use_string_buffers reports, although intermediate strings are used

[Hixie]

The first block in the following pattern triggers use_string_buffers but the second block (which uses StringBuffer) is ~7% slower:

const int count = 100000;

void test(String s) { }

void main() {
  Stopwatch t = new Stopwatch();

  // USING STRING
  t.start();
  String s = '';
  for (int i = 0; i < count; i += 1) {
    s = s + '-$i';
    test(s);
  }
  print(t.elapsedMicroseconds);

  // USING STRINGBUFFER
  t.reset();
  StringBuffer b = new StringBuffer();
  for (int i = 0; i < count; i += 1) {
    b.write('-$i');
    test(b.toString());
  }
  print(t.elapsedMicroseconds);
}

[lrhn]

This is a tricky edge-case where a string is being built by repeated concatenation, the classical case where you should use StringBuffer, but where all the intermediate strings are also used. I'm not sure this isn't an acceptable false-positive for the hint, especially since performance isn't really an issue.

The benchmark isn't very precise. Try https://dartpad.dartlang.org/8515b26b4d4b4807d24bbe7cb36a8f91 instead - it runs each of these tests in a loop for two seconds, counting how many rounds it manages to do in that time, then repeats that three times to ensure all methods should be warm. It also adds versions that avoid creating the intermediate '-$i' string.

In JavaScript, there is no measurable difference in speed between the two versions (and a big advantage to versions that do not create the '-$o' intermediate string), but then, V8 does constant time concatenation, and we never do anything to the strings that forces flattening. If we change test to do s.indexOf('-'), then it gets 300 times slower.

When run on the VM, you need to lower count to 10000 or it won't even do one round in two seconds on my home computer, and the StringBuffer versions are ~25% less efficient than the plain concatenation then. It gets better asymptotically, at count 100000, the StringBuffer versions are faster than the + concatenation.

[pq]

Bumping to P2 since this one is a proposed canonical lint (dart-lang/core#784).

[srawlins]

This is a duplicate request of #57619.