I think this goes against the grain: StringBuilder is a simple builder for
non-localized unformatted output. For formatted output there’s Formatter,
String.format and, with JEP 430[1], FMT.”…” to help output string data in more
advanced, formatted and localized ways. The latter has traditionally lacked a
bit in raw throughput, and String templates is a deep-cut approach to try and
improve both the programming model and the performance characteristics.
Let’s compare a few existing APIs (JDK 22 ea, —enable-preview enabled for the
FMT.”…” micro:
@Benchmark
public String toStringAppendHex8() {
StringBuilder result = new StringBuilder();
result.append(Integer.toHexString(2048));
result.append(Integer.toHexString(31337));
result.append(Integer.toHexString(0xbeefcace));
result.append(Integer.toHexString(9000));
result.append(Integer.toHexString(4711));
result.append(Integer.toHexString(1337));
result.append(Integer.toHexString(2100));
result.append(Integer.toHexString(2600));
return result.toString();
}
@Benchmark
public String toStringFMTHex8() {
int h1 = 2048, h2 = 31337, h3 = 0xbeefcace, h4 = 9000, h5 = 4711, h6 =
1337, h7 = 2100, h8 = 2600;
return FMT."%x\{h1}%x\{h2}%x\{h3}%x\{h4}%x\{h5}%x\{h6}%x\{h7}%x\{h8}";
}
@Benchmark
public String toStringFormatHex8() {
StringBuilder result = new StringBuilder();
Formatter f = new Formatter(result);
f.format("%x", 2048);
f.format("%x", 31337);
f.format("%x", 0xbeefcace);
f.format("%x", 9000);
f.format("%x", 4711);
f.format("%x", 1337);
f.format("%x", 2100);
f.format("%x", 2600);
return result.toString();
}
@Benchmark
public String toStringFormatCombinedHex8() {
return String.format("%x%x%x%x%x%x%x%x", 2048, 31337, 0xbeefcace, 9000,
4711, 1337, 2100, 2600);
}
Benchmark Mode Cnt Score Error Units
StringBuilders.toStringAppendHex8 avgt 15 19,521 ± 0,608 ns/op
StringBuilders.toStringFMTHex8 avgt 15 10,958 ± 0,838 ns/op
StringBuilders.toStringFormatHex8 avgt 15 378,327 ± 9,302 ns/op
StringBuilders.toStringFormatCombinedHex8 avgt 15 235,609 ± 6,315 ns/op
As we can see FMT.”%x…” is already a great deal faster here, while
String.format and Formatter leave something to be desired.
One thing missing with the FMT approach is a natural way to format into
pre-existing StringBuilder/Appendable (like the toStringFormatHex8 benchmark
models). I think adding that would be a way more interesting development than
to add more formatting methods to StringBuilder.
/Claes
[1] https://openjdk.org/jeps/430
30 sep. 2023 kl. 14:06 skrev 温绍锦(高铁) <shaojin.we...@alibaba-inc.com>:
There are many combinations of `append(Integer.toHexString(i))` and
`append(Long.toHexString(i))` in the JDK code. I request to add method
`appendHex(int)` and `appendHex(int)` to StringBuilder and StringBuffer. This
can reduce duplicate code within the JDK and will also be useful to users.
I submitted a PR ( https://github.com/openjdk/jdk/pull/15998 ), including the
code and tests for the new method. I also replaced the code that uses the
append + toHexString combination inside the JDK with appendHex.
Please review and don't hesitate to critique my approach and patch.
