@carlosame FYI, this change corrects the busted performance of jsoup in your DOM modification benchmark (code).

New results:

Benchmark                       Mode  Cnt     Score    Error  Units
DOMChangeMark.markChangeDOM    thrpt    3  1539.337 ± 63.919  ops/s
DOMChangeMark.markChangeDOM4J  thrpt    3   363.892 ± 55.999  ops/s
DOMChangeMark.markChangeJdk    thrpt    3  2736.017 ± 83.992  ops/s
DOMChangeMark.markChangeJsoup  thrpt    3   366.781 ± 10.708  ops/s

Where previously jsoup was a pathetic 6 ops/s on this machine. As mentioned in the commit, the flow of the benchmark's removal was running as O(n²); now it's O(n).

From a quick look at the JDK's DOM impl, the primary difference that gives the extra speed for this specific benchmark is that they are using a linked-list vs an array, and so the construction/removal is just a simple field change. And as it's XML they're not needing to lowercase normalize the tag names or attributes etc. There are pros and cons to using a linked list in a DOM; for jsoup's general use case I've found the list is better, but I'll noodle on that some more.

From a quick look at the JDK's DOM impl, the primary difference that gives the extra speed for this specific benchmark is that they are using a linked-list vs an array, and so the construction/removal is just a simple field change. And as it's XML they're not needing to lowercase normalize the tag names or attributes etc. There are pros and cons to using a linked list in a DOM; for jsoup's general use case I've found the list is better, but I'll noodle on that some more.

Thanks for the heads-up, I'll update the benchmarks once 1.21.2 is out.

Css4j's native DOM uses a linked list but has all the HTML overhead, it scores 1539 in the benchmark that you ran. The JDK DOM used to be about 55% faster than css4j DOM but has improved in recent JDKs. The JDK is generally faster because it does some weird internal optimizations; those sometimes trigger errors (e.g. ArrayIndexOutOfBoundsException) during thread-heavy benchmarks.