#13044: make it possible to apply GHC rewrite rules to class methods -------------------------------------+------------------------------------- Reporter: George | Owner: Type: task | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.0.1 Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: | Unknown/Multiple Type of failure: None/Unknown | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Changes (by nomeata): * cc: nomeata (added) -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/13044#comment:1 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler

#13044: make it possible to apply GHC rewrite rules to class methods -------------------------------------+------------------------------------- Reporter: George | Owner: Type: task | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.0.1 Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: | Unknown/Multiple Type of failure: None/Unknown | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Comment (by nomeata): Currently, rewrite rules, as well as unfolding (i.e. inlining) are active in certain phases, which are currently counted down from, I belive, 3 to 0. Exending this system of phases to something more expressive has always been a thought. Maybe it would help here? The smallest delta that might help is simply adding a a number of extra phases, say, from 7 to 4. The compiler would annotate all type class dictionary access functions (i.e. class methods) with `INLINE [~4]`, which would prevent them from being rewritten to the actual implementation. Then rewrite rules active in phases 7, 6 and 5 would have a pretty good chance of applying to class methods. Would this work? Obviously, one might want nicer ways of specifying phases (i.e. fixed names, or even arbitrary news with ordering constraints), but that is a slightly orthogonal discussion. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/13044#comment:5 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler

#13044: make it possible to apply GHC rewrite rules to class methods -------------------------------------+------------------------------------- Reporter: George | Owner: Type: task | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.0.1 Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: | Unknown/Multiple Type of failure: None/Unknown | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Comment (by bgamari): The limitation described here leads to quite a bit of boilerplate in GHC itself to ensure that users can write rules against, e.g. `(==)`. See, for instance, `GHC.Int.eqInt64`. Moreover, I have seen a number of well-known packages (`text` being one example) which mistakenly try to match overloaded operators in their rewrite rules. Thankfully we now produce warnings in these cases, but previously these were latent performance bugs just waiting to be noticed. It would be great to eliminate this sharp edge. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/13044#comment:7 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler