#15226: GHC doesn't know that seq# produces something in WHNF -------------------------------------+------------------------------------- Reporter: dfeuer | Owner: (none) Type: bug | Status: closed Priority: normal | Milestone: 8.6.1 Component: Compiler | Version: 8.4.3 Resolution: fixed | Keywords: Operating System: Unknown/Multiple | Architecture: | Unknown/Multiple Type of failure: Runtime | Test Case: performance bug | perf/should_run/T15226, 15226a Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Phab:D4796 Wiki Page: | -------------------------------------+------------------------------------- Comment (by simonpj):
Considering seq# strict can be rather bad, I believe.
I'm not sure about that. First, let's remember (I always forget this) that `seq#` has type {{{ seq# ::a -> State# s -> (# State# s, a #) }}} That is, it involves the IO monad. It's used to implement {{{ evaluate :: a -> IO a evaluate a = IO $ \s -> seq# a s -- NB. see #2273, #5129 }}} See Trac #5129. Now consider {{{ \s. let x = blah in let y = x+1 in case seq# x s of (# s', x' #) -> ... }}} It seems fine to me transform this to {{{ \x. case blah of x -> let y = x+1 in case seq# x s of (# s', x' #) -> ... }}} What if the `seq#` is after some other IO operations thus: {{{ \s. let x = blah in case f s of (# s1, r #) -> case seq# x s of (# s2, x' #) -> ...... }}} Now you might worry that `x` might be evaluated (and throw an exception) before `f` gets a chance to run. But it doesn't: there's a hack in the strictness analyser (see `See Note [IO hack in the demand analyser]` in `DmdAnal`) that will make `x`'s binding lazy; in effect the strictness analyser treats the `case f s of ...` as if it had an extra invisible alternative not mentioning `x`. It's not that important. But I think that `seq#` can safely be strict in `x`. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/15226#comment:20 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler