On Mon, Oct 24, 2005 at 12:37:37PM +1000, Ian Wienand said > Hi, > > I just built 1.4.2-2 on my IA64 machine and it does build OK. > > The relevant part where the buildd failed looks like > > ======= TESTING: id-tagging-defaults ======= > Test 1: id-tagging-defaults > Test 1: PASS > baz(30725): unaligned access to 0x60000fffffb7191e, > ip=0x40000000001095f0 > baz(30725): unaligned access to 0x60000fffffb71922, > ip=0x40000000001095f0 > baz(30725): unaligned access to 0x60000fffffb71926, > ip=0x40000000001095f0 > baz(30725): unaligned access to 0x60000fffffb7192a, > ip=0x40000000001095f0 > > Having a look > > [EMAIL PROTECTED]:/usr/lib/debug/usr/bin$ addr2line - e ./baz > 40000000001095f0 > /tmp/bazaar-1.4.2/src/hackerlab/hash/sha1.c:370 > > which dereferences the buffer in a sha1_context struct. > > I've seen problems with stack alignment and gcc 4 on IA64 and I'd > suggest this is similar. > > http://gcc.gnu.org/bugzilla/show_bug.cgi?id=22605 > > The two possible solutions I can see is that you either change > sha1_process_blocks to not assume the input is aligned or to change > sha1_process_blocks so it is defined as only taking aligned input. > Either is messy unfortunatley.
Hm...I think the fixes for this are in bazaar 1.5 already; if I try to backport them, would you be able to test build baz on itanium for me? > caballero is an Itanium 1 right? That has stricter alignment > requirements than Itanium 2. I'm not sure if this could cause the > problem on the other architectures too. It seems caballero has a kernel option enabled that makes unalighed accesses fatal. It is an Itanium 2, though. -rob -- I bet your lymph nodes are as big as cats! -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
