[prev in list] [next in list] [prev in thread] [next in thread] 

List:       hugs-bugs
Subject:    Re: [Hugs-bugs] internal error
From:       Ross Paterson <ross () soi ! city ! ac ! uk>
Date:       2004-10-22 10:20:55
Message-ID: 20041022102054.GA13023 () soi ! city ! ac ! uk
[Download RAW message or body]

On Fri, Oct 22, 2004 at 07:45:32AM +0000, fool@sdf-eu.org wrote:
> bla = (if True then (+) else (+)) 0 0

An elegant report.  Here's another version:

	bla = (if True then id else negate) (1::Int)

Unfortunately it seems to point at a whole class of longstanding bugs
in the peephole optimizer: it takes no account of branch targets.
(Here MKAP 1; MKAP 1 => MKAP 2, but the second MKAP was the target of
a TEST branch.)  Another of these optimizations was turned off two
years ago for a similar reason, but they all seem similarly flawed.

[prev in list] [next in list] [prev in thread] [next in thread]