RE: memory alignment errors (ITS#2760)

[richton@nbcs.rutgers.edu]

Thanks; make test appears much better now. I tried a new compile using
'long' (int is 32 bits on Solaris LP64; long is 64.) Every time I run with
that change, things are good until test 17 (syncreplication-refresh),
which always fails with the message:

"test failed - master and slave databases differ"

This is running against HEAD, so I realize I may be looking at a work in
progress. Is that test currently expected to pass?

I noticed Kurt's follow-up about using a union. If somebody thinks there's
a better fix, I'd be willing to try it.