Premature Evalutation when using -O
|Reported by:||remit||Owned by:||nobody|
|Type of failure:||None/Unknown||Test Case:|
|Related Tickets:||Differential Rev(s):|
Hi all, the following program crashes (Prelude.head: empty list) when compiled with optimizations by both ghc-6.0 and ghc-5.04.3 . When the out-commented line is activated it runs fine with 5.04.3, but still crashes when compiled by 6.0 . Without -O both run perfectly fine. Before each command the file is touched: Without the commented-line: % ghc-5.04.3 test1.hs && ./a.out % ghc-5.04.3 -O test1.hs && ./a.out Fail: Prelude.head: empty list % ghc-6.0 test1.hs && ./a.out % ghc-6.0 -O test1.hs && ./a.out Fail: Prelude.head: empty list With it: % ghc-5.04.3 test1.hs && ./a.out % ghc-5.04.3 -O test1.hs && ./a.out % ghc-6.0 test1.hs && ./a.out % ghc-6.0 -O test1.hs && ./a.out Fail: Prelude.head: empty list Full test done on: % ghc --version The Glorious Glasgow Haskell Compilation System, version 6.0 % uname -a Linux localhost 2.4.20 #1 Sat Nov 30 14:46:26 CET 2002 i686 unknown I also tested the prebuilt binaries for linux-x86 and win98, but only with ghc-5.04.3 -O and ghc-6.0 -O (same behaviour). AFAICS this really seems to be a bug in GHC, as a) Laziness should prevent `len' to be evaluated until after program-exit ;o) b) I fail to see why this behaviour would change when upgrading my compiler c) AFAIK -O shouldn't change semantics (well, not in this way at least) d) I don't see why the "_ <- return ()" would make any difference. (Except for messing up the optimizer ;)) If it isn't a bug, please enlighten me. Happy hacking, Remi module Main where import System import Monad main = do args <- getArgs when (null args) $ exitWith ExitSuccess -- _ <- return () let len = read (head args) :: Int print (show len) print len
Note: See TracTickets for help on using tickets.