#11735: Optimize coercionKind -------------------------------------+------------------------------------- Reporter: goldfire | Owner: (none) Type: task | Status: new Priority: normal | Milestone: Component: Compiler | Version: 7.10.3 Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: Type of failure: Compile-time | Unknown/Multiple performance bug | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Comment (by tdammers): Oh, and also this note further up: {{{ Note [Computing a coercion kind and role] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ To compute a coercion's kind is straightforward: see coercionKind. But to compute a coercion's role, in the case for NthCo we need its kind as well. So if we have two separate functions (one for kinds and one for roles) we can get exponentially bad behaviour, since each NthCo node makes a separate call to coercionKind, which traverses the sub-tree again. This was part of the problem in Trac #9233. Solution: compute both together; hence coercionKindRole. We keep a separate coercionKind function because it's a bit more efficient if the kind is all you want. }}} Which leads me to believe that maybe this particular input I'm testing doesn't hit the `NthCo` branch enough to make a dent there, and thus fares better with the separated-out implementation, but other code might. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/11735#comment:18 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler