• add code to support GNU bash’s “&> file” I/O redirection extension,

and make it fit into mksh’s model (also gives us a couple of things
  GNU bash doesn’t have
• add regression tests for all of these

Lukas “smultron” Upton from MidnightBSD spotted a script with /bin/sh
shebang invalidly using “&>” in some Apple backup toolkit, 10x

XXX why fds are limited to one digit?

check.t

@@ -1,4 +1,4 @@
-# $MirOS: src/bin/mksh/check.t,v 1.200 2008/06/21 19:30:49 tg Exp $
+# $MirOS: src/bin/mksh/check.t,v 1.201 2008/06/28 22:51:54 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 $
@@ -7,7 +7,7 @@
 # http://www.research.att.com/~gsf/public/ifs.sh
 
 expected-stdout:
-	@(#)MIRBSD KSH R34 2008/06/08
+	@(#)MIRBSD KSH R34 2008/06/28
 description:
 	Check version of shell.
 stdin:
@@ -5000,4 +5000,129 @@ stdin:
 expected-stdout:
 	okay
 ---
-
+name: bashiop-1
+description:
+	Check if GNU bash-like I/O redirection works
+	Part 1: this is also supported by GNU bash
+stdin:
+	exec 3>&1
+	function threeout {
+		echo ras
+		echo dwa >&2
+		echo tri >&3
+	}
+	threeout &>foo
+	echo ===
+	cat foo
+expected-stdout:
+	tri
+	===
+	ras
+	dwa
+---
+name: bashiop-2a
+description:
+	Check if GNU bash-like I/O redirection works
+	Part 2: this is *not* supported by GNU bash
+stdin:
+	exec 3>&1
+	function threeout {
+		echo ras
+		echo dwa >&2
+		echo tri >&3
+	}
+	threeout 3&>foo
+	echo ===
+	cat foo
+expected-stdout:
+	ras
+	===
+	dwa
+	tri
+---
+name: bashiop-2b
+description:
+	Check if GNU bash-like I/O redirection works
+	Part 2: this is *not* supported by GNU bash
+stdin:
+	exec 3>&1
+	function threeout {
+		echo ras
+		echo dwa >&2
+		echo tri >&3
+	}
+	threeout 3>foo &>&3
+	echo ===
+	cat foo
+expected-stdout:
+	===
+	ras
+	dwa
+	tri
+---
+name: bashiop-2c
+description:
+	Check if GNU bash-like I/O redirection works
+	Part 2: this is *not* supported by GNU bash
+stdin:
+	echo mir >foo
+	set -o noclobber
+	exec 3>&1
+	function threeout {
+		echo ras
+		echo dwa >&2
+		echo tri >&3
+	}
+	threeout &>>foo
+	echo ===
+	cat foo
+expected-stdout:
+	tri
+	===
+	mir
+	ras
+	dwa
+---
+name: bashiop-3a
+description:
+	Check if GNU bash-like I/O redirection fails correctly
+	Part 1: this is also supported by GNU bash
+stdin:
+	echo mir >foo
+	set -o noclobber
+	exec 3>&1
+	function threeout {
+		echo ras
+		echo dwa >&2
+		echo tri >&3
+	}
+	threeout &>foo
+	echo ===
+	cat foo
+expected-stdout:
+	===
+	mir
+expected-stderr-pattern: /.*: cannot (create|overwrite) .*/
+---
+name: bashiop-3b
+description:
+	Check if GNU bash-like I/O redirection fails correctly
+	Part 2: this is *not* supported by GNU bash
+stdin:
+	echo mir >foo
+	set -o noclobber
+	exec 3>&1
+	function threeout {
+		echo ras
+		echo dwa >&2
+		echo tri >&3
+	}
+	threeout &>|foo
+	echo ===
+	cat foo
+expected-stdout:
+	tri
+	===
+	ras
+	dwa
+---

lex.c

@@ -2,7 +2,7 @@
 
 #include "sh.h"
 
-__RCSID("$MirOS: src/bin/mksh/lex.c,v 1.61 2008/06/28 22:01:44 tg Exp $");
+__RCSID("$MirOS: src/bin/mksh/lex.c,v 1.62 2008/06/28 22:51:54 tg Exp $");
 
 /*
  * states while lexing word
@@ -771,7 +771,7 @@ yylex(int cf)
 		state = SBASE;
 
 	dp = Xstring(ws, wp);
-	if ((c == '<' || c == '>') && state == SBASE &&
+	if ((c == '<' || c == '>' || c == '&') && state == SBASE &&
 	    ((c2 = Xlength(ws, wp)) == 0 ||
 	    (c2 == 2 && dp[0] == CHAR && ksh_isdigit(dp[1])))) {
 		struct ioword *iop = (struct ioword *)alloc(sizeof (*iop),
@@ -780,12 +780,22 @@ yylex(int cf)
 		if (c2 == 2)
 			iop->unit = dp[1] - '0';
 		else
-			iop->unit = c == '>' ? 1 : 0;
+			iop->unit = c == '<' ? 0 : 1;
+
+		if (c == '&') {
+			if ((c2 = getsc()) != '>') {
+				ungetsc(c2);
+				goto no_iop;
+			}
+			c = c2;
+			iop->flag = IOBASH;
+		} else
+			iop->flag = 0;
 
 		c2 = getsc();
 		/* <<, >>, <> are ok, >< is not */
 		if (c == c2 || (c == '<' && c2 == '>')) {
-			iop->flag = c == c2 ?
+			iop->flag |= c == c2 ?
 			    (c == '>' ? IOCAT : IOHERE) : IORDWR;
 			if (iop->flag == IOHERE) {
 				if ((c2 = getsc()) == '-')
@@ -794,9 +804,9 @@ yylex(int cf)
 					ungetsc(c2);
 			}
 		} else if (c2 == '&')
-			iop->flag = IODUP | (c == '<' ? IORDUP : 0);
+			iop->flag |= IODUP | (c == '<' ? IORDUP : 0);
 		else {
-			iop->flag = c == '>' ? IOWRITE : IOREAD;
+			iop->flag |= c == '>' ? IOWRITE : IOREAD;
 			if (c == '>' && c2 == '|')
 				iop->flag |= IOCLOB;
 			else
@@ -809,6 +819,8 @@ yylex(int cf)
 		Xfree(ws, wp);	/* free word */
 		yylval.iop = iop;
 		return REDIR;
+ no_iop:
+		;
 	}
 
 	if (wp == dp && state == SBASE) {

mksh.1

@@ -1,4 +1,4 @@
-.\" $MirOS: src/bin/mksh/mksh.1,v 1.129 2008/06/08 16:57:52 tg Exp $
+.\" $MirOS: src/bin/mksh/mksh.1,v 1.130 2008/06/28 22:51:55 tg Exp $
 .\" $OpenBSD: ksh.1,v 1.122 2008/05/17 23:31:52 sobrado Exp $
 .\"-
 .\" Try to make GNU groff and AT&T nroff more compatible
@@ -30,7 +30,7 @@
 .el .xD \\$1 \\$2 \\$3 \\$4 \\$5 \\$6 \\$7 \\$8
 ..
 .\"-
-.Dd $Mdocdate: June 8 2008 $
+.Dd $Mdocdate: June 28 2008 $
 .Dt MKSH 1
 .Os MirBSD
 .Sh NAME
@@ -252,6 +252,7 @@ The meta-characters are used in building the following
 .Ql \*(Gt ,
 .Ql \*(Gt& ,
 .Ql \*(Gt\*(Gt ,
+.Ql &\*(Gt ,
 etc. are used to specify redirections (see
 .Sx Input/output redirection
 below);
@@ -2015,6 +2016,37 @@ indicating standard input is to be closed.
 Same as
 .Ic \*(Lt& ,
 except the operation is done on standard output.
+.It &\*(Gt Ar file
+Same as
+.Ic \*(Gt Ar file 2\*(Gt&1 .
+This is a GNU
+.Nm bash
+extension supported by
+.Nm
+which also supports the preceding explicit fd digit, for example,
+.Ic 3&\*(Gt Ar file
+is the same as
+.Ic 3\*(Gt Ar file 2\*(Gt&3
+in
+.Nm
+but a syntax error in GNU
+.Nm bash .
+.It Xo
+.No &\*(Gt| Ar file ,
+.No &\*(Gt\*(Gt Ar file ,
+.No &\*(Gt& Ar fd
+.Xc
+Same as
+.Ic \*(Gt| Ar file ,
+.Ic \*(Gt\*(Gt Ar file ,
+or
+.Ic \*(Gt& Ar fd ,
+followed by
+.Ic 2\*(Gt&1 ,
+as above.
+These are
+.Nm
+extensions.
 .El
 .Pp
 In any of the above redirections, the file descriptor that is redirected
@@ -5489,7 +5521,7 @@ and many other persons, and is currently maintained by
 .An Thorsten Glaser Aq tg@mirbsd.de .
 .Sh BUGS
 This document attempts to describe
-.Nm mksh\ R34
+.Nm mksh\ R35
 and up,
 compiled without any options impacting functionality, such as
 .Dv MKSH_SMALL ,

sh.h

@@ -100,9 +100,9 @@
 #define __SCCSID(x)	__IDSTRING(sccsid,x)
 
 #ifdef EXTERN
-__RCSID("$MirOS: src/bin/mksh/sh.h,v 1.219 2008/06/08 17:15:30 tg Exp $");
+__RCSID("$MirOS: src/bin/mksh/sh.h,v 1.220 2008/06/28 22:51:55 tg Exp $");
 #endif
-#define MKSH_VERSION "R34 2008/06/08"
+#define MKSH_VERSION "R34 2008/06/28"
 
 #ifndef MKSH_INCLUDES_ONLY
 
@@ -970,6 +970,7 @@ struct ioword {
 #define IOCLOB	BIT(6)	/* >|, override -o noclobber */
 #define IORDUP	BIT(7)	/* x<&y (as opposed to x>&y) */
 #define IONAMEXP BIT(8)	/* name has been expanded */
+#define IOBASH	BIT(9)	/* &> etc. */
 
 /* execute/exchild flags */
 #define XEXEC	BIT(0)		/* execute without forking */

syn.c

@@ -2,7 +2,7 @@
 
 #include "sh.h"
 
-__RCSID("$MirOS: src/bin/mksh/syn.c,v 1.21 2008/05/17 18:47:02 tg Exp $");
+__RCSID("$MirOS: src/bin/mksh/syn.c,v 1.22 2008/06/28 22:51:56 tg Exp $");
 
 struct nesting_state {
 	int start_token;	/* token than began nesting (eg, FOR) */
@@ -144,10 +144,17 @@ static struct ioword *
 synio(int cf)
 {
 	struct ioword *iop;
+	static struct ioword *nextiop = NULL;
 	int ishere;
 
+	if (nextiop != NULL) {
+		iop = nextiop;
+		nextiop = NULL;
+		return (iop);
+	}
+
 	if (tpeek(cf) != REDIR)
-		return NULL;
+		return (NULL);
 	ACCEPT;
 	iop = yylval.iop;
 	ishere = (iop->flag&IOTYPE) == IOHERE;
@@ -161,7 +168,18 @@ synio(int cf)
 		*herep++ = iop;
 	} else
 		iop->name = yylval.cp;
-	return iop;
+
+	if (iop->flag & IOBASH) {
+		nextiop = (struct ioword *)alloc(sizeof (*iop), ATEMP);
+
+		iop->flag &= ~IOBASH;
+		nextiop->unit = 2;
+		nextiop->flag = IODUP;
+		nextiop->name = shf_smprintf("%d", iop->unit);
+		nextiop->delim = NULL;
+		nextiop->heredoc = NULL;
+	}
+	return (iop);
 }
 
 static struct op *
@@ -211,9 +229,11 @@ get_command(int cf)
 			    (XPsize(args) == 0 ? ALIAS|VARASN : CMDWORD);
 			switch (tpeek(cf)) {
 			case REDIR:
-				if (iopn >= NUFILE)
+				while ((iop = synio(cf)) != NULL) {
-					yyerror("too many redirections\n");
+					if (iopn >= NUFILE)
-				iops[iopn++] = synio(cf);
+						yyerror("too many redirections\n");
+					iops[iopn++] = iop;
+				}
 				break;
 
 			case LWORD: