[ ghc-Bugs-428694 ] Simplifier and optimization flags

noreply@sourceforge.net noreply@sourceforge.net
Thu, 31 May 2001 04:26:12 -0700


Bugs item #428694, was updated on 2001-05-30 08:26
You can respond by visiting: 
http://sourceforge.net/tracker/?func=detail&atid=108032&aid=428694&group_id=8032

Category: Compiler
Group: 5.00
>Status: Closed
>Resolution: Wont Fix
Priority: 5
Submitted By: Axel Krauth (axxl)
Assigned to: Nobody/Anonymous (nobody)
Summary: Simplifier and optimization flags

Initial Comment:
compiling 

>module Parse where
>
>import Parsec
>
>whiteSpace  = skipMany1 (oneOf " ")
>

or 

>module Parse where
>
>import Parsec 
>
>ttoken :: Parser TToken
>ttoken = do { ; return TIf }
>
>data TToken = TIf

via

ghc -O2 -funbox-strict-fields
-funfolding-use-threshold30 -package text -c Parse.hs
-o Parse.o

results both times in

ghc-5.00.1: panic! (the `impossible' happened, GHC
version 5.00.1):
	simplCore/Simplify.lhs:1468: Non-exhaustive patterns
in function cat_evals

removing any option compiles.
(the latter also complains about some bad eta expands)

i'm using the linux version distributed from the ghc
homepage.

----------------------------------------------------------------------

>Comment By: Simon Peyton Jones (simonpj)
Date: 2001-05-31 04:26

Message:
Logged In: YES 
user_id=50165

Yes, GHC 5.00 and its patches won't work with -funbox-
strict-fields, if a library is compiled with different 
options to the library client.  This is fixed in the head, 
but it's too awkward to back-patch 5.00 (may introduce new 
bugs etc).

We should document this bug, but it's a wont-fix one.

----------------------------------------------------------------------

You can respond by visiting: 
http://sourceforge.net/tracker/?func=detail&atid=108032&aid=428694&group_id=8032