Fix Ctrl+D after ksh receives SIGWINCH (#208)

src/cmd/ksh93/edit/edit.c: ed_read():
- The loop that handles SIGWINCH assumes sfpkrd will return and
  set errno to EINTR if ksh is sent SIGWINCH. This only occurs
  when select(2) is used to wait for input, so tell sfpkrd to
  use select if possible. This is only done if the last argument
  given to sfpkrd is '2', which should avoid regressions.

src/lib/libast/sfio/sfpkrd.c: sfpkrd():
- Always use select if the last argument is 2. This allows
  sfpkrd() to intercept SIGWINCH when necessary.

Fixes: https://github.com/ksh93/ksh/issues/202

src/lib/libast/sfio/sfpkrd.c

@@ -24,11 +24,12 @@
 /*
  * The preferred method is POSIX recv(2) with MSG_PEEK, which is detected as 'socket_peek'.
  * On Solaris/Illumos (__sun), _stream_peek and _lib_select are needed, as _socket_peek doesn't work correctly.
- * On at least macOS and Linux, sfpkrd() runs significantly faster if we disable these.
+ * On at least macOS and Linux, sfpkrd() runs significantly faster if we disable these. However,
+ * ed_read() still needs to use select to intercept SIGWINCH, so if the last argument given
+ # to sfpkrd is '2' select is always used when available.
  */
 #if _socket_peek && !__sun
 #undef _stream_peek
-#undef _lib_select
 #endif
 
 #if __APPLE__ && !_socket_peek
@@ -55,6 +56,7 @@ long	tm;	/* time-out */
 int	action;	/* >0: peeking, if rc>=0, get action records,
 		   <0: no peeking, if rc>=0, get -action records,
 		   =0: no peeking, if rc>=0, must get a single record
+		   =2: same as >0, but always use select(2)
 		*/
 #endif
 {
@@ -118,7 +120,11 @@ int	action;	/* >0: peeking, if rc>=0, get action records,
 			(t&SOCKET_PEEK) )
 		{	r = -2;
 #if _lib_select
-			if(r == -2)
+			if(r == -2
+#if !__sun /* select(2) is always used on Solaris or if action == 2 on other OSes */
+				&& action == 2
+#endif
+				)
 			{	fd_set		rd;
 				struct timeval	tmb, *tmp;
 				FD_ZERO(&rd);