axis-c-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Kenneth Chiu <ch...@cs.indiana.edu>
Subject Re: Supporting stub classes functionality to C stubs
Date Mon, 14 Jun 2004 16:41:11 GMT
On Mon, 14 Jun 2004, Susantha Kumara wrote:
> On Mon, 2004-06-14 at 15:28, Samisa Abeysinghe wrote:
> > Working with axis transport (client side) I experienced the same problem. It was
a hell of a
> > trouble to add something new. (e.g. I added connection timeout and proxy and I had
to modify
> > around 6-7 classes to  add the feature.
>
> Now the transports are separate modules and can be written/developped
> without affecting the other parts of the code.
>
> Personally I too find it very difficult to improve the AxisTransport
> module. I would rather like to use libwww kinda library or re-design the
> existing AxisTransport and re-use some of its classes/code.
>
> >
> > I am trying to understand why exactly we should provide C support on top of C++.
(If it was other
> > way around it would have been more understandable)
> > However, having a good OO design is key as this is a C++ project.
>
> Earlier the idea was to go towards C because of performance reasons. We
> were not supposed to use any C++ features that leads to any performance
> degradation. Ex RTTI, Exceptions, Multiple inheritance.

In my opinion, any kind of blanket prohibition against a
feature for performance reasons is not ideal.  For example,
one dynamic_cast per character is definitely going to kill
your performance.  But one dynamic_cast per service
invocation is going to be down in the noise.

These kinds of considerations exist regardless of what you
allow or disallow.  Injudicious use of almost anything can
impact performance.  For example, excessive use of malloc()
can impact performance.

Mime
View raw message