ia64/sqlite problems

[Matthias Andree]

David Relson <relson at osagesoftware.com> writes:

> On Sun, 3 Jul 2005 22:54:18 -0400
> Clint Adams wrote:
>
>> sqlite tests fail on the ia64 autobuilder
>> http://buildd.debian.org/fetch.php?&pkg=bogofilter&ver=0.95.2-1&arch=ia64&stamp=1120235262&file=log&as=raw
>> 
>> Apparently coredumpsize is being limited, so no backtraces.
>> 
>> 
>> On a different Itanium machine, the tests pass, but I get this (these messages
>> appear to be from the Linux kernel).
>
> Hi Clint,
>
> SourceForge has an alpha running Debian 3.0.  On it I've built
> sqlite-3.2.2 and bogofilter-0.95.2 and successfully run "make check".
> That's as far as I can go with this problem.  Likely Matthias can go
> farther.  I've also emailed sqlite's author, but haven't heard back
> from him.

I can reproduce the problem on Solaris 8 on a SPARC64, and as far as the
backtrace goes (I needed to use dbx, gdb isn't up to the task), it
happens deep in the sqlite3 code, when db_set_dbvalue calls
sql_fastpath, which in turn calls sqlite3_stmt. The same machine is fine
when 32-bit code is compiled. The backtrace itself seem unconclusive to
me. AFAICS, we aren't passing junk, but if 64-bit code requires data on
64-bit boundaries, that might explain a bit. Note I used the sqlite3
3.2.1 sparcv9 (=64 bit) code from blastwave.org, so I haven't run the
sqlite3 self-test on the 64 bit machine, and I haven't yet looked at the
sqlite3 sources.

I haven't checked on David's assertion that other architectures are also
failing.

-- 
Matthias Andree