[Sbcl-devel] [slime-devel] Threading warnings on SBCL 1.0.19 Darwin x86

[Scott B. <se...@gm...>]

Apologies folks for the cross-threading, this is related to the issue
many have experienced running SLIME on a threaded SBCL under Darwin
x86.

On 29-Aug-08, at 3:22 AM, Nikodemus Siivola wrote:
No, sorry for my bogus knowledge. Last time I checked, which was
probably during 10.4, the only timed wait I found was the IIRC
SemaphoreWait or similar which did not like SBCL at all. (Mysterious
messages to terminal is not usually a good sign...)

Support for a timeout value appears to be there for futexes, and I
think all of the lutex platforms support pthread_cond_timedwait. It
should be fairly simple to support a :TIMEOUT optional parameter in
CONDITION-WAIT.

To address the issue, are there any major concerns with the following
approach?

Index: src/code/target-thread.lisp
===================================================================
RCS file: /cvsroot/sbcl/sbcl/src/code/target-thread.lisp,v
retrieving revision 1.93
diff -r1.93 target-thread.lisp
393,398c393,399
<     (progn
<       ;; FIXME: This doesn't look interrupt safe!
<       (setf (mutex-%owner mutex) nil)
<       (with-lutex-address (queue-lutex-address (waitqueue-lutex queue))
<         (with-lutex-address (mutex-lutex-address (mutex-lutex mutex))
<           (%lutex-wait queue-lutex-address mutex-lutex-address)))
---
>     (unwind-protect
> 	 (progn
> 	   (setf (mutex-%owner mutex) nil)
> 	   (with-lutex-address (queue-lutex-address (waitqueue-lutex queue))
> 	     (with-lutex-address (mutex-lutex-address (mutex-lutex mutex))
> 	       (%lutex-wait queue-lutex-address mutex-lutex-address))))
>       ;; restore the owner of the mutex if we are interrupted

Regards,

- Scott Bell