given a proposed Core-Core transformation, does the transformation meet
#3693: Show stack traces -------------------------------------+------------------------------------ Reporter: jpet | Owner: Tarrasch Type: feature request | Status: new Priority: normal | Milestone: 7.10.1 Component: Runtime System | Version: 6.10.4 Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: Unknown/Multiple Type of failure: None/Unknown | Difficulty: Unknown Test Case: | Blocked By: Blocking: | Related Tickets: -------------------------------------+------------------------------------ Comment (by scpmw): the specfification or not? Simon covered the basic idea pretty well. From the transformation point of view, the general rule of thumb is that source notes want to stay on covered code (even if transformed), but don't care too much about new code getting added in ("soft scope"). So we can for example work with fusion by simply floating "upwards": {{{ src<1.hs> foldr k z (src<2.hs> build g) ===> src<1.hs> src<2.hs> g k z }}} Cost-centres would disallow this kind of transformation, as the new stacks have little to do with the old one. Source notes on the other hand want the transformation to still happen, so we just annotate the new expression with everything that appeared on the deconstructed source expression. The paper goes into (way) more detail here, but the short version is that while not always optimal, this approach is always legal. The only special case so far is `let`- and `case`-floating, where we run into the same problems as cost-centres (and use pretty much the same solution). Additionally we decide to not care too much about allocation costs at a number of points where it would have made things too messy. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/3693#comment:65 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler