diff options
Diffstat (limited to 'src/jdk/nashorn/internal/codegen/LocalVariableTypesCalculator.java')
-rw-r--r-- | src/jdk/nashorn/internal/codegen/LocalVariableTypesCalculator.java | 7 |
1 files changed, 7 insertions, 0 deletions
diff --git a/src/jdk/nashorn/internal/codegen/LocalVariableTypesCalculator.java b/src/jdk/nashorn/internal/codegen/LocalVariableTypesCalculator.java index af1a8786..ac3c2934 100644 --- a/src/jdk/nashorn/internal/codegen/LocalVariableTypesCalculator.java +++ b/src/jdk/nashorn/internal/codegen/LocalVariableTypesCalculator.java @@ -93,6 +93,13 @@ import jdk.nashorn.internal.parser.TokenType; * variable to its widest used type after the join point. That would eliminate some widenings of undefined variables to * object, most notably those used only in loops. We need a full liveness analysis for that. Currently, we can establish * per-type liveness, which eliminates most of unwanted dead widenings. + * NOTE: the way this class is implemented, it actually processes the AST in two passes. The first pass is top-down and + * implemented in {@code enterXxx} methods. This pass does not mutate the AST (except for one occurrence, noted below), + * as being able to find relevant labels for control flow joins is sensitive to their reference identity, and mutated + * label-carrying nodes will create copies of their labels. A second bottom-up pass applying the changes is implemented + * in the separate visitor sitting in {@link #leaveFunctionNode(FunctionNode)}. This visitor will also instantiate new + * instances of the calculator to be run on nested functions (when not lazy compiling). + * */ final class LocalVariableTypesCalculator extends NodeVisitor<LexicalContext>{ |