#14154: Some cocktail of features causes GHC panic -------------------------------------+------------------------------------- Reporter: Iceland_jack | Owner: (none) Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.3 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 goldfire): Could we fix one case this with a well-placed zonk? More broadly (re: could this happen elsewhere?) I've grown increasingly uncomfortable with the way GHC handles tyvars and zonking. When the type- checker fills in a metavariable, it then assumes that the metavar and its new value are equal. Except, of course, pure code can (and will, to our dismay) spot the difference. So, we could make `tc` versions of various type-manipulation functions (we indeed already do) and make them monadic, looking through metavars (this bit is new). That's disappointing, somehow. But I don't really see a better option. If we did this -- and did it reliably -- I think we could remove `zonkTcType` and friends (keeping only `zonkTcTypeToType` and friends). -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/14154#comment:7 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler