Mike Hearn wrote:
On Mon, 2004-12-20 at 17:56 +0000, Robert Shearman wrote:
Sorry Mike, but we want to *remove* the current hacks for the RPC-runtime emulating named pipe code, not adding more.
Yes, of course, but we also want to keep the code in CVS in a roughly working situation. As it is, the code is guaranteed to crash with all InstallShields (or indeed any program that uses out of process COM). If Alexandre was to do a new tarball release tomorrow that's a lot of users who are now inconvenienced.
Currently table marshalling isn't implemented, nor is thread affinity. They're not 5-line patches either. They are on the todo list though and when they're done we can redo the threading to work more like native and remove the hacks.
IIRC the patch I wrote for thread affinity at the start of the summer was never submitted precisely because it used a hack to ensure the thread could find its wine_pipe again (a TLS slot/stack thing). I think for thread affinity to work correctly we need to at least rework how pipes and threads interact, maybe even switch it on top of the RPC runtime transports entirely so re-entrancy works correctly.
The patch really isn't as big as it looks, there are some formatting changes in there and some renaming+commenting.
thanks -mike
I just want to say that I tried out this patch. It does not fix the problem for InstallShield 6. I have a bug report in bugzilla if you are interested. Right now AFAIKT InstallSheild is seriously broken (for quite some time) and I am unable to use Native ole to workaround the problem. If there is anyway I can help this process I please tell me.
--
Tony Lambregts.