Re[2]: GREAT socket speedup on win32

[Arseny S. <am...@ic...>]

Just realised that I didn't CC this message to the list.

Hi Joerg.

> Thanks  you very much for this one! About one year ago, I discovered
> in  the  win32.d  source that indeed, CLISP used to start one thread
> per  I/O  operation.
I remember your messages of that time.

> I've not yet reviewed or tested it, but I'm wondering why you say it
> only  speeds  up  sockets.  IIRC,  all I/O was affected?
You're right! I missed one use of DoInterruptible. The patch was
initiated by idea, that came to my mind in another project, not the
dead-end with some clisp project.
I tested the patch with sample web-server by Devon Sean McCullough,
that he was posted to clisp-list last year, with MSVC and mingw
clisps and speed-up factor was about 8. Cygwin (not mingw) version
performs faster than WIN32_NATIVE one anyway. Just tested the patch
with ordinary I/O of text files. Buffered I/O is affected not much,
speed-up of test program was about 10%, but times with :buffered nil
differ greatly (26 times).

-- 
Best regards,
 Arseny