Fixed a subtle bug in the macro for the << redirection: The port

opened by open-input-string could be closed by the GC before the exec().
Now << stores the port in a global variable.
The test to reveal the bug was:

(let lp ()
  (run (head) (<< "a"))
  (lp))

printing a "closed channel" message after some a few hundred interations.

scsh/syntax-helpers.scm

@@ -167,6 +167,10 @@
 	 (%q (lambda (x) (list (rename 'quote) x)))
 	 (%close (rename 'close))
 	 (%move->fdes (rename 'move->fdes))
+	 (%set! (rename 'set!))
+	 (%<<-port-holder (rename '<<-port-holder))
+	 (%let (rename 'let))
+	 (%port (rename 'port))
 	 (%stdports->stdio (rename 'stdports->stdio)))
     (cond ((pair? redir)
 	   (let ((args (cdr redir)))
@@ -184,9 +188,13 @@
 ;		(receive (fdes epf) (parse-spec args 0)
 ;		  `(,%dup-port (,%run/port . ,epf) ,fdes))) ; Add a WITH-PORT.
 
+	       ;; We save the port in the global variable <<-port-holder to prevent a
+	       ;; GC from closing the port before the exec().
 	       ((<<)
 		(receive (fdes exp) (parse-spec args 0)
-		  `(,%move->fdes (,%open-string-source ,exp) ,fdes)))
+		  `(,%let ((,%port (,%open-string-source ,exp)))
+		     (,%set! ,%<<-port-holder ,%port)
+		     (,%move->fdes ,%port ,fdes))))
 
 	       ((>>)
 		(receive (fdes fname) (parse-spec args 1)

scsh/syntax.scm

@@ -10,6 +10,10 @@
 ;;; The three basic forms for running an extended process form:
 ;;; EXEC-EPF, &, and RUN. EXEC-EPF is the foundation.
 
+;; This is used by the macro for the << redirection to prevent the temporary 
+;; port from being closed by a GC before the process exec's
+(define <<-port-holder)
+
 (define-syntax exec-epf
   (lambda (form rename compare)
     (transcribe-extended-process-form (cdr form) rename compare)))