[jdk17] RFR: JDK-8263321: Regression 8% in javadoc-steady in 17-b11

[Pavel Rappo]

On Mon, 14 Jun 2021 13:43:40 GMT, Hannes Wallnöfer <hannesw at openjdk.org> wrote:

> This change fixes a performance regression caused by passing instances of `com.sun.tools.javac.util.Name` to a `html.markup.Text` object, where their content is accessed using`CharSequence` methods which are implemented very inefficiently. 
> 
> Since improving `javac.util.Name` will likely require a bit more work (see JDK-8268622), a safe short term solution for JDK 17 is to make sure instances of `Name` are converted to `String` before invoking said `CharSequence` methods. The main fix is to convert to `String` early in `Entity.escapeHtmlChars(CharSequence)` which is invoked by the `Text` constructor. There are a few more `.toString()` invocations added in places where a `Name` is appended to a `StringBuilder`, although these have less performance impact.

Based on these numbers, 8% degradation followed by 5% improvement, this patch reverses approximately half of the initial degradation introduced by JDK-8260223, right? (Had to remind myself how tricky percentages can be.) If so, then where is the other half? Should we be concerned about it?

-------------

PR: https://git.openjdk.java.net/jdk17/pull/42