Re: Feasibility of using Ada in new development

["Randy Brukardt" <randy@rrsoftware.com>]

"Wojtek Narczynski" <wojtek@power.com.pl> wrote in message
news:5ad0dd8a.0409080110.1be46d74@posting.google.com...
> Randy,
>
> > This is how tasking works in the Janus/Ada compiler. So, clearly there
is
> > something comparable available. But I wouldn't go so far as to call it
high
> > performance (it was designed to make tasking available on as many
targets as
> > possible).
>
> You mean you can put a task to sleep untill an IO descriptor is ready
> for reading or writing?

I don't know if you can even do that on Windows. I know that the API exists,
but it was seriously broken the last time I looked. Perhaps its been fixed.

In any case, the answer is no, there is no I/O support in our tasking (its
the big flaw, which I why I said it wasn't high performance). But it doesn't
use the OS threads. It would actually be fairly easy to build a special I/O
library that could put a task to sleep (no threads here - only Ada :-). Just
replace the implementation of Read and Write. Of course that wouldn't work
for sockets, because sockets aren't I/O in Windows. The Claw sockets library
supports various forms of socket waits, but experience has shown that (at
least on Windows 2000), there is no performance to be gained there. Every
one we tried was worse than the basic busy-wait a bunch of worker tasks
implementation. (This was in the context of Ada Server, the web server that
runs Adaic.com and the search engine for AdaIC, Ada-Auth.Org, and the ARM.)

                                 Randy.