[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Bug#458133: Details, please

On Mon, Dec 31, 2007 at 05:05:38PM +0100, Sune Vuorela wrote:
> On Monday 31 December 2007, Steve M. Robbins wrote:
> > Hi,
> >
> > I don't have any reason to doubt what you say about qt's SIGBUS on
> > hppa.  But where is this documented?  Without an documented
> > evidence trail, we can't collaborate on solving this; instead,
> > each person using "qmake" will run into this independently and
> > waste time re-doing the legwork that you have already done.
> >
> > Rather than relying on supposition and rumour, maybe we could
> > leave this bug open until the actual facts are in evidence.
> > The bug was closed on the strength of
> >
> >     According to some hppa people it is a misconfiguration somewhere
> >     on the buildds, so we are currently doing nothing except poking
> >     hppa buildd people occasionally.
> This is from right after midnight CET in #debian-devel on the day change to 
> the 31st
> 00:05 < pusling> lamont: what is "bus error" on hppa ?
> 00:06 < lamont> pusling: unaligned load/store
> 00:06 < lamont> pusling: more specically, valid address, permission fault
> 00:06 < pusling> lamont: is unaligned load/store enough to make builds fail ?
> 00:06 < lamont> pusling: was - I turned that off rather recently

So it sounds to me like there are two issues:

  1.  There is a bug, somewhere, that is generating an unaligned access
  2.  HPPA can be configured to handle, somehow, unaligned accesses.

Your contention is that HPPA buildd machines should be reconfigured to
handle unaligned access.  That's fair enough, though in the past it
has been claimed that this imposes a large performance penalty [1].

Irrespective of how the buildd machines are configured, however, there
is a bug.  I think it should be tracked.

[1] http://lists.debian.org/debian-devel/2007/11/msg00871.html

Attachment: signature.asc
Description: Digital signature

Reply to: