Core Java
Why a synchronized StringBuffer was never a good idea
Introduction
StringBuffer is a synchronized class for mutable strings. The main problem with making it synchronized is that
- It was usually used as a local variable so making it synchronized just made it slower.
- It was never a good idea to use it in a multi-threaded way. This problem is that developers assumed that methods which used StringBuffer were themselves thread safe when they were not.
The problem with StringBuffer
This is an example from a real class which is used in production in many trading systems. It’s not a commonly used but you might assume that StringBuffer gives you thread safety, when it doesn’t.
public void addProperty(String name, String value) { if (value != null && value.length() > 0) { if (sb.length() > 0) { sb.append(','); } sb.append(name).append('=').append(value); } }
While individual calls are thread safe, multiple calls are not. It is almost impossible to find a good use for StringBuffer that doesn’t involve multiple calls (including toString)
A puzzle
Imagine three threads call (in no particular order)
T1: addProperty("a", "b"); T2: addProperty("c", "d"); T3: sb.toString();
write a program which will generate every possible output of T3’s sb.toString() I found 89. With thread safety, you might reduce this to 4.
Note
If you used StringBuilder it would be worse, but at least you might not assume your method is thread safe when it is not. e.g. SimpleDateFormat uses StringBuffer
Reference: Why a synchronized StringBuffer was never a good idea from our JCG partner Peter Lawrey at the Vanilla Java blog.
>While individual calls are thread safe, multiple calls are not.
It would be better for understanding to name them not multiple but chained calls.
They do not necessarily need to be chained[1]. How about ‘consecutive’?
[1] http://en.wikipedia.org/wiki/Method_chaining