rewrite funcs.c:c_test(), i.e. test(1) and [(1), to follow POSIX and XSI

in the cases where they are defined unambiguously; bug reported by
Jilles Tjoelker in <20111129232526.GC14357@stack.nl> due to a report
by Stefano Lattarini on bug-autoconf

in the ambiguous case, I stick to traditional pdksh behaviour, which means
	test ! a = b	vs.	test a = b
and
	test ! a -o b	vs.	test a -o b
behave different from each other (in the second case, the NOT operator
binds strong; POSIX demands a reduction to 3 arguments and negating
that result in the first case), so we're at two known not-ok in the
FreeBSD® testsuite. (81 and 82 in regress.sh,v 1.3)

check.t

@@ -1,4 +1,4 @@
-# $MirOS: src/bin/mksh/check.t,v 1.496 2011/11/26 00:45:18 tg Exp $
+# $MirOS: src/bin/mksh/check.t,v 1.497 2011/11/30 21:34:10 tg Exp $
 # $OpenBSD: bksl-nl.t,v 1.2 2001/01/28 23:04:56 niklas Exp $
 # $OpenBSD: history.t,v 1.5 2001/01/28 23:04:56 niklas Exp $
 # $OpenBSD: read.t,v 1.3 2003/03/10 03:48:16 david Exp $
@@ -23,9 +23,12 @@
 #-
 # You may also want to test IFS with the script at
 # http://www.research.att.com/~gsf/public/ifs.sh
+#
+# More testsuites at:
+# http://www.freebsd.org/cgi/cvsweb.cgi/src/tools/regression/bin/test/regress.sh?rev=HEAD
 
 expected-stdout:
-	@(#)MIRBSD KSH R40 2011/11/25
+	@(#)MIRBSD KSH R40 2011/11/30
 description:
 	Check version of shell.
 stdin:

funcs.c

@@ -38,7 +38,7 @@
 #endif
 #endif
 
-__RCSID("$MirOS: src/bin/mksh/funcs.c,v 1.199 2011/11/19 17:42:24 tg Exp $");
+__RCSID("$MirOS: src/bin/mksh/funcs.c,v 1.200 2011/11/30 21:34:12 tg Exp $");
 
 #if HAVE_KILLPG
 /*
@@ -2724,8 +2724,10 @@ c_mknod(const char **wp)
 int
 c_test(const char **wp)
 {
-	int argc, res;
+	int argc, rv, invert = 0;
 	Test_env te;
+	Test_op op;
+	const char *lhs, **swp;
 
 	te.flags = 0;
 	te.isa = ptest_isa;
@@ -2747,63 +2749,96 @@ c_test(const char **wp)
 	te.wp_end = wp + argc;
 
 	/*
-	 * Handle the special cases from POSIX.2, section 4.62.4.
+	 * Attempt to conform to POSIX special cases. This is pretty
-	 * Implementation of all the rules isn't necessary since
+	 * dumb code straight-forward from the 2008 spec, but unless
-	 * our parser does the right thing for the omitted steps.
+	 * the old pdksh code doesn't live from so many assumptions.
 	 */
-	if (argc <= 5) {
+	switch (argc - 1) {
-		const char **owp = wp, **owpend = te.wp_end;
+	case 0:
-		int invert = 0;
+		return (1);
-		Test_op op;
+	case 1:
-		const char *opnd1, *opnd2;
+ ptest_one:
-
+		op = TO_STNZE;
-		if (argc >= 2 && ((*te.isa)(&te, TM_OPAREN))) {
+		goto ptest_unary;
-			te.pos.wp = te.wp_end - 1;
+	case 2:
-			if ((*te.isa)(&te, TM_CPAREN)) {
+ ptest_two:
-				argc -= 2;
-				te.wp_end--;
-				te.pos.wp = owp + 2;
-			} else {
-				te.pos.wp = owp + 1;
-				te.wp_end = owpend;
-			}
-		}
-
-		while (--argc >= 0) {
-			if ((*te.isa)(&te, TM_END))
-				return (!0);
-			if (argc == 3) {
-				opnd1 = (*te.getopnd)(&te, TO_NONOP, 1);
-				if ((op = (*te.isa)(&te, TM_BINOP))) {
-					opnd2 = (*te.getopnd)(&te, op, 1);
-					res = (*te.eval)(&te, op, opnd1,
-					    opnd2, 1);
-					if (te.flags & TEF_ERROR)
-						return (T_ERR_EXIT);
-					if (invert & 1)
-						res = !res;
-					return (!res);
-				}
-				/* back up to opnd1 */
-				te.pos.wp--;
-			}
-			if (argc == 1) {
-				opnd1 = (*te.getopnd)(&te, TO_NONOP, 1);
-				res = (*te.eval)(&te, TO_STNZE, opnd1,
-				    NULL, 1);
-				if (invert & 1)
-					res = !res;
-				return (!res);
-			}
 		if ((*te.isa)(&te, TM_NOT)) {
-				invert++;
+			++invert;
-			} else
+			goto ptest_one;
+		}
+		if ((op = (*te.isa)(&te, TM_UNOP))) {
+ ptest_unary:
+			rv = (*te.eval)(&te, op,
+			    (*te.getopnd)(&te, op, true), NULL, true);
+ ptest_out:
+			return ((invert & 1) ? rv : !rv);
+		}
+		/* let the parser deal with anything else */
+		break;
+	case 3:
+ ptest_three:
+		swp = te.pos.wp;
+		/* skip lhs, without evaluation */
+		(*te.getopnd)(&te, TO_NONOP, false);
+		if ((op = (*te.isa)(&te, TM_BINOP))) {
+			const char *rhs;
+
+			/* read rhs, with evaluation */
+			rhs = (*te.getopnd)(&te, op, true);
+			/* back up to lhs */
+			te.pos.wp = swp;
+			/* re-read lhs, with evaluation */
+			lhs = (*te.getopnd)(&te, op, true);
+			/* finally run the test of lhs op rhs */
+			rv = (*te.eval)(&te, op, lhs, rhs, true);
+			goto ptest_out;
+		}
+		/* back up to lhs */
+		te.pos.wp = swp;
+		if ((*te.isa)(&te, TM_NOT)) {
+			++invert;
+			goto ptest_two;
+		}
+		if ((*te.isa)(&te, TM_OPAREN)) {
+			swp = te.pos.wp;
+			/* skip operand, without evaluation */
+			(*te.getopnd)(&te, TO_NONOP, false);
+			/* check for closing parenthesis */
+			op = (*te.isa)(&te, TM_CPAREN);
+			/* back up to operand */
+			te.pos.wp = swp;
+			/* if there was a closing paren, handle it */
+			if (op)
+				goto ptest_one;
+			/* backing up is done before calling the parser */
+		}
+		/* let the parser deal with it */
+		break;
+	case 4:
+		if ((*te.isa)(&te, TM_NOT)) {
+			++invert;
+			goto ptest_three;
+		}
+		if ((*te.isa)(&te, TM_OPAREN)) {
+			swp = te.pos.wp;
+			/* skip two operands, without evaluation */
+			(*te.getopnd)(&te, TO_NONOP, false);
+			(*te.getopnd)(&te, TO_NONOP, false);
+			/* check for closing parenthesis */
+			op = (*te.isa)(&te, TM_CPAREN);
+			/* back up to first operand */
+			te.pos.wp = swp;
+			/* if there was a closing paren, handle it */
+			if (op)
+				goto ptest_two;
+			/* backing up is done before calling the parser */
+		}
+		/* defer this to the parser */
 		break;
 	}
-		te.pos.wp = owp + 1;
-		te.wp_end = owpend;
-	}
 
+	/* "The results are unspecified." */
+	te.pos.wp = wp + 1;
 	return (test_parse(&te));
 }
 

mksh.1

@@ -1,4 +1,4 @@
-.\" $MirOS: src/bin/mksh/mksh.1,v 1.278 2011/11/11 22:14:02 tg Exp $
+.\" $MirOS: src/bin/mksh/mksh.1,v 1.279 2011/11/30 21:34:14 tg Exp $
 .\" $OpenBSD: ksh.1,v 1.141 2011/09/03 22:59:08 jmc Exp $
 .\"-
 .\" Copyright © 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009,
@@ -72,7 +72,7 @@
 .\" with -mandoc, it might implement .Mx itself, but we want to
 .\" use our own definition. And .Dd must come *first*, always.
 .\"
-.Dd $Mdocdate: November 11 2011 $
+.Dd $Mdocdate: November 30 2011 $
 .\"
 .\" Check which macro package we use
 .\"
@@ -3549,7 +3549,7 @@ The file type may be
 (character type device),
 or
 .Cm p
-(named pipe).
+.Pq named pipe , Tn FIFO .
 The file created may be modified according to its
 .Ar mode
 (via the
@@ -4296,7 +4296,8 @@ instead of
 .Ql xtrace .
 .It Fl p Ar file
 .Ar file
-is a named pipe.
+is a named pipe
+.Pq Tn FIFO .
 .It Fl r Ar file
 .Ar file
 exists and is readable.
@@ -4394,22 +4395,22 @@ a mathematical term or the name of an integer variable:
 x=1; [ "x" \-eq 1 ]	evaluates to true
 .Ed
 .Pp
-Note that some special rules are applied (courtesy of POSIX)
+Note that some special rules are applied (courtesy of
-if the number of
+.Px
-arguments to
+) if the number of arguments to
 .Ic test
-or
+or inside the brackets
 .Ic \&[ ... \&]
 is less than five: if leading
 .Ql \&!
-arguments can be stripped such that only one argument remains then a string
+arguments can be stripped such that only one to three arguments remain,
-length test is performed (again, even if the argument is a unary operator); if
+then the lowered comparison is executed; (thanks to XSI) parentheses
-leading
+.Ic \e( ... \e)
-.Ql \&!
+lower four- and three-argument forms to two- and one-argument forms,
-arguments can be stripped such that three arguments remain and the second
+respectively; three-argument forms ultimately prefer binary operations,
-argument is a binary operator, then the binary operation is performed (even
+followed by negation and parenthesis lowering; two- and four-argument forms
-if the first argument is a unary operator, including an unstripped
+prefer negation followed by parenthesis; the one-argument form always implies
-.Ql \&! ) .
+.Fl n .
 .Pp
 .Sy Note :
 A common mistake is to use

sh.h

@@ -151,9 +151,9 @@
 #endif
 
 #ifdef EXTERN
-__RCSID("$MirOS: src/bin/mksh/sh.h,v 1.505 2011/11/26 00:45:21 tg Exp $");
+__RCSID("$MirOS: src/bin/mksh/sh.h,v 1.506 2011/11/30 21:34:15 tg Exp $");
 #endif
-#define MKSH_VERSION "R40 2011/11/25"
+#define MKSH_VERSION "R40 2011/11/30"
 
 #ifndef MKSH_INCLUDES_ONLY