Optimisation puzzle with list of list folds and mconcat
As observed in !4890 (comment 326560) optimisation of some list of list folds is inconsistent across superficially similar forms. Specifically:
print $ foldl' (+) 0 $ mconcat $ map (\i -> [0..i]) [0..n]performs noticeably better (half the allocations and a just over half the runtime of:
print $ foldl' (+) 0 $ foldr (++) [] $ map (\i -> [0..i]) [0..n]@simonpj notes differences in the rules that fired in the two cases: !4890 (comment 326990)
Some more comments on this subtopic in:
This topic is a tangent from the original concerns of !4890 (closed) and so is best explored and tracked separately.