Abstract
Any decent optimising compiler for a lazy language like Haskell must include a strictness analyser. The results of this analysis allow the compiler to use call-by-value instead of call-by-need, and that leads to big performance improvements. It turns out that strictness analysis is an interesting problem from a theoretical point of view, and the 1980’s saw a huge rash of papers on the subject. There were fewer, many, many fewer, papers that described real implementations.
This paper presents the fruits of a decade-long experience with strictness analysis, in the context of the Glasgow Haskell Compiler, an optimising compiler for Haskell. In particular, we recently re-engineered the existing strictness analyser that used forward abstract interpretation, replacing it with a new one that uses backward analysis instead.
This unpublished draft is better explained, and much closer to the actual implementation in GHC, than our earlier version.