I'm trying to build an app under OSF 1.3 on an Alpha. Can
anyone tell me how to get gdb or dbx to stop at the C line
causing an 'Unaligned access' error message to be printed?

In case someone as seen my problem, let me describe it.
When I use gdb to stop at the pc contained in an unaligned
access message (via 'break *0x1a2b3c4d5'), it claims the error
comes from a line where I've got a link list and I'm trying
to set a pointer to NULL; like

head->next->next = (struct_type *) NULL;

I don't see why an assignment should cause this kind of error!
Any one seen this or know how I can investigate further?

Regards,
Chuck Packard                          cpack...@mathworks.com
The Mathworks, Inc.                    508.653.1415
-------------------------------------------------------------
If we had known 10 years ago today would be 10 years from now,
would we spend tomorrows yesterdays and make it last somehow.
                      - from "No Time to Kill" by Clint Black

There might be two reasons:
        (a) either head or head->next are pointers to unaligned structures
        (b) head->next->next is a pointer to unaligned storage.
where "unaligned" means "not on 8-byte boundary" since we are talking
(8-byte) pointers here.

You can examine the instruction at which the access is unaligned (dbx:
$pc/i, gdb: x/i $pc); if it's a ldq (load quad), it's an unaligned
fetch.  A stq indicates an unaligned store (store quad).  That should
tell you which of the two cases above is the one you have.

--
John Kohl <j...@atria.com> or <jtk...@mit.edu>
working for but not representing:       Atria Software
---------
What are you pretending not to know?

|> I'm trying to build an app under OSF 1.3 on an Alpha. Can
|> anyone tell me how to get gdb or dbx to stop at the C line
|> causing an 'Unaligned access' error message to be printed?
|>

using the pc in the unaligned access message, in dbx:
(dbx) 0xyyyyyyyy/i

where yyyyyyyy is the address of the pc in the message. This will tell
you the function and line where the unaligned access occured. To stop
at that point:

        stopi at 0xyyyyyyyy

|> In case someone as seen my problem, let me describe it.
|> When I use gdb to stop at the pc contained in an unaligned
|> access message (via 'break *0x1a2b3c4d5'), it claims the error
|> comes from a line where I've got a link list and I'm trying
|> to set a pointer to NULL; like
|>
|> head->next->next = (struct_type *) NULL;
|>

is one of your nexts an int ?

|> I don't see why an assignment should cause this kind of error!
|> Any one seen this or know how I can investigate further?
|

It only does if one of the "addresses" in not on a pointer boundry.
It is quite possible that head or head->next is not on a proper
boundry but is pointing to something that is not fatal.