Re: [Sbcl-devel] A bug in SBCL 1.0.20?

[Nikodemus S. <nik...@ra...>]

On Sun, Oct 5, 2008 at 6:07 PM, Sölvi Páll Ásgeirsson <so...@gm...> wrote:
> Hi everybody,
>
> Every once in a while(every 1-3 days?) I get the following error popping up
> in SLIME.
> I'm running SBCL 1.0.20 on Fedora 9, x86_64,
> Linux 2.6.26.3-29.fc9.x86_64
> This doesn't seem to be tied to any usage pattern.
> This is the full backtrace, please let me know if I can give you any more
> information.

The proximate cause is a bug in HANDLER-DESCRIPTORS-ERROR, which
signals an error even if there were no bad file descriptors -- as is
the case here.

The reason HANDLER-DESCRIPTROS-ERROR is called in the first place is
the call to select() returning with an error other than EINTR.

SBCL 1.0.21.8 deals with both issues: HANDLER-DESCRIPTORS-ERROR
doesn't signal the error if there were no bogus descriptors, gets
called only in case of EBADF, and the user is informed about other
unexpected errors from select().

If you build that version from the CVS and run with it, then in 1-3
days you should get an error saying

 "Unix system call select() failed: ..."

with the actual problem description where the dots are. Please let us
know what the issue is when that happens. (Selecting the CONTINUE
restart at that point should be harmless, but depending on what is
going on may just lead to the same error being immediately signaled
again.)

Cheers,

 -- Nikodemus