We’ve been spending a lot of times looking at performance for one of our enterprise customers. As part of the investigation
he came up with an interesting and very important find… He found that
hashMap.get("String") was much
slower under the new VM than under the old VM. Its these sort of “simple” finds that help narrow down bigger issues
in performance that might impact a lot of things in Codename One, since HashMap is so ubiquitous the implications
of improving it can be huge.
One of the advantages of the new VM is in the fact that we now also control the full API including the HashMap API and
can rewrite large portions of it natively for full performance advantage. But it seems that a lot of the things that were
really affecting performance were in the more “subtle” pieces touched by the
String class and
We moved many small bottlenecks into native code e.g.
But one of the bigger performance improvements came from improving the behavior of the finalizer methods
which reduces CPU usage and potential stuttering.
We’ve made some updates to the developer guide specifically a lot of details related to the
regenerate native interfaces for older cn1libs to be compatible. If you don’t do that compilation with such
libraries can result in a compilation error.