Hi,

At Fri, 31 Jan 2003 07:09:32 +0900,
Joel VanderWerf wrote:
> >>>does anyone know how to specify SA_RESTART in the signal mask such that
> >>>interupted system calls are automatically restarted on solaris, as they are on
> >>>linux?
> >>
> >>If all of sigprocmask, sigaction and SA_RESTART are available,
> >>ruby always sets it.
> > if that is the case, then shouldn't it be impossible for system
> > calls to fail
> > with EINTR? (see thread regarding solaris/flock for history)
> 
> I'm wondering too. I see this not only in Ara's test code, but in my
> own, with heavy demand by threads and processes doing shared reads and
> exclusive writes. After getting all(?) my concurrency bugs out, I
> still had to put in a few "rescue Errno::EINTR; retry" statements, or
> else there would be some (very infrequent) failures. Interestingly, it
> only seems to happen during the shared reads, so it looks like
> something one thread is doing causes a system call in another thread
> to be interrupted.

In Solaris, it apparently needs to handle EINTR at every I/O
operations.  Also fseek()?


Index: file.c
===================================================================
RCS file: /cvs/ruby/src/ruby/file.c,v
retrieving revision 1.130
diff -u -2 -p -r1.130 file.c
--- file.c	29 Jan 2003 23:28:47 -0000	1.130
+++ file.c	30 Jan 2003 22:21:47 -0000
@@ -1990,4 +1990,5 @@ rb_file_flock(obj, operation)
 	fflush(GetWriteFile(fptr));
     }
+  retry:
     TRAP_BEG;
     ret = flock(fileno(fptr->f), NUM2INT(operation));
@@ -2001,4 +2002,9 @@ rb_file_flock(obj, operation)
 #endif
               return Qfalse;
+	  case EINTR:
+#if defined(ERESTART)
+	  case ERESTART:
+#endif
+	    goto retry;
         }
 	rb_sys_fail(fptr->path);


-- 
Nobu Nakada