apr-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From M Joonas Pihlaja <jpihl...@cc.helsinki.fi>
Subject Re: [VOTE] 0.9.8 releases
Date Mon, 30 Jan 2006 08:39:11 GMT


On Sun, 29 Jan 2006, Garrett Rooney wrote:

> If you could check and see if those tests failed in 0.9.7 that would
> be interesting...

In regards to Solaris 9 and 0.9.8/0.9.7:

The exact same errors appear when testing 0.9.7 on Solaris 9.
There are only two real failures, I believe.  The first one is in
testdir.c:test_rmkdir_nocwd(), which tries to rmdir(2) a
directory that it just changed to.  Solaris doesn't like that and
returns EINVAL.  The test case is then aborted, leaving the the
remaining test cases are likely confused.  I didn't check them,
as removing the failing test case and rerunning makes most of the
other failures go away.

The second failure is in testsockets.c:sendto_receivefrom().
For some reason bind() is returning EADDRNOTAVAIL when
sendto_receivefrom() calls apr_socket_bind(sock, to).  This is a
datagram socket that is supposed to bind to "::1" port 7772.
Below is the struct sockaddr's sin6 field before the address is
passed to bind():

        sin6 = {
            sin6_family   = 26U
            sin6_port     = 7772U
            sin6_flowinfo = 0
            sin6_addr     = {
                _S6_un = {
                    _S6_u8     = ""
                    _S6_u32    = (0, 0, 0, 1)
                    __S6_align = 0
            sin6_scope_id = 0
            __sin6_src_id = 0

I don't really understand why it's not binding.  I notice there
are no other UDP sockets using IPv6 on this machine; only TCP
sockets.  That's a bit odd, as there's a big bunch of bound IPv4
UDP sockets and usually on other OS's both v6 and v4 sockets get
bound at the same time.  (But I'm a Solaris newbie, so this might
just be one of those things, I guess.)



View raw message