Jason Tishler wrote:
>Andrew,
>
>On Fri, Oct 10, 2003 at 12:25:01AM -0400, Andrew Dunstan wrote:
>
>
>>WinXP/cygwin/gcc version 3.3.1 (cygming special)
>>
>>
>
>XP Home or Pro? What version of Cygwin?
>
XP-HE. I ran a cygwin upgrade last night (to pick up the latest bison
and get cygutils - it picked up gcc along the way). Then I installed the
ipc-daemon2 as a service and started it, and ran make check.
>
>
>
>>gives these
>>
>>tablecmds.c:3528: warning: dereferencing type-punned pointer will break
>>strict-aliasing rules
>>[snip]
>>
>>
>
>FWIW, I saw warnings I never saw before when compiling Python under
>Cygwin gcc 3.3.1 yesterday. I presume gcc 3.3.1 is pickier or more
>"chatty" than it was before. Anyway, Python still passed its full
>regression test.
>
>
Right. It's been sen on other platforms with the latest gcc too, I
believe. As I understand the latest versions of gcc turn this on with
-O2, so perhaps we need to add -fno-strict-aliasing as a short term
measure if we are using gcc. But it should be fixed - IIRC the Linux
kernel folks went through some agony over this and eventually made
almost the whole kernel type-pun safe. After all, this enables better
optimisation.
>
>
>>make check fails (hangs) consistently on parallel tests
>>
>>
>
>Are you getting hangs or connection refused errors. The Cygwin
>PostgreSQL README documents the following issue:
>
>
hangs - I have to kill the psql process to continue.
> 1. make check can generate spurious regression test failures due to
> overflowing the the listen() backlog queue which generates
> connection refused errors. Note that make installcheck does not
> have this problem since it runs all tests sequentially instead of in
> large concurrent groups.
>
>
>
>>max_connections set to 100, shared_buffers set to 1000 by initdb (CVS
>>version, not mine).
>>
>>
>
>The above will not have any affects under Cygwin. IIRC, the listen()
>backlog is 200 under server versions of Windows (e.g., 2000 Advanced
>Server) and 5 on all other versions (e.g., XP Pro).
>
>Can you try make installcheck instead?
>
>
will do tonight after work.
cheers
andrew