Attached test case is much slower with -n: -- js : 0,121 ms js -m -a : 0,253 ms js -n : 5,400 ms js -n -a : 28,622 ms js -m -n -a : 50,447 ms -- Release build, 32-bit OS X, revision 9ff00d53b5a5. This is a generated testcase but based on one of my pet projects, and I can imagine websites doing things like this.

The problem here was the quadratic search needed to find the right type object to use for new arrays/objects in scripts containing tons of initializers. Type objects for allocation sites are now tracked in a compartment-wide hashtable, which gives better asymptotic complexity and is a better use of memory (most type objects created are not initializers). Now I get these times (computed by timing how long a load(...) takes, as much of the overhead is during parsing). js 72 js -m 72 js -m -a 183 js -m -n 113 js -m -n -a 344 d8 162