Re: [Pan-users] Win32 build of 0.138 up

[Steve Davies]

On 18 June 2012 21:26, Duncan <address@hidden> wrote:
> Steve Davies posted on Mon, 18 Jun 2012 18:39:07 +0100 as excerpted:
>
>> I am closing in on part of the issue. It MIGHT be that it does not like
>> using a non-standard port number in servers.xml. My news server provides
>> TLS secured NNTP on port 443 for convenience. If I enable that, I can be
>> sure it will start crashing on me next time I restart.
>
> There was a thread with a similar problem, but on Linux (Mint 13 Maya,
> went back to Debian Sid/Unstable with 0.137-1 and it worked).  See the
> "New 0.138 version on Pan ppa" thread from Friday/Saturday (Sat, 16 Jun
> 2012 00:28:24 +0000 (UTC)), by Bob.
>
> There it seemed to be some sort of issue between pan and gnutls and
> disabling tls but still trying to use the tls port and crashing when it
> got an encrypted channel when it was either expecting plain-text
> (disabled ssl/tls) or trying to enable tls but crashing due to version
> mis-match.
>
> The details weren't fully worked out, but for now he seems to be fine
> using either 0.137 with ssl, or 0.138 with gnutls/ssl disabled at build-
> time and an appropriate plain-text port chosen.
>
> So it seems pan still has some instability with ssl/tls.  I'd guess
> (warning non-coder trying to use coder language!) that pan needs an
> assert or some such, to either popup a warning or at least fail
> gracefully if some condition it's expecting isn't matched, that without,
> it's simply crashing.
>
> As I said in that thread, pan's ssl/tls support is still experimental
> enough and changing enough between versions that people who don't
> actually need it and prefer "just working" to being testers, probably
> want to continue to run with an unencrypted-only connection configured.
>
> But obviously, being the de-facto Windows build guy, you'll need to work
> thru the problem eventually, and where's the fun in playing it safe,
> anyway. =:^)


For a non-coder, you did okay Duncan :)  Now I have GNU/TLS support in
Windows, I have no intention of disabling it. "stunnel" works great,
but is just a pain.

I think it may be some thread-locks that are needed, or perhaps some
simple reference counting so that you do not destroy a socket while it
is still being used by another thread. This is always a horrible issue
in multi-threaded apps.

In the meantime, I am attaching a patch which improves things a lot for me...

Hope it helps others too.

Cheers,
Steve

pan.diff
Description: Binary data