On Fri, 2013-12-20 at 10:54 -0300, Alvaro Herrera wrote:
> I don't see how can the pg_upgrade check fail in this way but not the
> regular regression test. This patch includes the following hunk to
> pg_config.h.in:
>
> +/* Enable large inode numbers on Mac OS X 10.5. */
> +#ifndef _DARWIN_USE_64_BIT_INODE
> +# define _DARWIN_USE_64_BIT_INODE 1
> +#endif
>
>
> Evidently something is not going well in ReadRecord. It should have
> reported the read failure, but didn't. That seems a separate bug that
> needs fixed.
This is enabling large-file support on OS X, so that seems kind of
important. It's not failing with newer versions of OS X, so that leaves
the following possibilities, I think:
- Large files never worked on 10.5. That would be strange because
Autoconf explicitly refers to that version.
- New OS X versions have this on by default (could someone check?), so
it already worked before, and it's something specific to 10.5 that's
broken.
- It's something specific to PowerPC or endianness.
- That build farm member has a strange setup.
If anyone has any of these components, please help isolate the problem.
It's conceivable that inconsistent include file order would cause this
kind of problem, but we're pretty strict about including the right
things first.
It someone can reproduce the problem, it would also help dissecting the
pg_upgrade test script to see exactly where the failure is introduced.