Chun Tian (binghe) wrote:

So, to make UNIX-FAST-SELECT wait for ever, both TIMEOUT-SECS and TIMEOUT-USECS must be NIL. I think your patch is right, that's a usocket bug.

Commited to trunk as r483, thanks very much!

Cool!

I did take a look at the source code of unix-fast-select. The thing that concerned me was,

(unless *interrupts-enabled* (note-dangerous-select))

under the T clause of the cond. Having looked at things a bit more closely now, I think I do understand better what this means.

Cheers, Chaitanya

--binghe

On 2008-12-24, at 01:13, Chaitanya Gupta wrote:

...

Hi,

On SBCL, providing :timeout NIL to wait-for-input returns immediately even when there is no input on any of the provided sockets (the documentation says it should wait indefinitely).

Attached is a patch (svn diff) which fixes this, but I am not very sure about how sb-unix:unix-fast-select works, so its be best if someone who knows better can review it before committing this patch. :)

Cheers, Chaitanya

Index: backend/sbcl.lisp

--- backend/sbcl.lisp (revision 482) +++ backend/sbcl.lisp (working copy) @@ -353,7 +353,7 @@ (1+ (reduce #'max (wait-list-%wait sockets) :key #'sb-bsd-sockets:socket-file-descriptor)) (sb-alien:addr rfds) nil nil

•            (when timeout secs) musecs)
  

•            (when timeout secs) (when timeout musecs))
     (if (null count)
     (unless (= err sb-unix:EINTR)
       (error (map-errno-error err)))
  

---

usocket-devel mailing list usocket-devel@common-lisp.net http://common-lisp.net/cgi-bin/mailman/listinfo/usocket-devel

-- Chun Tian (binghe) NetEase.com, Inc. P. R. China