Two qq: 1) Is this for trunk or 1.6?2) Does this do anything to fix issue #3064<http://code.google.com/p/google-web-toolkit/issues/detail?id=3064>? Seems like that bug is highly related, and might be a compelling enough reason to put this into 1.6.
Scott On Fri, Nov 7, 2008 at 10:35 AM, Lex Spoon <[EMAIL PROTECTED]> wrote: > The attached patch tweaks ControlFlowAnalyzer to do its upref visiting > eagerly. This results in 1-permutation Showcase compiling about 1-2 > seconds faster for me, an improvement of 3-6%. > > The reason for upref visiting at all is that if you rescue a virtual > method, you also need rescue any overriding methods in subclasses. > The current algorithm does this by scanning the whole program, in a > loop, to find any such overriding methods. The attached patch rescues > these methods eagerly. It takes three changes to make this happen: > > 1. When a method is rescued, immediately scan all overriding methods. > 2. To make #1 efficient, build a reverse index of method overriding > when the analyzer starts up. > 3. When a type is rescued, immediately scan all of its methods, just > like upref visitor did. > > It also adds PerfLogger calls to GWTCompiler, to make it easy to time > an entire compile. > > Can you review it Scott, or toss it to someone else? > > -Lex > --~--~---------~--~----~------------~-------~--~----~ http://groups.google.com/group/Google-Web-Toolkit-Contributors -~----------~----~----~----~------~----~------~--~---
