[Mono-dev] Fundamental performance problems with Mono

[James Mansion]

Robert Jordan wrote:
> So you're actually blaming a wrapper whose solely intention was
> to gain .NET 2.0 compatibility.
>   
To me, that looks like a very good target for blame - a lame 
implementation done in
a hurry to tick a box.
> Patches welcome :)
>   
I'd personally be much more comfortable about 'Design proposal welcome', 
and even
that assumes we all share a vision of the objective and constraints.

In this particular case it worth considering how the cross-platform 
differences in
event indications or completions will be handled (and that should 
ideally include
IOCP on Windows rather than an attempt to use Windows fake select). So if
for example you hook into the libevent 2.0 development (libev doesn't do 
IOCP)
then you'd have to decide whether to depend on it or grab a snapshot, 
and then
whether to code to it using P/Invoke or implement the lot in C.

I think its worth doing right and suggesting to go from zero to code is 
irresponsible.

Not that I have time at the moment (or indeed any inclination to code in 
C rather
than C++ - I suspect boost::asio would probably handle the API quite well).

James