Only codeblocks for functions that need an activation, which we can determine statically, need to keep their SymbolTable around after codegen.
Created attachment 25873 [details] patch
Comment on attachment 25873 [details] patch r=me
Oliver just pointed out that this will cause problems for the debugger -- you should check needsFullScopeChain on the CodeBlock instead, since that also includes the case of debugging.
Comment on attachment 25873 [details] patch removing r+ flag due to regression Sam found -- f.arguments.argumentName fails :-(
Created attachment 25903 [details] updated patch
Fixed in r39157.