have a large zip file which contains a huge xml file to be parsed. In Java, it provides a standard unzip and xml parser class which is quite fast, and we want to do the same things in Dart VM, but Dartlang didn't provide such API for us to use.

After doing some testing, we figure out that the Dart String operation could be the problem.

Here is the simple example to demonstrate such issue (Note: Java's XML parser doesn't need to OP with String value, it could be more faster, see at the bottom info)

import 'dart:io';

// DartLang : Simulate a XML parsing.
main() {
  var start = new DateTime.now().millisecondsSinceEpoch;
  String data = new File('large.xml')
      .readAsStringSync();
  print('File Length: ${data.length}');
  int i = 0;
  List<int> units = data.codeUnits;
  var foo;
  var value;
  for (int j = 0; j < units.length; j++) {
    if (j % 6 == 0) {
      value = [];
    }
    value.add(units[j]);
    if (j % 6 == 5) {
      i++;
      foo = new Foo(new String.fromCharCodes(value));
      value = null;
    }
  }
  print('Parsing time: ${new DateTime.now().millisecondsSinceEpoch - start}ms');
  print('Object creation times: ${i}');
}

class Foo {
  String value;
  Foo(this.value);
}

Note: StringBuffer in Dartlang is much slower than this codeUnits implementation.

// Java
public static void main(String[] args) throws IOException {
    long start = System.nanoTime();
    String f = FileUtils.readFileToString(new File("large.xml"), Charsets.UTF_8);
    Foo foo = null;
    System.out.println(f.length());
    int i = 0;
    char[] chars = f.toCharArray();
    StringBuilder value = null;
    for (int j = 0, k = chars.length; j < k; j++) {
        if (j % 6 == 0) {
            value = new StringBuilder();
        }
        value.append(chars[j]);
        if (j % 6 == 5) {
            i++;
            foo = new Foo(value.toString());
            value = null;
        }
    }
    System.out.println("Parsing time: " + TimeUnit.MILLISECONDS
            .convert(System.nanoTime() - start, TimeUnit.NANOSECONDS) + "ms");
    System.out.println(i);
}
static class Foo {
    String value;
    Foo(String value) {
        this.value = value;
    }
}

The result

FileSize	Java	Dart
20 MB	File Length: 20375011 Parsing time: 572ms Object creation times: 3395835	File Length: 20375011 Parsign time: 938ms Object creation times: 3395835
163 MB	File Length: 163000011 Parsing time: 4060ms Object creation times: 27166668	File Length: 163000011 Parsign time: 6223ms Object creation times: 27166668
326 MB	File Length: 326000011 Parsing time: 7202ms Object creation times: 54333335	File Length: 326000011 Parsign time: 11871ms Object creation times: 54333335

Extra Info

Here is the Java XML Parser example to parse the 326MB file.

    long start = System.nanoTime();
    XMLInputFactory factory = XMLInputFactory.newInstance();
    XMLStreamReader xmlStreamReader = factory
            .createXMLStreamReader(new FileInputStream(new File("large.xml")));
    int i = 0;
    String tagName;

    while(xmlStreamReader.hasNext()) {
        if (xmlStreamReader.next() == XMLStreamConstants.START_ELEMENT) {
            tagName = xmlStreamReader.getName().getLocalPart();
            i++;
        }
    }
    System.out.println("Parsing time: " + TimeUnit.MILLISECONDS
            .convert(System.nanoTime() - start, TimeUnit.NANOSECONDS) + "ms");
    System.out.println(i);  

It takes only 4658 ms to parse.

Note: The attached file for 326MB large.xml.zip

[floitschG]

I had a short look, but couldn't find anything obvious. The VM developers will hopefully have more information. Btw. There is a Stopwatch class that is designed for these kinds of measurements.

Below is the same example using a StringBuffer. Normally it should be faster (or at least as fast), but currently it's ~50% slower.

Also interesting: replacing the j % 6 with the similar (and in theory faster) j.remainder(6) actually made things much worse.

import 'dart:io';

// DartLang : Simulate a XML parsing.
main() {
  var sw = new Stopwatch()..start();
  String data = new File('large.xml')
      .readAsStringSync();
  print('File Length: ${data.length}');
  int i = 0;
  List<int> units = data.codeUnits;
  var foo;
  var buffer = new StringBuffer();
  for (int j = 0; j < units.length; j++) {
    buffer.writeCharCode(units[j]);
    if (j % 6 == 5) {
      i++;
      foo = new Foo(buffer.toString());
      buffer.clear();
    }
  }
  print('Parsing time: ${sw.elapsedMilliseconds}ms');
  print('Object creation times: ${i}');
}

class Foo {
  String value;
  Foo(this.value);
}