zsh-3.1.4 bug, parsing case ... esac

zsh-3.1.4 bug, parsing case ... esac

[SL Baur]

The following code, when executed in zsh running as /bin/sh produces
the error messages:
zsh: parse error near `echo'
zsh: parse error in command substitution

"BISON=`if [ -f /usr/src/egcs-1.1a/gcc/../bison/bison ] ; then
		case . in
		/*) echo /usr/src/egcs-1.1a/gcc/../bison/bison -L ./../bison/ ;;
		*) echo /usr/src/egcs-1.1a/gcc/../bison/bison -L /usr/src/egcs-1.1a/gcc/./../bison/ ;;
		esac else echo bison ;
	fi`"

adding a semicolon after the esac makes the messages go away, but I
think this should work as is.

This test case came from egcs-1.1a and causes the build to break.

Re: zsh-3.1.4 bug, parsing case ... esac

[Bart Schaefer]

On Sep 2,  5:53pm, SL Baur wrote:
} Subject: zsh-3.1.4 bug, parsing case ... esac
}
} zsh: parse error near `echo'
} zsh: parse error in command substitution
} 
} 		esac else echo bison ;
} 
} adding a semicolon after the esac makes the messages go away, but I
} think this should work as is.

Blech.  That sure doesn't parse according to the bash manual.  The
stuff between "then" and "else" has to be a properly-terminated list,
which means it has to end with ; & or newline.

Anybody have access to a ksh manual?

It might be possible to special-case (ahem) this because esac is a
reserved word, but zsh's parser is doing exactly what it is documented
to do, so it can't be called a bug (except in egcs's makefile).

I'd report it to the egcs folks and see what they have to say.  My guess is
this wasn't intentional on their part.  It does appear that bash accepts
"esac", "done" or "fi" as termination of a "list", so it even accepts the
following gobbledygook:

if false; then if false; then if false; then; fi fi fi
if false; then case x in x);; esac else while true; do break; done fi
for y in y; do if true; then case x in x) break;; esac fi done

This only works in "list" context, e.g., if you run the above three lines
together it won't parse the "fi if".  I suspect all the above is terribly
unportable and should not be used in a reasonable shell script or makefile.

-- 
Bart Schaefer                                 Brass Lantern Enterprises
http://www.well.com/user/barts              http://www.brasslantern.com

Re: zsh-3.1.4 bug, parsing case ... esac

[Zoltan Hidvegi]

> On Sep 2,  5:53pm, SL Baur wrote:
> } Subject: zsh-3.1.4 bug, parsing case ... esac
> }
> } zsh: parse error near `echo'
> } zsh: parse error in command substitution
> } 
> } 		esac else echo bison ;
> } 
> } adding a semicolon after the esac makes the messages go away, but I
> } think this should work as is.
> 
> Blech.  That sure doesn't parse according to the bash manual.  The
> stuff between "then" and "else" has to be a properly-terminated list,
> which means it has to end with ; & or newline.

Actually, the shell grammar allows this, this is really a zsh bug.
For example,

if true; then if true; then echo foo; fi else echo bar; fi

is accepted by zsh although there is no semicolon between fi and else.
Similarily

if ((some math)) then ((some other math)) fi

works.  Same for [[ ... ]].

Zoli


*** Src/parse.c~	Sat May  2 03:45:37 1998
--- Src/parse.c	Thu Sep  3 00:36:33 1998
***************
*** 542,557 ****
  
  	while (tok == SEPER)
  	    yylex();
! 	if (tok == OUTBRACE) {
! 	    yylex();
  	    break;
- 	}
  	if (tok != STRING)
  	    YYERRORV;
! 	if (!strcmp(tokstr, "esac")) {
! 	    yylex();
  	    break;
- 	}
  	str = ncalloc(strlen(tokstr) + 2);
  	*str = ';';
  	strcpy(str + 1, tokstr);
--- 542,553 ----
  
  	while (tok == SEPER)
  	    yylex();
! 	if (tok == OUTBRACE)
  	    break;
  	if (tok != STRING)
  	    YYERRORV;
! 	if (!strcmp(tokstr, "esac"))
  	    break;
  	str = ncalloc(strlen(tokstr) + 2);
  	*str = ';';
  	strcpy(str + 1, tokstr);
***************
*** 618,627 ****
  	addlinknode(pats, str);
  	addlinknode(lists, par_list());
  	n++;
! 	if ((tok == ESAC && !brflag) || (tok == OUTBRACE && brflag)) {
! 	    yylex();
  	    break;
- 	}
  	if(tok == SEMIAMP)
  	    *str = '&';
  	else if (tok != DSEMI)
--- 614,621 ----
  	addlinknode(pats, str);
  	addlinknode(lists, par_list());
  	n++;
! 	if ((tok == ESAC && !brflag) || (tok == OUTBRACE && brflag))
  	    break;
  	if(tok == SEMIAMP)
  	    *str = '&';
  	else if (tok != DSEMI)
***************
*** 630,635 ****
--- 624,632 ----
  	incmdpos = 0;
  	yylex();
      }
+ 
+     incmdpos = 1;
+     yylex();
  
      cc->pats = (char **)alloc((n + 1) * sizeof(char *));
  

Re: zsh-3.1.4 bug, parsing case ... esac

[Bart Schaefer]

On Sep 3, 12:43am, Zoltan Hidvegi wrote:
} Subject: Re: zsh-3.1.4 bug, parsing case ... esac
}
} > } 		esac else echo bison ;
} > 
} > Blech.  That sure doesn't parse according to the bash manual.  The
} > stuff between "then" and "else" has to be a properly-terminated list,
} 
} Actually, the shell grammar allows this, this is really a zsh bug.

Obviously, this is a surprise to me.  Can you point me at "the shell
grammar" that specifies this?  The informal one in the man pages/info
sure doesn't.

} Similarily
} 
} if ((some math)) then ((some other math)) fi
} 
} works.  Same for [[ ... ]].

Yes, I knew about those, but I thought they were handled specially.

-- 
Bart Schaefer                                 Brass Lantern Enterprises
http://www.well.com/user/barts              http://www.brasslantern.com

PATCH: 3.0.5 - Re: zsh-3.1.4 bug, parsing case ... esac

[Bart Schaefer]

On Sep 3, 12:43am, Zoltan Hidvegi wrote:
} Subject: Re: zsh-3.1.4 bug, parsing case ... esac
} > }
} > } zsh: parse error near `echo'
} > } zsh: parse error in command substitution
} > } 
} > } 		esac else echo bison ;
} 
} Actually, the shell grammar allows this, this is really a zsh bug.

Here's Zoltan's patch tweaked for 3.0.5.

Index: Src/parse.c
===================================================================
--- parse.c	1998/07/02 15:56:02	1.6
+++ parse.c	1998/09/03 14:55:26
@@ -482,16 +482,12 @@
 
 	while (tok == SEPER)
 	    yylex();
-	if (tok == OUTBRACE) {
-	    yylex();
+	if (tok == OUTBRACE)
 	    break;
-	}
 	if (tok != STRING)
 	    YYERRORV;
-	if (!strcmp(tokstr, "esac")) {
-	    yylex();
+	if (!strcmp(tokstr, "esac"))
 	    break;
-	}
 	str = tokstr;
 	incasepat = 0;
 	incmdpos = 1;
@@ -556,16 +552,17 @@
 	addlinknode(pats, str);
 	addlinknode(lists, par_list());
 	n++;
-	if ((tok == ESAC && !brflag) || (tok == OUTBRACE && brflag)) {
-	    yylex();
+	if ((tok == ESAC && !brflag) || (tok == OUTBRACE && brflag))
 	    break;
-	}
 	if (tok != DSEMI)
 	    YYERRORV;
 	incasepat = 1;
 	incmdpos = 0;
 	yylex();
     }
+
+    incmdpos = 1;
+    yylex();
 
     cc->pats = (char **)alloc((n + 1) * sizeof(char *));
 


-- 
Bart Schaefer                                 Brass Lantern Enterprises
http://www.well.com/user/barts              http://www.brasslantern.com

Re: zsh-3.1.4 bug, parsing case ... esac

[Zoltan Hidvegi]

> } Actually, the shell grammar allows this, this is really a zsh bug.
> 
> Obviously, this is a surprise to me.  Can you point me at "the shell
> grammar" that specifies this?  The informal one in the man pages/info
> sure doesn't.

You can look at the Single Unix Specification -> Commands & Utilities
-> Shell Command Language -> Shell Grammar Rules:

if_clause        : If compound_list Then compound_list else_part Fi
                 | If compound_list Then compound_list           Fi
                 ;

compound_list    :              term
                 | newline_list term
                 |              term separator
                 | newline_list term separator
                 ;

term             : term separator and_or
                 |                and_or
                 ;

and_or           :                         pipeline
                 | and_or AND_IF linebreak pipeline
                 | and_or OR_IF  linebreak pipeline
                 ;


pipeline         :      pipe_sequence
                 | Bang pipe_sequence
                 ;

pipe_sequence    :                             command
                 | pipe_sequence '|' linebreak command
                 ;

command          : simple_command
                 | compound_command
                 | compound_command redirect_list
                 | function_definition
                 ;

compound_command : brace_group
                 | subshell
                 | for_clause
                 | case_clause
                 | if_clause
                 | while_clause
                 | until_clause
                 ;

case_clause      : Case WORD linebreak in linebreak case_list Esac
                 | Case WORD linebreak in linebreak           Esac
                 ;

There is more, but as you see, there is no required separator.
Obviously the manual and info pages are not that accurate.  People are
already complaining aboud the complicated and hard to understand
manual, a formal grammar description would be even more difficult to
read.

Zoli