#14927: Hyperbolic area sine is unstable for (even moderately) big negative arguments. -------------------------------------+------------------------------------- Reporter: leftaroundabout | Owner: (none) Type: bug | Status: new Priority: normal | Milestone: Component: libraries/base | Version: 8.2.1 Resolution: | Keywords: Floating | IEEE754 trigonometric Operating System: Unknown/Multiple | Architecture: Type of failure: Incorrect result | Unknown/Multiple at runtime | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Comment (by leftaroundabout): Yes, the cut-offs come out different, but it doesn't really matter: `sqrt (1 + x*x)` is still stable and accurate for a long way even when `x` is so big that `1 + x*x == x*x`. So, whereas it's important that `huge` should not be ''smaller'' than `1/sqrt epsilon ≈ 6.7e7`, it ''can'' actually be much bigger: `1e100` would also work in case of `Double`. Therefore, `1/epsilon ≈ 4.5e15` is a safe compromise that's presumably less likely to give new surprising corner-cases. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/14927#comment:6 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler