2.0.0-pre : Failed to open socket.

[David Wood]

Hi Arran and all,

In message <467978FF.4010403 at sussex.ac.uk>, Arran Cudbard-Bell 
<A.Cudbard-Bell at sussex.ac.uk> writes
>Debashis Prusty wrote:
>> No. I have tried this. As I have mentioned earlier versions like 
>>1.1.4 & 1.1.6  are working fine. Problem is with version 2.0.0, where 
>>the listen part is not comented. Lets think of something else.
>>
>
>As I said earlier , but will say again for clarity.
>
>It *is* a bug in pre1, Alan was trying something out that broke binding
>in some BSD based operating systems and looks like Solaris too ...

Following a little bit of detective work with gdb, I realised that the 
problem is with the udpfromto code in -pre1, at least on FreeBSD. 
Passing --without-udpfromto to configure on FreeBSD means 2.0.0-pre1 
works on FreeBSD 6.2-RELEASE. Supposedly udpfromto works on BSD like 
operating systems, but I've never got it to work on FreeBSD, even in 
1.1.x.


>The code has been taken out in the CVS head... if you want to use the
>new features of 2** (of which there are many) use the CVS head not pre1.

Has the faulty code really been taken out or fixed, or is the resolution 
of this situation a side-effect of udpfromto being disabled in HEAD at 
the moment?


>In my opinion the biggest advantage of 2.*.* is the FreeRADIUS unlang
>(see man unlang) which wasn't included in pre1 anyway.

I have a working FreeBSD port for 2.0.0-pre1 on my system, but I'm loath 
to ask for it to be committed. I have to patch for bugs #452, 453 and 
454 (thanks to Nicolas for committing my enhanced patch for #454 to HEAD 
and the 1.1 branch), and pass --without-udpfromto to get the thing to 
work at all - though it is now working on my live system.

Further, pre1 has features missing compared to HEAD (not least the 
sites-available / sites-enabled stuff in raddb, which leads to quite a 
few changes in the configuration file), the PGP signature for the pre1 
.tar.gz doesn't verify and the .tar.bz2 isn't PGP signed, also 
raddb/certs/bootstrap doesn't work for me in pre1. I haven't bothered to 
try to debug raddb/certs/bootstrap yet; I have my own way of building 
the necessary certificates.

Nevertheless, if any FreeBSD users want a tarball of my 2.0.0-pre1 port, 
please email me. At the moment, it couldn't be committed to the ports 
tree because the patches are organised incorrectly, but it does work on 
my machine.


I had a relatively painless migration to a version 2 configuration - 
svn_load_dirs to update the configuration vendor branch in my private 
Subversion repository, svn copy to tag that vendor drop, svn merge the 
vendor branch changes to my live configuration, a bit of manual conflict 
resolution followed svn resolved (though less than I'd feared - 
Subversion got most things right by itself), then svn ci when I'd done.

That said, all I've done this far is port my version 1.1.6 configuration 
to version 2 configuration files; I've not taken advantage of anything 
new, nor will I do so without sites-available / sites-enabled and some 
other things that not in pre1.


Is there any hope of a 2.0.0-pre2 release any time soon? I realise that 
version 2 is still under active development, that HEAD has just gone 
through a period of being uninstallable for a while so it may be wise to 
let things settle a while longer, and that the priority is on 
development rather than release of even a "pre" version. That said, I 
feel that there may be sufficient broken or missing in 2.0.0-pre1 to 
discourage further testing now, which I understood to be the whole point 
of shipping 2.0.0-pre1.

Whilst "it's fixed in HEAD" is fine for those who hang out here, FreeBSD 
ports are not really supposed to depend on a CVS or Subversion checkout 
for their main tarball. Recent discussion on freebsd-ports has suggested 
that, at most, using a checkout should be a non-default option. As 
development on 2.x continues, the FreeBSD package list is changing, so 
that really leaves my only options at the moment as patching 2.0.0-pre1 
or creating an unofficial 2.0.0-pre2 based on a tarball I'd have to host 
myself.


On that note, whilst I know this isn't the best place to report it (what 
is - bug database?), there's a typo in HEAD. raddb/Makefile version 1.26 
has a typo in the second line of the install target - it should be 
sites-available not sites-evailable.


Best wishes to you all,




David
-- 
David Wood
david at wood2.org.uk