In this experience report we present our early work at improving Smalltalk performance by inlining message sends during compilation. Smalltalk developers typically write small method bodies with one or two statements, this limits a compiler’s ability to perform many optimizations, e.g. common sub-expression elimination. The overhead of message sends in a dynamically typed language introduces a high overhead cost when looking up the class of the message receiver. Inlining messages into a method body produces methods with fewer message sends and more statements allowing the compiler to optimize and generate efficient executable code and avoid lookup overhead. There are several challenges to inlining messages in Smalltalk that need to be resolved.

We describe a generalized inlining approach that can be applied regardless of the execution target - bytecode, threaded-execution, or machine code - allowing interactive debugging with perfect verisimilitude to the JITed code. In this experience report, we describe the inlining approach taken for the Zag Smalltalk compiler that solves for these issues and improves performance.