Greetings Geoff,
1. It works fine with --disable-bigfile and I'd be inclined to leave=20
it at that for 3.2.0b5. (If people have indexes over 4GB, the I say=20
eliminating redundancy from the database structure is a higher=20
priority than enabling big file support...)
2. The problem is that the arguments to pwrite in db/os_rw.cc (of=20
type 'off_t') seem to be treated as 32 bit instead of 64 bit. This=20
causes the page size to be treated as 0, and all pages to be written=20
over page 0. (This problem would not manifest itself on a=20
little-endian architecture.)
3. The configure script assumes a lot of variable are the same for=20
both the C and C++ compilers, which is not the same when using cc =20
and g++. That may or may not be the problem here.
4. Again, this may not be relevant, but the SunOS configurations I've=20
tested
#define open open64
and call a separate set of libraries to support big files. This=20
interferes with C++, and is #undef'd in stream.h.
Good luck with the grant review :)
Lachlan
On Sun, 1 Jun 2003 13:42, Geoff Hutchison wrote:
> > The problem with make check on SunOS with native cc is the size
> > of off_t, the size of an offset in a file. This seems to be
> > related to the --enable-bigfile configure option.
>
> This is new.
>
> I'm trying to hunt for the exact message in the developer list, but
> can't find it. What is the exact problem with native Sun CC? Many
> type problems can be fixed with configure results. (e.g. the
> various hacks in Connection.cc)
|
