Imported From: https://issues.scala-lang.org/browse/SI-5852?orig=1
Reporter: @magarciaEPFL
Assignee: @magarciaEPFL
Blocks #5850

@magarciaEPFL said:
A prototype to evaluate delta-based typeflow-analysis can be found at

https://github.com/magarciaEPFL/scala/compare/master...inliner211 

It also performs two micro-optimizations for speed:

• Inlining ascribes type-flow-computed type to added locals, giving them in general more specific types than those declared in the callee's formal params.
• Some methods calls are resolved as by-product of inlining analysis.
  However those calls to a more specific method (say, j.l.String.equals() instead of its Objectcounterpart) can't be devirtualized (ie madeinvokespecialinstead ofinvokevirtual`) because the JVM verifier complains (looks like our type-flow analysis, which tracks non-nullness, is more precise).

All in all, the single-pass approach isn't much faster than the current implementation. Therefore looks like we'll stick to the current implementation. Closing this ticket as Won't fix.

The technique to track non-nullness should be added to the current implementation, #5850 .