Re: rhug failure on alpha

[Andrew Haley <aph at cambridge dot redhat dot com>]

Jeff Sturm writes:
 > On 17 Mar 2002, Tom Tromey wrote:
 > > I built 3.1 on Linux/alpha and then used it to build rhug.  I get this
 > > failure:
 > 
 > I'd guess this doesn't happen on x86?  What revision of alpha (output of
 > config.guess)?
 > 
 > Early alphas didn't have byte load/store instructions, so perhaps a
 > BIT_FIELD_REF is generated somewhere to mask out the needed bits.  I don't
 > know why this would appear up at the tree level though.

I've seen this happen before, but not for some time.  Fairly innocuous
field references can be changed into bit field refs.

I remember looking at this and coming to the conclusion that we really
need a fold() with Java semantics.  I think it would be a deal simpler
than the C/C++ version.

Andrew.