The SYNTAX/FNC2 FAQ

[Source Mihaela Sighireanu]

Yes, there is a sun4 version of fnc2 availiable. This version is quite new, so there may exist little differences of behaviour with the sun5 version due to the different architectures.

[Source Mihaela Sighireanu and Bruno Vivien]

If your directories for syntax and fnc2 are resp. /common/Syntax and /common/Fnc2, then the following environment variables shall be set :

setenv sx /common/Syntax
setenv f2 /common/Fnc2
set f2arch = `$f2/doc/sys`
setenv F2PATH .:$f2/lib/$f2arch
set path = ($path $sx/bin/$f2arch $f2/bin/$f2arch $f2/doc)

alias FNC2      "(set path = (/usr/local/gnu/bin $path) ; setenv CFLAGS
g" ; setenv LDFLAGS "-g" ;$f2/doc/FNC2)"

[Source Bruno Vivien]

• If you compile without the FNC2 shell: add option "OPTION nnull" for all ASX grammars, for example

ASX elotos-base elotos-base import
OPTION nnull 
end import ;

• If you use the FNC2 shell, use the option "u" at construction.
• If you use the FNC2 command-line mode, add the option "-nnull" to your FNC2 call

[Source Mihaela Sighireanu]

The fnc2 use by default the option "-total" while compiling attribute grammars. This option forces the evaluation of all attributes either if they are not used to compute attributes defined for the root node. When you have this "incomplete" grammars, you obtain the errors at Compilation, for example:

BINARY
cc -o /local_home/users/sighirea/Traian/Code/bin/sun5/traian
Undefined                       first referenced
symbol                             in file
v_141_53_29_26_108_14_12           
/local_home/users/sighirea/Traian/Code/lib/sun5/elotos_int_bind_auto.o
v_141_53_29_26_108_14_13           
/local_home/users/sighirea/Traian/Code/lib/sun5/elotos_int_bind_auto.o
v_29_108_14                        
/local_home/users/sighirea/Traian/Code/lib/sun5/elotos_int_bind_auto.o
ld: fatal: Symbol referencing errors No output written to
/local_home/users/sighirea/Traian/Code/bin/sun5/traian
rm: /local_home/users/sighirea/Traian/Code/bin/sun5/traian: No such file
or directory
gmake: *** [/local_home/users/sighirea/Traian/Code/bin/sun5/traian]
Error 1

Note that the "ntotal" option is not documented in the chapter 18 of the FNC2 manual.

[Source Mihaela Sighireanu]

While using "-ntotal" option, the evaluation of attributes which are not needed by the computation of the "result" attributes is not forced. The compilation with FNC2 does not signal such cases. However, you may obtain an executable which "core dumps" on function `f2___oper_attr`.

To avoid such situations, while working with incomplete grammars (option "ntotal") you have to check if your exported attributes are used in the computation of the resulting attributes (the attributes declared as the result of the attributes grammar in the ag_XXXX.olga file). You can check this by looking to the listing produced by fnc2 for your grammar (tmp/ag_XXXX.olga.l file) at the section RESULTS OF STRONGLY NON-CIRCULAR TEST

[Source Mihaela Sighireanu (BUG 016 in v1.16)]

An imported attribute ***cannot be used*** as a global attributed in the importing grammar, either it was declared as "global" in the source grammar. This is a known bug of FNC2.

For example, you have the following declarations :

attribute grammar AG1 (ASX1 in ph-axiom ; ASX2 out ph-axiom) : ($correct1 : bool)
is
attribute
	{ exported }
	global $I_A (ph-stmt) : bool;
{ *** your grammar *** }
end grammar
grammar ASX2 is
root is ph-axiom;
	$I_A (ph-stmt) : bool;
end grammar
attribute grammar AG2 (ASX2 in ph-axiom ; ASX3 out ph-axiom) : ($correct2 : bool)
is
attribute
{ *** this grammar uses $I_A [ph-stmt] *** }
end grammar

[Source Mihaela Sighireanu (BUG 019 in v1.16)]

One of the bugs of the "evalgen" tool concerns the evaluation of attributes while visiting list phylum. This bug appears at the generation of the evaluator for the compound attributes defined on list phylum. For example, the definition given below for the inherited part of the compound attribute $A1 may give strange results due of this bug:

where op-top-level-declarations -> 
ph-top-level-declaration  *
use
	$h.A1 (ph-top-level-declaration) := $h.A1;
end where

where op-top-level-declarations -> 
ph-top-level-declaration  *
use
	$h.A1 (ph-top-level-declaration) := 
		case position us
			first :	$h.A1;
			other : let t := $s.A1 (ph-top-level-declaration.left) 
				in $h.A1;
		end case;
end where

[Source Mihaela Sighireanu (BUG 019 in v1.16)]

Non, the map expressions using local attributes (variables) are not correctly treated by fnc2 (translation to C). For example, the following expression may give strange results at the execution of the generated compiler:

where op-module-expression-list -> 
ph-module-expression  * 
declare
	value
	CORRECT_SIZE : bool := ... ;
use 
	$CORRECT := 
		map left & 
value CORRECT_SIZE
other $CORRECT (ph-module-expression) 
end map; 
end where;

where op-module-expression-list -> 
ph-module-expression  * 
declare
	value
	CORRECT_SIZE : bool := ... ;
	CORRECT_MAP : bool :=
		map left & 
value true
other $CORRECT (ph-module-expression) 
end map; 
	
use 
	$CORRECT := CORRECT_SIZE & CORRECT_MAP;
end where;

[Source Mihaela Sighireanu]

No, you have to use the op-merge function. For example:

ph-flat-definitions = op-flat-definitions;
op-flat-definitions -> ph-flat-definition *;
map left op-flat-definitions-merge { and not + }
value op-flat-definitions ()
other $s_FLAT_definitions (ph-declarations)
end map

[Source Mihaela Sighireanu]

No, the "op-all" pattern may be use only in atc files. Into an AG you have to use your (recursive) functions and the simple pattern-matching ("case ... is") to obtain the same effect. Example: The destination (exit) abstract grammar is :

ph-flat-behaviour-match-list  = op-flat-behaviour-match-list ;
op-flat-behaviour-match-list -> ph-flat-behaviour-match *;
ph-flat-behaviour-match       = op-flat-behaviour-match ;              
op-flat-behaviour-match      -> ph-flat-pattern
ph-flat-value-expression
ph-flat-behaviour ;

s_FLAT_behaviour_match_list :=
case $s_FLAT_pattern_list (ph-pattern-list) is
op-flat-pattern-list-all (Pi) :
map tree Pi left op-flat-behaviour-match-list-merge 
value op-flat-behaviour-match-list ()
other case Pi is
op-flat-pattern-list (Pj) :
op-flat-behaviour-match-list (
op-flat-behaviour-match (Pj,
s_FLAT_value_expression (ph-value-expression),
s_FLAT_behaviour (ph-behaviour)));
other :
op-flat-behaviour-match-list ();
end case
end map;
other : { erreur }
op-flat-behaviour-match-list ();
end case;

	$s_FLAT_behaviour_match_list :=
		DISTRIBUTE_P1_Pn_B ($s_FLAT_pattern_list (ph-pattern-list), $s_FLAT_value_expression (ph-value-expression), $s_FLAT_behaviour (ph-behaviour));

function DISTRIBUTE_P1_Pn_B (P1_Pn : ph-flat-pattern-list; VE : ph-flat-value-expression; B : ph-flat-behaviour)
		: ph-flat-behaviour-match-list
is
	case P1_Pn is
	op-flat-pattern-list-pre (P1, tail_P1_Pn) :
		op-flat-behaviour-match-list-pre (
			op-flat-behaviour-match (P1, VE, B),
			DISTRIBUTE_P1_Pn_B (tail_P1_Pn, VE, B));
	other : 
		op-flat-behaviour-match-list ();
	end case
end function;

[Source Mihaela Sighireanu (BUG 020 in v1.16)]

No, the C code generated actually by fnc2 is incorrect. To avoid this bug write an auxiliary function accessing the attribute; you will obtain only a warning at compilation of the code generated by fnc2. Example:

function CREATE_SEQUENCE_E (E1, E2 : ph-flat-expression) 
ph-flat-expression 
is 
case E1 is 
op-flat-null-expression () : E2; 
other : 
case E2 is 
op-flat-null-expression () : E1; 
other : op-flat-sequential-expression (E1, E2) with $TYPE := $TYPE(E2) { access of the $TYPE attribute } end
with; 
end case; 
end case 
end function; 

function XXX(E:ph-flat-expression) : TYPE_EXTENDED_ID_LIST is 
TYPE (E) 
end function ; 
function CREATE_SEQUENCE_E (E1, E2 : ph-flat-expression) 
ph-flat-expression 
is 
case E1 is 
op-flat-null-expression () : E2; 
other : 
case E2 is 
op-flat-null-expression () : E1; 
other : op-flat-sequential-expression (E1, E2) with $TYPE := XXX(E2) end
with; 
end case; 
end case 
end function; 

[Source Mihaela Sighireanu (BUG 022 in v1.16)]

No, the "case arity" can not be used for attributes declared in the "declare value" section of a rule. Example:

where op-flat-gate-match-list -> ph-flat-gate-match  *
declare value
	DOMAINE : TYPE_DOMAINE :=
		case arity is
			0 :	DOMAINE_VIDE;
			other :	$s.DOMAINE (ph-flat-gate-match.last);
		end case;
	IS_EXHAUSTIVE : bool :=
		if !empty (DOMAINE) then
			message-lotosnt (LOTOSNT_NOT_EXHAUSTIVE_PATTERNS,
					MODE_DEBUG,
					"Gate match list:",
					$scx)
		elsif (EXISTS_GUARD & !EXISTS_ANY) then
			message-lotosnt (LOTOSNT_MAYBE_NOT_EXHAUSTIVE_PATTERNS,
					MODE_DEBUG,
					"Gate match list:",
					$scx)
		else
			true
		end if;
use
	$CORRECT_PH1 := IS_EXHAUSTIVE;

	$CORRECT_PH1 :=
		case arity is
			0 :	message-lotosnt (LOTOSNT_NOT_EXHAUSTIVE_PATTERNS,
					MODE_DEBUG,
					"Gate match list:",
					$scx);
			other :	if !empty ($s.DOMAINE (ph-flat-gate-match.last)) then
				message-lotosnt (LOTOSNT_NOT_EXHAUSTIVE_PATTERNS,
						MODE_DEBUG,
						"Gate match list:",
						$scx)
				elsif (EXISTS_GUARD & !EXISTS_ANY) then
				message-lotosnt (LOTOSNT_MAYBE_NOT_EXHAUSTIVE_PATTERNS,
						MODE_DEBUG,
						"Gate match list:",
						$scx)
				else
					true
				end if; 
		end case;
	

[Source Mihaela Sighireanu]

• Declare the ASX1 and ASX2 abstract attributed trees with roots resp. ph-root1 and ph-root2
• Declare the grammar ag_AG_transf.olga as follows:

attribute grammar AG1 (ASX1 in ph-root1)
($ASX2_ROOT : ph-root2; $CORRECT_AG1 : bool) is

• Declare the declaration module for AG1 as follows:

declaration module AG1 is
import grammar ASX1 ;
import grammar ASX2 ;
function AG1 (n1 : ph-root1) : ph-root2
end module ;

• Call the grammar in the XXXX_smp.c file with the classical form. Let r_AG1 be the result of this call.
• Use the root r_AG1->ASX2_ROOT (and not f2atcvar.atc_lv.abstract_tree_root) as the new root for the call of the grammars or the ppats working in the ASX2.
• NOTE that the $scx predefined attribute can not be used on the ASX2 phyla directly because it was lost by the tree transformation. You have to specify at each node construction the $scx attribute. For list phyla, this does not work (see BUG 030). The fnc2 does not signal such (incorrect) uses of $scx and you can have core dump at the execution of the generated binary.

[Source Bruno Vivien]

If you write comments inside an atc rule as follows :

	_non-terminal_= _other_non_terminals_
{_commented_non_terminal_}  
			_other_non_terminals_

[Source Christophe Discours]

Yes. See FNC2 manual vX.X page ???.

[Source Hubert Garavel]

Waiting response.

[Source Mihaela Sighireanu]

If you use the "null-*" nodes (you don't compile with "-nnull" option), then you should not have such problems. If you use "-nnull" option, atc cannot build correctly the nodes obtained by error recovering. So, you have to test if a syntax error appears in the syntax analysis before the beginning of the semantic analysis. This may be done in your "XXXX_smp.c" file, in the function "XXXXsmp", by testing the variables sxerrmngr.nbmess[1] (nb of errors) and sxerrmngr.nbmess[2] (nb of warnings). For example:

	if ((sxerrmngr.nbmess[1] + sxerrmngr.nbmess[2]) != 0) {
	fprintf (sxstderr, "XXXX:\t Cannot recover from previous syntax errors!\n\tquit\n");

[Source Mihaela Sighireanu]

• Check if a ";" symbol lacks in the `.asx' file compiled. For example:

op-type-array-definition -> 
		   ph-identifier
ph-array-range
ph-identifier
ph-array-range  = op-array-range ;

No, because the fnc2 compiler use only "*" operators to compute attributes. If you use "+" operator you can have unexpected "core dump" at the execution of the generated binary.

[99/01/05 - Claude Chaudet] The "+" operator does not produce core dump but acts just like the "*" operator.

[Source Bruno Vivien]

If syntax do a error recovering on a lexical entity, for example %INT, the text is unknown and syntax gives the value error-token to the token.

[Source Mark Jorgensen]

In the src/XXXX_smp.c file, in the XXXsmp() function, add the line :

	/* Impression de l'arbre (X,Y,Z,T) X: la racine,Y:marge,Z:table pour
la structure des noeud,Z:fichier) */
f2___print_attr_arbre(f2atcvar.atc_lv.abstract_tree_root, 0, f2___atc_table, sortie);

[Source Marc Jorgensen]

The following functions on strings defined at page 103 of the manual seems to be not available:

substr omit insert index search verify

You have to use external definitions written in C to define these functions.

[Source Marc Jorgensen]

No, the "member" operator described at page 122 of the manual uses type inference (meta-type @T) and, as the file $f2/doc/BUGS says, it is not yet implemented.

[Several sources]

As said in Fnc2/doc/BUGS, overloaded functions are not supported. Description du bug : Quand on a deux fonctions surchargees declarees dans un dec_X.olga

function surchargee (N1 : int; N2 : int) : int;
fonction surchargee (R1 : real; R2 : real) : real;

Pour les declarations de dec|def_X.olga FNC-2 genere le code C suivant dans src_fnc2/X_b_df.c

int surchargee (N1 : int, N2 : int) { ... }
real surchargee__real__real (R1 : real, R2 : real) { ... } 

[Source Mihaela Sighireanu]

No, because the SYNTAX tool define the "INIT" identifier in the `sxunix.h` file as follows:

#define INIT SXINIT
SXINIT -1

[Source Mihaela Sighireanu]

The C code generated for this operator is actually incorrect. For example:

message "op-top-level-declarations " + string (CORRECT_PH3)
	position value $scx [ph-top-level-declarations]
severity 2
value CORRECT_PH3;

sxput_error( = arg_1;
%s %s.", sxtab_ptr
err_titles[2],plus___string("op-top-level-declarations "
string___bool(CORRECT_PH3)));
return CORRECT_PH3;

Moreover, due to this wrong translation, while the "message" operator is used into a expression like "... & message .... &", you have to put the message into a "let", like:

let t : bool := message "PH3 " + string ($CORRECT_PH3
(ph-compilation-unit))
position ph-compilation-unit
severity 1
value $CORRECT_PH3 (ph-compilation-unit)
in
t & ...; { or ... & ... when value is already here }

function tutu (arg1:bool;sx:source-index;arg2:bool) : bool is
message "ph-top-level-declaration " + string (arg2)
position sx
severity 2
value arg1 & arg2
end function; 

CORRECT_PH3 :=
map left tutu
value   true
other   $scx(ph-top-level-declaration), 
				$CORRECT_PH3 (ph-top-level-declaration)
end map;

CORRECT_PH3 :=
map left tutu
value   true
other   $scx(ph-top-level-declaration), 
				message "ph-top-level-declaration " + string (arg2)
position sx
severity 2
value $CORRECT_PH3 (ph-top-level-declaration)
end map; 

[Source Bruno Vivien]

The correct syntax of the "case" expression is "case...is" instead of "case ... match" as suggested by the FNC2 manual (appendix B).

[Several sources]

There are no debug tools for OLGA sources. Use C debuggers on the sources generated by fnc2. You can use "message position ... severity ... value ..." construct to write traces.

[Several sources]

When a grammar enriches an existent tree, the declaration order of the attributes no modified should be the same in both (input and output) .asx files. This bug is already known.

[Source Bruno Vivien]

In order to avoid cumbersome tests on error-token, you can directly test the generation of a error-token after the syntax pass, in the XXXX_smp.c file. For this, test the following variables:

sxerrmngr.nbmess[2]/* nombre de messages d'erreurs */
sxerrmngr.nbmess[1]/* nombre de warnings */

[Source Mihaela Sighireanu]

The constants 'beginning-of-text' and 'end-of-text' of type source-index can be used only into an attributed grammar. To use these constants and other ones in a definition or declaration module you can:

• if the function using this value is called from an AG you have to pass the source-index value as parameter
• when the function is not called from a AG, you have to define an external function written in C. For example, the following C function returns the 'beginning-of-text' value:

#include "olga_predefinit.h"
struct sxsource_coord beginning_of_text ()
{
return beginning__of__text;
}

[Source Mihaela Sighireanu]

The implementation of the "+" operation on strings makes a copy of the first argument and then concatenates (using the C "strcat" function) with the second argument. So, you have to limit its use because the memory may be overfilled.

[Source Mihaela Sighireanu]

• Declare the type into a file spec/dec_XXXX.olga as follows:

declaration module XXX is
type TYPE_EXTERNAL ;
end module ;

• Create a file incl/XXXX_t_df.h containing the definitions of the external types declared in dec_XXXX.olga. In this file you can use the OLGA types defined in dec_XXXX.olga.
• Create a file incl/XXXX_b_df.h containing the declarations of the external functions declared in dec_XXXX.olga.
• Create a file src/XXXX_b_df.c as follows:

#include "olga_predefinit.h"
#include "XXXX_b.h"
/* implementation of external functions declared in dec_XXXX.olga */

[Source Mihaela Sighireanu and Claude Chaudet]

The problem is caused by the fact that the declaration of variable "prgentname" was removed by you from file main_xxx.c. This problem seems to occur only under Linux.

[Source Mark Jorgensen]

This message appears when you access directly (without using the "case position" construct) an element of a phylum which is a list.

For example, the following code is WRONG:

where stmt_list -> STMT  + use
	   $h.tool(STMT.first) := "aldebaran" ;
end where ;

where stmt_list -> STMT  + use
	   $h.tool(STMT) := case position is
		   first : "aldebaran" ;
		   other : $s.tool(STMT.left);
	   end case;
end where ;

[Source Marc Jorgensen (Fixed in v1.16)]

Yes, the bug described below is fixed in the version >= v1.16.

Je veux utiliser plusieurs fois PPAT. (J'ai besoin d'imprimer l'arbre de deux manieres differentes). Le seul probleme, est que le fichier SVL.c (SVL est le nom de mon langage) est mauvais, car il contient deux fois les definitions de Largeur, Filename, et les include sont aussi fait deux fois, ce qui cause probleme. Du coup, je dois patcher le fichier SVL.c a la main, et m'arranger pour que FNC2 ne le regenere pas.

[Source Marc Jorgensen (Fixed in v1.16)]

Yes, the bug described below is fixed in the version >= v1.16.

Lorsque j'imprime un token (ici, le token text) avec ppat, celui-ci ne contient pas la bonne donnee. Je contourne le probleme en attachant aux noeuds ou il y a un token text, un attribut contenant le string de text, et j'imprime ce string avec ppat.

[Source Christophe Discours (BUG 025)] Yes, if the attribute is declared in the input grammar of your ppat. ATTENTION: As long as the calls of attributed grammars and ppat are directly written in XXXX_smp.c, the typechecking is not done. So, if you use a local attribute of a previous grammar, ppat accepts this, but the binary obtained produces a core dump (see BUG 025).

PPAT will print on the stdout if the name of the file given as parameter is empty (i.e. ""). For example:

	ppat___block_debug (R_BLOCK->BLOCK_AXIOM, 80, "");

[Source Mihaela Sighireanu]

The prefnc2 tool creates a `langage.recor' and you have to modifying it for your application.

[Source Marc Jorgensen]

Probleme dans le code engendre par FNC2 : En gros, j'ai des problemes lorsque j'utilise des fonctions "look-ahead" dans ma grammaire attribue. Cette fonction (case B1 is beh_src(SOURCE) ci-dessous) fait que le code produit pour le fichier SVL__check_regles.c va chercher dans la table de symboles : SVL__base_asx_table qui n'existe pas. Seuls SVL__in_asx_table et SVL__out_asx_table sont definis.

ag_SVL-check.olga:
attribute grammar SVL-check (SVL-in, SVL-out) is 
where beh_rv_parallel -> B1 B2 ID_L use
	$network_filename(B1) := case B1 is
beh_src(SOURCE) :
network_filename;
other :
new_exp_file();
end case ;
end where ;

[Source Mihaela Sighireanu]

With the version 1.15 of the FNC2, the attribute grammar shall be specified as follows:

attribute grammar _name_
	( _asx_file_ [in|out] _axiom_ { ; _asx_file_ [in|out] _axiom_ } ) 
	: ( _result_attribute_ : _type_ { ; _result_attribute_ : _type_ } ) is
		etc.
end grammar

[Source Mihaela Sighireanu]

The listing generated with `sxlisting_output` swaps the number of errors and the number of warning. For example 14 warnings are reported but the final message in the ".l" file is: "14 errors and 0 warnings are reported."

[Source Mihaela Sighireanu]

On the `warn_12.tar.gz` example.

[Source Mihaela Sighireanu]

The tool bnf_read of SYNTAX cut the names of files to 10 characters. So, the name of the language should have maximum 10 characters.

[Source Mihaela Sighireanu]

The $scx predefined attribute cannot be used on the phyla of a transformed tree (the tree not having an atc source). At the initial tree transformation this information is lost. You have to define and to compute your attribute on the transformed tree.. The fnc2 does not signal such (incorrect) uses of $scx and you can have core dump at the execution of the generated binary.

[Source Mihaela Sighireanu]

You have to search in the listing file tmp/*.bn.l the "error" word. An explanation of the error is given.

[Source Mihaela Sighireanu]

You have to look in the listing file tmp/*.la.l. An explanation for each ambiguity is given.

[Souce Mihaela Sighireanu]

Dans la description de la grammaire il faudra pouvoir specifier si on veut eliminer ou non les regles pour "null-...". Par exemple,

	    attribute grammar nnull X ( ... ) : ... >>

• To report a bug, please send an e-mail to Didier.Parigot@inria.fr (after consulting this FAQ list).
• For the users of the VASY team, please put Mihaela.Sighireanu@inria.fr in the CC of you message and fill the following form:

	Number:	_empty_
	Date:	_the_message_date_
	Source:	_your_name_
	Status: _no_for_the_moment_
	Nature:
		_the_bug_description_with_examples_