[slime-devel] SBCL's with-timeout failed with slime-input-stream
Hi, I'm using SLIME (2010-11-07) with SBCL(1.0.44.22) on Linux x8664. The following code should signal an sb-ext:timeout condition but with-timeout isn't functioning on slime-input-stream with-timeout works with SBCL started from terminal. (sb-ext:with-timeout 1 (read *standard-input*)) CL-USER> *standard-input* #<SWANK-BACKEND::SLIME-INPUT-STREAM {1002C88D31}> Cheers, -- Jianshi Huang
(sb-ext:with-timeout 1 (read *standard-input*))
with-deadline also doesn't work from inside slime, but generally you should prefer that to with-timeout in production systems. (with-timeout is not safe due to async unwinds at random points in the code (unless the entire codebase is prepared for that)). hth, -- attila
Hi Attila, On Wed, Dec 1, 2010 at 8:09 PM, Attila Lendvai <attila.lendvai@gmail.com> wrote:
(sb-ext:with-timeout 1 (read *standard-input*))
with-deadline also doesn't work from inside slime, but generally you should prefer that to with-timeout in production systems.
Thanks for the advice. Being able to restrict the signal in specific code block is very nice.
(with-timeout is not safe due to async unwinds at random points in the code (unless the entire codebase is prepared for that)).
That makes sense, and I suppose using without-interrupts is unacceptable in the body part. Since the part I want timeouts are all IO function calls, I think it should be safe. But I think I'd better use with-deadline. :) -- Jianshi
* Jianshi Huang [2010-12-01 09:42] writes:
Hi,
I'm using SLIME (2010-11-07) with SBCL(1.0.44.22) on Linux x8664.
The following code should signal an sb-ext:timeout condition but with-timeout isn't functioning on slime-input-stream
with-timeout works with SBCL started from terminal.
(sb-ext:with-timeout 1 (read *standard-input*))
SLIME also uses with-timeout internally. Essentially we have a situation like (sb-ext:with-timeout 1 (loop (sb-ext:with-timeout 0.2 (handler-case (sleep 10000) (sb-ext:timeout () (print 'timeout) (force-output)))))) The handler doesn't know that the condition was raised by some outer with-timeout. That's probably just one of many problems of the with-timeout construct and we would like to get rid of it and replace it with a timeout argument to sb-thread:condition-wait. That's not a new issue and adding the timeout argument doesn't sound like rocket science but it doesn't seem to be a priority for the SBCL maintainers. Maybe somebody could hire Nikodemus to finally implement it. Helmut
Hi Helmut, On Thu, Dec 2, 2010 at 12:30 AM, Helmut Eller <heller@common-lisp.net> wrote:
SLIME also uses with-timeout internally. Essentially we have a situation like
(sb-ext:with-timeout 1 (loop (sb-ext:with-timeout 0.2 (handler-case (sleep 10000) (sb-ext:timeout () (print 'timeout) (force-output))))))
The handler doesn't know that the condition was raised by some outer with-timeout.
I see. But why (handler-case (sb-ext:with-timeout 1 (read *standard-input*)) (sb-ext:timeout (c) c)) or (handler-case (sb-sys:with-deadline (:seconds 1) (read *standard-input*)) (sb-sys:deadline-timeout (c) c)) also doesn't work. Interestingly, a #<SB-SYS:DEADLINE-TIMEOUT> is returned in the second case after I slime-interrupted it. I think I'm still confused. -- Jianshi
participants (3)
-
Attila Lendvai -
Helmut Eller -
Jianshi Huang