source: doc/theses/aaron_moss_PhD/phd/conclusion.tex@ 1dda8de

\chapter{Conclusion}

Decades after its first standardization, the C language remains a widely-used tool and a vital part of the software development landscape. 
The \CFA{} language under development at the University of Waterloo represents an evolutionary modernization of C with expressive modern language features paired with strong C backwards-compatibility. 
This thesis has contributed to these project goals in a variety of ways, including the addition of a generic-types language feature (Chapter~\ref{generic-chap}) and refinement of the \CFA{} overload selection rules to produce a more expressive and intuitive model (Section~\ref{conv-cost-sec}). 
Based on the technical contribution of the resolver prototype system (Section~\ref{rp-features-sec}), I have also made significant improvements to \CFA{} compilation performance, including un-combined bottom-up expression traversal (Section~\ref{arg-parm-matching-sec}), deferred-cached assertion satisfaction (Section~\ref{assn-sat-sec}), and a novel persistent union-find type environment data structure (Section~\ref{env-persistent-union-find}). 
The combination of these practical improvements and added features significantly improve the viability of \CFA{} as a practical programming language.

Further improvements to the \CFA{} type-system are still possible, however. 
One area suggested by this work is development of a scheme for user-defined conversions; to integrate properly with the \CFA{} conversion model, there would need to be a distinction between safe and unsafe conversions, and possibly a way to denote conversions as explicit-only or non-chainable. 
Another place for ongoing effort is improvement of compilation performance; I believe the most promising direction for that is rebuilding the \CFA{} compiler on a different framework than Bilson's \CFACC{}. 
The resolver prototype presented in this work has good performance and already has the basics of \CFA{} semantics implemented, as well as many of the necessary core data structures, and would be a viable candidate for a new compiler architecture. 
An alternate approach would be to fork an existing C compiler such as Clang~\cite{Clang}, which would need to be modified to use one of the resolution algorithms discussed here, as well as various other features introduced by Bilson~\cite{Bilson03}.