Towards a flexible Pharo Compiler

International audienceThe Pharo Smalltalk-inspired language and environment started its development with a codebase that can be traced back to the original Smalltalk-80 release from 1983. Over the last years, Pharo has been used as the basis of many research projects. Often these experiments needed changes related to the compiler infrastructure. However, they did not use the existing compiler and instead implemented their own experimental solutions. This shows that despite being an impressive achievement considering its age of over 35 years, the compiler infrastructure needs to be improved. We identify three problems: (i) The architecture is not reusable, (ii) compiler can not be parametrized and (iii) the mapping between source code and bytecode is overly complex. Solving these problems will not only help researchers to develop new language features, but also the enhanced power of the infrastructure allows many tools and frameworks to be built that are important even for day-to-day development, such as debuggers and code transformation tools. In this paper we discuss the three problems, show how these are solved with a new Compiler model. We present an implementation, Opal, and show how Opal is used as the bases for many important tools for the everyday development of Pharo 3