Dead Code Elimination (DCE) behavior during Haxe 4 production optimization

Preventing DCE Pruning in Haxe 4

To preserve specific classes, methods, or fields from Dead Code Elimination (DCE) without disabling global optimizations, use the @:keep metadata annotation. This instructs the Haxe compiler to treat the annotated symbol as "used," ensuring it remains in the final production binary or script regardless of whether a static reference exists in the AST.

Implementation Methods

Depending on whether your code is static or generated via macros, use one of the following two mechanisms:

• Manual Annotation: Apply @:keep directly above the class, function, or field definition.

  @:keep
  class DynamicHandler {
      public function execute() { 
          trace("Handled dynamically"); 
      }
  }

• Programmatic Retention: If you are using a macro to generate code or handle dynamic registration, use the haxe.macro.Compiler.keep method to force the compiler to retain specific types or fields during the compilation pass.

Likely Cause of Production Discrepancies

The discrepancy between local and production builds usually stems from the -D dce flag. Development builds often use std or no, which are permissive. Production builds typically utilize full DCE. When you access a class via Type.getClassByName or Reflect, the compiler cannot statically trace the dependency chain, leading it to assume the code is unreachable and safe to prune.

Verification Steps

1. Compile the project using the production flag: haxe -D dce=full [other-args].
2. Execute the specific path that relies on dynamic reflection.
3. If the feature fails in production but works in debug, verify that the target class is explicitly marked with @:keep.

Note: @:keep prevents the symbol from being removed by DCE, but it does not necessarily prevent other optimizations like inlining. If you are targeting a platform with strict reflection requirements, ensure the entire class hierarchy is considered.

To provide a more specific recommendation, are you using a custom target (e.g., a specific C++ or JS target) or the standard Haxe targets?