[wx-dev] Fix for Reentrance problems in events (Socketsproblem, even more?) (Was: Status of the SOC2006_SOCKET branch)

[Albert van Peppen]

> I'll take a look at them myself, too, for the SOC branch.

I think it should be rather straight forward.


> I'd love to get some sample for this behaviour, as server-client
connection, both TCP and UDP,
> with or without proxies, are working on my side, both base and
event-driven, at least TCP and
> UDP are working perfectly, and I didn't really touch IPC much. I'll
check IPC anyway and unix
> sockets. But if you can actually tell me what's wrong instead of
"problems in the send/receive
> protocol handling or something" it will be extremely helpful. A simple
code sample showing that
> error would be the best.

What I basically did was compile the library; which was bit of a pain on
VC.NET 2003/2005.
And then compile the client/server examples in both DDE and IPC mode.
The samples where unable to work. When compiling the samples against the
2_6_Branch / Head branch they run ok.
My application as well; when compiling against SOC it doesn't work. As
far as I could see the requests are ok but the reply on the request is
invalid.
But as said I didn't got too deep into it due to the fact that I had a
problem and just 2 days to fix it.


> The 10 minutes timeout issue has been discussed here more than once,
or in bugtracker at least,
> and never got to a definite answer, other than everyone saying "oh god
that's too much".
> Changing it is something in my TODO for sure.

It is fixed in the patch I supplied; it's a solution which you like or
not but it is a solid solution and should be easy to understand ;)


> I'll check the patch as I said above, about the reentrant threading
problems. That's a very
> sensitive issue, and every time I seen issues in that matter were
caused by user code more
> than the actual library, but it'll be nice to discover there's
something fishy actually going on there.

Please feel free to check what I did and what the implications are.
This patch should make things better; I saw various problems which are
related to this problem. (Bug-1574240 should be fixed as well with the
patch, but only if all eventhandlers are set to be non-reentrant)

I have set the default eventhandler behaviour to allow reentrance, but I
think this is wrong and the default should be to disallow reentrance in
the eventhandler, unless stated otherwise. But I would like to know who
is in favour of this / against this and why.


> And for everyone else: I'm fully moved to the new flat, have internet,
and rested
> a little from all the work needed to actually move everything. So I'll
get back to
> the branch and commit my latest work on base/gui implementations, to
move on diferent
> build systems than I used (VS comes to mind), and get feedback from
users to see issues
> I don't see myself on my side (like the one discussed above).
>
> In short, I'm back.

Congrat and welcome back ;)


Albert