Parse::Eyapp::debuggingtut - Solving ambiguities and fixing lexical, syntactic and semantic errors
The sources of error when programming with eyapp are many and various. Some of them are minor, as having a nonterminal without production rules or a terminal that is never produced by the lexical analyzer. These kind of errors can be caught with the help of the %strict directive.
eyapp
%strict
In the following sections we will discuss three main kind of errors that correspond to three development stages:
Conflict errors:
Conflicts with the grammar: the grammar is ambiguous or is not clear - perhaps due to the fact that eyapp uses only a lookahead symbol - which sort of tree must be built for some inputs
Tree building errors:
There are no conflicts but the parser does not build the syntax tree as expected. May be it rejects correct sentences or accepts incorrect ones. Or may be it accepts correct ones but the syntax tree has not the shape we want (i.e. we have a precedence problem).
Semantic errors:
We have solved the conflicts and trees are satisfactory but we have errors inside the semantic actions.
Each time you discover an error write a test that covers that error. Section "TREE EQUALITY" deals with the problem of checking if the generated abstract syntax tree has the correct shape and attributes.
As Andreas Zeller points out in his article "Beautiful Debugging" finding the causes of a failing program must follow the scientific method:
-v
yydebug
By default, identifiers appearing in the rule section will be classified as terminal if they don't appear in the left hand side of any production rules.
The directive %strict forces the declaration of all tokens. The following eyapp program issues a warning:
pl@nereida:~/LEyapp/examples/eyapplanguageref$ cat -n bugyapp2.eyp 1 %strict 2 %% 3 expr: NUM; 4 %% pl@nereida:~/LEyapp/examples/eyapplanguageref$ eyapp bugyapp2.eyp Warning! Non declared token NUM at line 3 of bugyapp2.eyp
To keep silent the compiler declare all tokens using one of the token declaration directives (%token, %left, etc.)
%token
%left
pl@nereida:~/LEyapp/examples/eyapplanguageref$ cat -n bugyapp3.eyp 1 %strict 2 %token NUM 3 %% 4 expr: NUM; 5 %% pl@nereida:~/LEyapp/examples/eyapplanguageref$ eyapp bugyapp3.eyp pl@nereida:~/LEyapp/examples/eyapplanguageref$ ls -ltr | tail -1 -rw-r--r-- 1 pl users 2395 2008-10-02 09:41 bugyapp3.pm
It is a good practice to use %strict at the beginning of your grammar.
Token and production priorities are used to solve conflicts. Recall the main points of yacc-like parsers related to priorities:
The directives
%left %right %nonassoc
can be used in the head section to declare the priority of a token
The later the declaration line the higher the priority
The precedence of a production rule (right hand side) is the precedence of the last token in the right hand side
In a shift-reduce conflict the default action is to shift. This action can be changed if the production and the token have explicit priorities
If the precedence of the production rule is higher the shift-reduce conflict is solved in favor of the reduction
If the precedence of the token is higher the shift-reduce conflict is solved in favor of the shift
If the precedence of the token is the same than the precedence of the rule, and is left the shift-reduce conflict is solved in favor of the reduction
If the precedence of the token is the same than the precedence of the rule, and is right the shift-reduce conflict is solved in favor of the shift
If the precedence of the token is the same than the precedence of the rule, and is nonassoc the presence of a shift-reduce conflict means an error. This is used to describe operators, like the operator .LT. in FORTRAN, that may not associate with themselves. That is, because
.LT.
A .LT. B .LT. C
is invalid in FORTRAN, .LT. would be described with the keyword %nonassoc in eyapp.
%nonassoc
The default precedence of a production can be changed using the %prec TOKEN directive. Now the rule has the precedence and associativity of the specified TOKEN.
%prec TOKEN
TOKEN
The program Precedencia.eyp illustrates the way priorities work in eyapp:
Precedencia.eyp
pl@europa:~/LEyapp/examples/debuggingtut$ eyapp -c Precedencia.eyp %token NUM %left '@' %right '&' dummy %tree %% list: | list '\n' | list e ; e: %name NUM NUM | %name AMPERSAND e '&' e | %name AT e '@' e %prec dummy ; %%
See an execution:
pl@europa:~/LEyapp/examples/debuggingtut$ ./Precedencia.pm Expressions. Press CTRL-D (Unix) or CTRL-Z (Windows) to finish: 2@3@4 2@3&4 2&3@4 2&3&4 <CTRL-D> AT(AT(NUM(TERMINAL[2]),NUM(TERMINAL[3])),NUM(TERMINAL[4])) AT(NUM(TERMINAL[2]),AMPERSAND(NUM(TERMINAL[3]),NUM(TERMINAL[4]))) AT(AMPERSAND(NUM(TERMINAL[2]),NUM(TERMINAL[3])),NUM(TERMINAL[4])) AMPERSAND(NUM(TERMINAL[2]),AMPERSAND(NUM(TERMINAL[3]),NUM(TERMINAL[4])))
See if you are able to understand the output:
2@3@4: The phrase is interpreted as (2@3)@4 since the rule e '@' e has the precedence of the token dummy which is stronger that then priority of token @. The conflict is solved in favor of the reduction
2@3@4
(2@3)@4
e '@' e
dummy
@
2@3&4: The rule e '@' e has the precedence of dummy which is the same than the token &. The associativity decides. Since they were declared %right the conflict is solved in favor of the shift. The phrase is interpreted as 2@(3&4)
2@3&4
&
%right
2@(3&4)
2&3@4: The rule e '&' e has more precedence than the token @. The phrase is interpreted as (2&3)@4
2&3@4
e '&' e
(2&3)@4
2&3&4: Both the rule and the token have the same precedence. Since they were declared %right, the conflict is solved in favor of the shift. The phrase is interpreted as 2&(3&4)
2&3&4
2&(3&4)
The following simplified eyapp program has some errors. The generated language is made of lists of declarations (D stands for declaration) followed by lists of sentences (S stands for statement) separated by semicolons:
D
S
pl@nereida:~/LEyapp/examples/debuggingtut$ cat -n Debug.eyp 1 %{ 2 =head1 SYNOPSIS 3 4 This grammar has an unsolved shift-reduce conflict. 5 6 Be sure C<DebugTail.pm> is reachable. 7 Compile it with 8 9 eyapp -b '' Debug.eyp 10 11 See the C<Debug.output> file generated. 12 Execute the generated modulino with: 13 14 ./Debug.pm -d # to activate debugging 15 ./Debug.pm -h # for help 16 17 The generated parser will not recognize any input, since its shifts forever. 18 Try input C<'D; D; S'>. 19 20 =head1 See also 21 22 http://search.cpan.org/perldoc?Parse::Eyapp::debuggingtut 23 24 Debug1.eyp Debug2.eyp DebugLookForward.eyp DebugDynamicResolution.eyp 25 26 =cut 27 28 our $VERSION = '0.01'; 29 use base q{DebugTail}; 30 31 %} 32 33 %token D S 34 35 %% 36 p: 37 ds ';' ss 38 | ss 39 ; 40 41 ds: 42 D ';' ds 43 | D /* this production is never used */ 44 ; 45 46 ss: 47 S ';' ss 48 | S 49 ; 50 51 %% 52 53 __PACKAGE__->main('Provide a statement like "D; D; S" and press <CR><CTRL-D>: ') unless caller;
Sometimes the presence of actions, attribute names and support code makes more difficult the readability of the grammar. You can use the -c option of eyapp, to see only the syntactic parts:
-c
$ eyapp -c examples/debuggingtut/Debug.eyp %token D S %% p: ds ';' ss | ss ; ds: D ';' ds | D ; ss: S ';' ss | S ; $
It is clear now that the language generated by this grammar is made of non empty sequences of D followed by non empty sequences of <S> separated by semicolons.
When compiling this grammar, eyapp produces a warning message announcing the existence of a conflict:
pl@nereida:~/LEyapp/examples$ eyapp Debug.eyp 1 shift/reduce conflict (see .output file) State 4: shifts: to state 8 with ';'
.output
The existence of warnings triggers the creation of a file Debug.output containing information about the grammar and the syntax analyzer.
Debug.output
Let us see the contents of the Debug.output file:
pl@nereida:~/LEyapp/examples$ cat -n Debug.output 1 Warnings: 2 --------- 3 1 shift/reduce conflict (see .output file) 4 State 4: shifts: 5 to state 8 with ';' 6 7 Conflicts: 8 ---------- 9 State 4 contains 1 shift/reduce conflict 10 11 Rules: 12 ------ 13 0: $start -> p $end 14 1: p -> ds ';' ss 15 2: p -> ss 16 3: ds -> D ';' ds 17 4: ds -> D 18 5: ss -> S ';' ss 19 6: ss -> S 20 21 States: 22 ------- 23 State 0: 24 25 $start -> . p $end (Rule 0) 26 27 D shift, and go to state 4 28 S shift, and go to state 1 29 30 p go to state 2 31 ss go to state 3 32 ds go to state 5 33 .. ......................................... 55 State 4: 56 57 ds -> D . ';' ds (Rule 3) 58 ds -> D . (Rule 4) 59 60 ';' shift, and go to state 8 61 62 ';' [reduce using rule 4 (ds)] 63 .. ......................................... 84 State 8: 85 86 ds -> D ';' . ds (Rule 3) 87 88 D shift, and go to state 4 89 90 ds go to state 11 91 .. ......................................... 112 State 12: 113 114 p -> ds ';' ss . (Rule 1) 115 116 $default reduce using rule 1 (p) 117 118 119 Summary: 120 -------- 121 Number of rules : 7 122 Number of terminals : 4 123 Number of non-terminals : 4 124 Number of states : 13
The parser generated by Parse::Eyapp is based on a deterministic finite automaton. Each state of the automaton remembers what production rules are candidates to apply and what have been seen from the right hand side of the production rule. The problem, according to the warning, occurs in state 4. State 4 contains:
Parse::Eyapp
55 State 4: 56 57 ds -> D . ';' ds (Rule 3) 58 ds -> D . (Rule 4) 59 60 ';' shift, and go to state 8 61 62 ';' [reduce using rule 4 (ds)] 63
An state is a set of production rules with a marker (the dot in rules 3 and 4) somewhere in its right hand side. If the parser is in state 4 is because the production rules ds -> D ';' ds and ds -> D are potential candidates to build the syntax tree. That they will win or not depends on what will happen next when more input is processed.
ds -> D ';' ds
ds -> D
The dot that appears on the right hand side means position in our guessing. The fact that ds -> D .';' ds is in state 4 means that if the parser is in state 4 we have already seen D and we expect to see a semicolon followed by ds (or something derivable from ds). If such thing happens this production will be the right one (will be the handle in the jargon). The comment
ds -> D .';' ds
ds
60 ';' shift, and go to state 8
means that if the next token is a semicolon the next state will be state 8:
84 State 8: 85 86 ds -> D ';' . ds (Rule 3) 87 88 D shift, and go to state 4 89 90 ds go to state 11
As we see state 8 has the item ds -> D ';' . ds which means that we have already seen a D and a semicolon.
ds -> D ';' . ds
The fact that ds -> D . is in state 4 means that we have already seen D and since the dot is at the end of the rule, this production can be the right one, even if a semicolon is just waiting in the input. An example that it will be correct to "reduce" by the rule ds -> D . in the presence of a semicolon is given by the input D ; S. A rightmost derivation for such input is:
ds -> D .
D ; S
p => ds ; ss => ds ; S => D ; S
that is processed by the LALR(1) algorithm following this sequence of actions:
+----------+---------+---------+ | rule | read | input | | | | D ; S $ | | | D | ; S $ | | ds->d | ds | ; S $ | | | ds ; | S $ | | | ds ; S | $ | | ss->s | ds ; ss | $ | | p->ds;ss | p | | +----------+---------+---------+
Since it is correct to reduce in some cases by the production ds -> D . and others in which is correct to shift the semicolon, eyapp complains about a shift/reduce conflict with ';'. State 4 has two rules that compete to be the right one:
';'
pl@nereida:~/LEyapp/examples$ eyapp Debug.eyp 1 shift/reduce conflict (see .output file)
We can guess that the right item (the rules with the dot, i.e. the states of the automaton are called LALR(0) items in the yacc jargon) is ds -> D .';' ds and shift to state 8 consuming the semicolon, expecting to see something derivable from ds later or guess that ds -> D . is the right LR(0) item and reduce for such rule. This is the meaning of the comments in state 4:
60 ';' shift, and go to state 8 61 62 ';' [reduce using rule 4 (ds)]
To illustrate the problem let us consider the phrases D;S and D;D;S.
D;S
D;D;S
For both phrases, after consuming the D the parser will go to state 4 and the current token will be the semicolon.
For the first phrase D;S the correct decision is to use rule 4 ds -> D (to reduce in the jargon). For the second phrase D;D;S the correct decision is to follow rule 3 ds -> D . ';' ds.
ds -> D . ';' ds
The parser generated by eyapp would be able to know which rule is correct for each case if it were allowed to look at the token after the semicolon: if it is a S is rule 4, if it is a D is rule 3. But the parsers generated by Eyapp do not lookahead more than the next token (this is what the "1" means when we say that Parse::Eyapp parsers are LALR(1)) and therefore is not in condition to decide which production rule applies.
Eyapp
Unfortunately this is the sort of conflict that can't be solved by assigning priorities to the productions and tokens as it was done for the calculator example in Parse::Eyapp::eyappintro. If we run the analyzer it will refuse to accept correct entries like D;D;S:
pl@europa:~/LEyapp/examples/debuggingtut$ eyapp -b '' -o debug.pl Debug.eyp 1 shift/reduce conflict (see .output file) State 4: shifts: to state 8 with ';' pl@europa:~/LEyapp/examples/debuggingtut$ ./debug.pl D;D;S ---------------------------------------- In state 0: Stack:[0] Need token. Got >D< Shift and go to state 4. ---------------------------------------- In state 4: Stack:[0,4] Need token. Got >;< Shift and go to state 8. ---------------------------------------- In state 8: Stack:[0,4,8] Need token. Got >D< Shift and go to state 4. ---------------------------------------- In state 4: Stack:[0,4,8,4] Need token. Got >;< Shift and go to state 8. ---------------------------------------- In state 8: Stack:[0,4,8,4,8] Need token. Got >S< Syntax error near input: 'S' line num 1
The default parsing action is to shift the token ; giving priority to the production
;
over the production
Since no ds production starts with S, the presence of S is (erroneously) interpreted as an error.
You may wonder why the productions
ss: S ';' ss | S ;
do not also produce a shift-reduce conflict with the semicolon. This is because the reduction by ss -> S always corresponds to the last S in a derivation:
ss -> S
ss => S ; ss => S ; S ; ss => S ; S; S
and thus, the reduction by ss -> S only occurs in the presence of the end of input token and never with the semicolon. The FOLLOW set of a syntactic variable is the set of tokens that may appear next to such variable in some derivation. While the semicolon ; is in the FOLLOW of dd, it isn't in the FOLLOW of ss.
end of input
dd
ss
To solve the former conflict the Eyapp programmer has to reformulate the grammar modifying priorities and reorganizing the rules. Rewriting the recursive rule for ds to be let recursive solves the conflict:
pl@nereida:~/LEyapp/examples/debuggingtut$ sed -ne '/^ds:/,/^;/p' Debug1.eyp | cat -n 1 ds: 2 %name D2 3 ds ';' D 4 | %name D1 5 D 6 ;
Now, for any phrase matching the pattern D ; ... the action to build the tree is to reduce by ds -> D.
D ; ...
The rightmost reverse derivation for D;D;S is:
Derivation | Tree --------------------------------------+----------------------------- D;D;S <= ds;D;S <= ds;S <= ds;ss <= p | p(ds(ds(D),';',D),';',ss(S))
while the rightmost reverse derivation for D;S is:
Derivation | Tree --------------------------------------+----------------------------- D;S <= ds;S <= ds;ss <= p | p(ds(D),';',ss(S))
When we recompile the modified grammar no warnings appear:
pl@nereida:~/LEyapp/examples$ eyapp Debug1.eyp pl@nereida:~/LEyapp/examples$
The problem here is that Eyapp/Yapp/Yacc etc. produce LALR(1) parsers. They only look the next token. We can decide how to solve the conflict by rewriting the lexical analyzer to peer forward what token comes after the semicolon: it now returns SEMICOLONS if it is an S and SEMICOLOND if it is an D. Here is a solution based in this idea:
Eyapp/Yapp/Yacc
SEMICOLONS
SEMICOLOND
pl@nereida:~/LEyapp/examples/debuggingtut$ cat -n DebugLookForward.eyp 1 /*VIM: set ts=2 */ 2 %{ 3 =head1 SYNOPSIS 4 5 See 6 7 http://search.cpan.org/perldoc?Parse::Eyapp::debuggingtut 8 file DebugLookForward.eyp 9 10 This grammar fixes the conflicts an bugs in Debug.eyp and Debug1.eyp 11 12 Be sure C<DebugTail.pm> is reachable 13 compile it with 14 15 eyapp -b '' DebugLookForward.eyp 16 17 execute the generated modulino with: 18 19 ./DebugLookForward.pm -t 20 21 =head1 See also 22 23 Debug.eyp Debug1.eyp Debug2.eyp 24 25 =cut 26 27 our $VERSION = '0.01'; 28 use base q{DebugTail}; 29 30 %} 31 32 %token D S 33 %syntactic token SEMICOLONS SEMICOLOND 34 35 %tree 36 37 %% 38 p: 39 %name P 40 ds SEMICOLONS ss 41 | %name SS 42 ss 43 ; 44 45 ds: 46 %name D2 47 D SEMICOLOND ds 48 | %name D1 49 D 50 ; 51 52 ss: 53 %name S2 54 S SEMICOLONS ss 55 | %name S1 56 S 57 ; 58 59 %% 60 61 __PACKAGE__->lexer( 62 sub { 63 my $self = shift; 64 65 for (${$self->input()}) { 66 s{^(\s+)}{} and $self->tokenline($1 =~ tr{\n}{}); 67 return ('',undef) unless $_; 68 69 return ($1,$1) if s/^([sSDd])//; 70 return ('SEMICOLOND', 'SEMICOLOND') if s/^;\s*D/D/; 71 return ('SEMICOLONS', 'SEMICOLONS') if s/^;\s*S/S/; 72 die "Syntax error at line num ${$self->tokenline()}: ${substr($_,0,10)}\n"; 73 } 74 return ('',undef); 75 } 76 ); 77 78 __PACKAGE__->main unless caller();
Though Debug1.pm seems to work:
Debug1.pm
pl@nereida:~/LEyapp/examples/debuggingtut$ ./Debug1.pm -t Try first "D;S" and then "D; D; S" (press <CR><CTRL-D> to finish): D;D;S P(D2(D1(TERMINAL[D]),TERMINAL[D]),S1(TERMINAL[S]))
There are occasions where we observe an abnormal behavior:
pl@nereida:~/LEyapp/examples/debuggingtut$ ./Debug1.pm -t Try first "D;S" and then "D; D; S" (press <CR><CTRL-D> to finish): D ; D ; S Syntax error near end of input line num 3. Expecting (;)
We can activate the option yydebug => 0xF in the call to the parser method YYParser. The integer parameter yydebug of new and YYParse controls the level of debugging. Different levels of verbosity can be obtained by setting the bits of this argument. It works as follows:
yydebug => 0xF
YYParser
new
YYParse
/============================================================\ | Bit Value | Outputs | |------------+-----------------------------------------------| | 0x01 | Token reading (useful for Lexer debugging) | |------------+-----------------------------------------------| | 0x02 | States information | |------------+-----------------------------------------------| | 0x04 | Driver actions (shifts, reduces, accept...) | |------------+-----------------------------------------------| | 0x08 | Parse Stack dump | |------------+-----------------------------------------------| | 0x10 | Error Recovery tracing | \============================================================/
Let us see what happens when the input is D;S. We have introduced some white spaces and carriage returns between the terminals:
pl@nereida:~/LEyapp/examples/debuggingtut$ ./Debug1.pm -d Try first "D;S" and then "D; D; S" (press <CR><CTRL-D> to finish): D ; D ; S ---------------------------------------- In state 0: Stack:[0] Need token. Got >D< Shift and go to state 4. ---------------------------------------- In state 4: Stack:[0,4] Don't need token. Reduce using rule 4 (ds --> D): Back to state 0, then go to state 5. ---------------------------------------- In state 5: Stack:[0,5] Need token. Got >< Syntax error near end of input line num 3. Expecting (;)
What's going on? After reading the carriage return
Need token. Got >D<
the parser receives an end of file. ¿Why?. Something is going wrong in the communications between lexical analyzer and parser. Let us review the lexical analyzer:
pl@nereida:~/LEyapp/examples/debuggingtut$ sed -ne '/lexer/,/^)/p' Debug1.eyp | cat -n 1 __PACKAGE__->lexer( 2 sub { 3 my $self = shift; 4 5 for (${$self->input()}) { # contextualize 6 s{^(\s)}{} and $self->tokenline($1 =~ tr{\n}{}); 7 8 return ('', undef) unless $_; 9 return ($1, $1) if s/^(.)//; 10 } 11 return ('', undef); 12 } 13 );
The error is at line 6. Only a single white space is eaten! The second carraige return in the input does not match lines 8 and 9 and the contextualizing for finishes. Line 11 then unconditionally returns the ('',undef) signaling the end of input.
for
('',undef)
The grammar in file Debug2.eyp fixes the problem: Now the analysis seems to work for this kind of input:
Debug2.eyp
pl@nereida:~/LEyapp/examples/debuggingtut$ eyapp -b '' Debug2.eyp pl@nereida:~/LEyapp/examples/debuggingtut$ ./Debug2.pm -t -d Provide a statement like "D; D; S" and press <CR><CTRL-D>: D ; D ; S ---------------------------------------- In state 0: Stack:[0] Need token. Got >D< Shift and go to state 4. ---------------------------------------- In state 4: Stack:[0,4] Don't need token. Reduce using rule 4 (ds --> D): Back to state 0, then go to state 5. ---------------------------------------- In state 5: Stack:[0,5] Need token. Got >;< Shift and go to state 8. ---------------------------------------- In state 8: Stack:[0,5,8] Need token. Got >D< Shift and go to state 11. ---------------------------------------- In state 11: Stack:[0,5,8,11] Don't need token. Reduce using rule 3 (ds --> ds ; D): Back to state 0, then go to state 5. ---------------------------------------- In state 5: Stack:[0,5] Need token. Got >;< Shift and go to state 8. ---------------------------------------- In state 8: Stack:[0,5,8] Need token. Got >S< Shift and go to state 1. ---------------------------------------- In state 1: Stack:[0,5,8,1] Need token. Got >< Reduce using rule 6 (ss --> S): Back to state 8, then go to state 10. ---------------------------------------- In state 10: Stack:[0,5,8,10] Don't need token. Reduce using rule 1 (p --> ds ; ss): Back to state 0, then go to state 2. ---------------------------------------- In state 2: Stack:[0,2] Shift and go to state 7. ---------------------------------------- In state 7: Stack:[0,2,7] Don't need token. Accept. P(D2(D1(TERMINAL[D]),TERMINAL[D]),S1(TERMINAL[S]))
The YYParse methods implements the generic LR parsing algorithm. It very much works Parse::Yapp::YYParse and as yacc/bison yyparse. It accepts almost the same arguments as Class->new (Being Class the name of the generated class).
Parse::Yapp::YYParse
yyparse
Class->new
Class
The parser uses two tables and a stack. The two tables are called the action table and the goto table. The stack is used to keep track of the states visited.
At each step the generated parser consults the action table and takes one decision: To shift to a new state consuming one token (and pushing the current state in the stack) or to reduce by some production rule. In the last case the parser pops from its stack as many states as symbols are on the right hand side of the production rule. Here is a Perl/C like pseudocode summarizing the activity of YYParse:
action
1 my $parser = shift; # The parser object 2 push(@stack, $parser->{startstate}); 3 $b = $parser->YYLexer(); # Get the first token 4 FOREVER: { 5 $s = top(0); # Get the state on top of the stack 6 $a = $b; 7 switch ($parser->action[$s->state][$a]) { 8 case "shift t" : 9 my $t; 10 $t->{state} = t; 11 $t->{attr} = $a->{attr}; 12 push($t); 13 $b = $parser->YYLexer(); # Call the lexical analyzer 14 break; 15 case "reduce A->alpha" : 16 # Call the semantic action with the attributes of the rhs as args 17 my $semantic = $parser->Semantic{A ->alpha}; # The semantic action 18 my $r; 19 $r->{attr} = $semantic->($parser, top(|alpha|-1)->attr, ... , top(0)->attr); 20 21 # Pop as many states as symbols on the rhs of A->alpha 22 pop(|alpha|); 23 24 # Goto next state 25 $r->{state} = $parser->goto[top(0)][A]; 26 push($r); 27 break; 28 case "accept" : return (1); 29 default : $parser->YYError("syntax error"); 30 } 31 redo FOREVER; 32 }
Here |alpha| stands for the length of alpha. Function top(k) returns the state in position k from the top of the stack, i.e. the state at depth k. Function pop(k) extracts k states from the stack. The call $state->attr returns the attribute associated with $state. The call $parser->Semantic{A ->alpha} returns the semantic action associated with production A ->alpha.
|alpha|
alpha
top(k)
k
pop(k)
$state->attr
$state
$parser->Semantic{A ->alpha}
A ->alpha
Let us see a trace for the small grammar in examples/debuggingtut/aSb.yp:
examples/debuggingtut/aSb.yp
pl@nereida:~/LEyapp/examples$ /usr/local/bin/paste.pl aSb.yp aSb.output | head -5 %% | Rules: S: { print "S -> epsilon\n" } | ------ | 'a' S 'b' { print "S -> a S b\n" } | 0: $start -> S $end ; | 1: S -> /* empty */ %% | 2: S -> 'a' S 'b'
The tables in file aSb.output describe the actions and transitions to take:
aSb.output
pl@nereida:~/LEyapp/examples$ cat -n aSb.output . ......................................... 7 States: 8 ------- 9 State 0: 10 11 $start -> . S $end (Rule 0) 12 13 'a' shift, and go to state 2 14 15 $default reduce using rule 1 (S) 16 17 S go to state 1 18 19 State 1: 20 21 $start -> S . $end (Rule 0) 22 23 $end shift, and go to state 3 24 25 State 2: 26 27 S -> 'a' . S 'b' (Rule 2) 28 29 'a' shift, and go to state 2 30 31 $default reduce using rule 1 (S) 32 33 S go to state 4 34 35 State 3: 36 37 $start -> S $end . (Rule 0) 38 39 $default accept 40 41 State 4: 42 43 S -> 'a' S . 'b' (Rule 2) 44 45 'b' shift, and go to state 5 46 47 State 5: 48 49 S -> 'a' S 'b' . (Rule 2) 50 51 $default reduce using rule 2 (S) 52 53 54 Summary: 55 -------- 56 Number of rules : 3 57 Number of terminals : 3 58 Number of non-terminals : 2 59 Number of states : 6
When executed with yydebug set and input aabb we obtain the following output:
aabb
pl@nereida:~/LEyapp/examples/debuggingtut$ eyapp -b '' -o use_aSb.pl aSb pl@nereida:~/LEyapp/examples/debuggingtut$ ./use_aSb.pl -d Provide a statement like "a a b b" and press <CR><CTRL-D>: aabb ---------------------------------------- In state 0: Stack:[0] Need token. Got >a< Shift and go to state 2. ---------------------------------------- In state 2: Stack:[0,2] Need token. Got >a< Shift and go to state 2. ---------------------------------------- In state 2: Stack:[0,2,2] Need token. Got >b< Reduce using rule 1 (S --> /* empty */): S -> epsilon Back to state 2, then go to state 4. ---------------------------------------- In state 4: Stack:[0,2,2,4] Shift and go to state 5. ---------------------------------------- In state 5: Stack:[0,2,2,4,5] Don't need token. Reduce using rule 2 (S --> a S b): S -> a S b Back to state 2, then go to state 4. ----------------------------------------
As a result of reducing by rule 2 the three last visited states are popped from the stack, and the stack becomes [0,2]. But that means that we are now in state 2 seeing a S. If you look at the table above being in state 2 and seeing a S we go to state 4.
[0,2]
In state 4: Stack:[0,2,4] Need token. Got >b< Shift and go to state 5. ---------------------------------------- In state 5: Stack:[0,2,4,5] Don't need token. Reduce using rule 2 (S --> a S b): S -> a S b Back to state 0, then go to state 1. ---------------------------------------- In state 1: Stack:[0,1] Need token. Got >< Shift and go to state 3. ---------------------------------------- In state 3: Stack:[0,1,3] Don't need token. Accept.
A third type of error occurs when the code inside a semantic action doesn't behave as expected.
The semantic actions are translated in anonymous methods of the parser object. Since they are anonymous we can't use breakpoints as
b subname # stop when arriving at sub ''name''
or
c subname # contine up to reach sub ''name''
Furthermore the file loaded by the client program is the generated .pm. The code in the generated module Debug.pm is alien to us - Was automatically generated by Parse::Eyapp - and it can be difficult to find where our inserted semantic actions are.
.pm
Debug.pm
To watch the execution of a semantic action is simple: We use the debugger f file.eyp option to switch the viewing filename to our grammar file. The following session uses the example in the directory examples/Calculator:
f file.eyp
examples/Calculator
pl@nereida:~/LEyapp/examples/Calculator$ perl -wd scripts/calc.pl Loading DB routines from perl5db.pl version 1.3 Editor support available. Enter h or `h h' for help, or `man perldebug' for more help. main::(scripts/calc.pl:8): Math::Calc->main(); DB<1> f lib/Math/Calc.eyp 1 2 3 4 5 6 7 #line 8 "lib/Math/Calc.eyp" 8 9: use base q{Math::Tail}; 10: my %s; # symbol table
Lines 37 to 41 contain the semantic action associated with the production exp -> VAR (see file examples/Calculator/lib/Math/Calc.eyp):
exp -> VAR
examples/Calculator/lib/Math/Calc.eyp
DB<2> l 37,41 37: my $id = $VAR->[0]; 38: my $val = $s{$id}; 39: $_[0]->semantic_error("Accesing undefined variable $id at line $VAR->[1].\n") 40 unless defined($val); 41: return $val;
now we set a break at line 37, to see what happens when a non initialized variable is used:
DB<3> b 37
We issue now the command c (continue). The execution continues until line 37 of lib/Math/Calc.eyp is reached:
c
lib/Math/Calc.eyp
DB<4> c Expressions. Press CTRL-D (Unix) or CTRL-Z (Windows) to finish: a = 2+b # user input Math::Calc::CODE(0x191da98)(lib/Math/Calc.eyp:37): 37: my $id = $VAR->[0];
Now we can issue any debugger commands (like x, p, etc.) to investigate the internal state of our program and determine what are the reasons of any abnormal behavior.
x
p
DB<4> n Math::Calc::CODE(0x191da98)(lib/Math/Calc.eyp:38): 38: my $val = $s{$id}; DB<4> x $id 0 'b' DB<5> x %s empty array
Most of the time reduce-reduce conflicts are due to some ambiguity in the grammar, as it is the case for this minimal example:
pl@nereida:~/LEyapp/examples/debuggingtut$ sed -ne '/%%/,/%%/p' minimalrr.eyp | cat -n 1 %% 2 s: 3 %name S_is_a 4 'a' 5 | A 6 ; 7 A: 8 %name A_is_a 9 'a' 10 ; 11 12 %%
In case of a reduce-reduce conflict, Parse::Eyapp reduces using the first production in the text:
pl@nereida:~/LEyapp/examples/debuggingtut$ eyapp -b '' minimalrr.eyp 1 reduce/reduce conflict pl@nereida:~/LEyapp/examples/debuggingtut$ ./minimalrr.pm -t Try "a" and press <CR><CTRL-D>: a S_is_a
If we change the order of the productions
pl@nereida:~/LEyapp/examples/debuggingtut$ sed -ne '/%start/,40p' minimalrr2.eyp | cat -n 1 %start s 2 3 %% 4 A: 5 %name A_is_a 6 'a' 7 ; 8 9 s: 10 %name S_is_a 11 'a' 12 | %name A 13 A 14 ; 15 %%
the selected production changes:
pl@nereida:~/LEyapp/examples/debuggingtut$ eyapp -b '' minimalrr2 1 reduce/reduce conflict pl@nereida:~/LEyapp/examples/debuggingtut$ ./minimalrr2.pm -t Try "a" and press <CR><CTRL-D>: a A(A_is_a)
In this example the programmer has attempted to define a language made of mixed lists IDs and NUMbers :
ID
NUM
~/LEyapp/examples/debuggingtut$ eyapp -c typicalrr.eyp %token ID NUM %tree %% s: /* empty */ | s ws | s ns ; ns: /* empty */ | ns NUM ; ws: /* empty */ | ws ID ; %%
The grammar has several reduce-reduce conflicts:
~/LEyapp/examples/debuggingtut$ eyapp -b '' typicalrr.eyp 3 shift/reduce conflicts and 3 reduce/reduce conflicts
There are several sources of ambiguity in this grammar:
Statments like
NUM NUM NUM
are ambiguous. The following two left-most derivations exists:
s =*=> ns ns =*=> NUM NUM ns => NUM NUM NUM and s =*=> ns ns =*=> NUM ns =*=> NUM NUM NUM
the same with phrases like ID ID ID
ID ID ID
The empty word can be generated in many ways. For example:
s => empty
s => s ns => s empty => empty
etc.
The generated parser loops forever if feed with a list of identifiers:
~/LEyapp/examples/debuggingtut$ ./typicalrr.pm -d Try inputs "4 5", "a b" and "4 5 a b"(press <CR><CTRL-D> to finish): ab ---------------------------------------- In state 0: Stack:[0] Don't need token. Reduce using rule 1 (s --> /* empty */): Back to state 0, then go to state 1. ---------------------------------------- In state 1: Stack:[0,1] Need token. Got >ID< Reduce using rule 4 (ns --> /* empty */): Back to state 1, then go to state 2. ---------------------------------------- In state 2: Stack:[0,1,2] Reduce using rule 3 (s --> s ns): Back to state 0, then go to state 1. ---------------------------------------- In state 1: Stack:[0,1] Reduce using rule 4 (ns --> /* empty */): Back to state 1, then go to state 2. ---------------------------------------- In state 2: Stack:[0,1,2] Reduce using rule 3 (s --> s ns): Back to state 0, then go to state 1. ---------------------------------------- ^C
The problem is easily solved designing an equivalent non ambiguous grammar:
pl@europa:~/LEyapp/examples/debuggingtut$ cat -n correcttypicalrr.eyp 1 %token ID NUM 2 3 %% 4 s: 5 /* empty */ 6 | s ID 7 | s NUM 8 ; 9 10 %%
See also these files in the examples/debuggintut/ directory:
examples/debuggintut/
typicalrr2.eyp is equivalent to typicalrr.eyp but has %name directives, to have a nicer tree
typicalrr2.eyp
typicalrr.eyp
%name
typicalrr_fixed.eyp eliminates the ambiguity using a combination of priorities and elimination of the redundant empty productions. Explicit precedence via %prec directives are given to produce right recursive lists
typicalrr_fixed.eyp
%prec
typicalrr_fixed_rightrecursive.eyp is almost equal to typicalrr_fixed.eyp but eliminates the of %prec directives by making the list production right recursive
typicalrr_fixed_rightrecursive.eyp
We can also try to disambiguate the former example using priorities. For that we need to give an explicit priority to the end-of-input token. To refer to the end-of-input token in the header section, use the empty string ''. In the file examples/debuggingtut/typicalrrwithprec.eyp there is a priority based solution:
''
examples/debuggingtut/typicalrrwithprec.eyp
~/LEyapp/examples/debuggingtut$ eyapp -c typicalrrwithprec.eyp %right LNUM %right NUM %right ID %right '' # The string '' refers to the 'End of Input' token %tree bypass %% s: %name EMPTY /* empty */%prec '' | %name LIST s ws | %name LIST s ns ; ns: %name EMPTYNUM /* empty */%prec NUM | %name NUMS NUM ns ; ws: %name EMPTYID /* empty */%prec LNUM | %name IDS ID ws ; %%
Observe the use of %right '' in the header section: it gives a priority to the end-of-input token.
%right ''
~/LEyapp/examples/debuggingtut$ ./typicalrrwithprec.pm -t Try "4 5 a b 2 3" (press <CR><CTRL-D> to finish): 4 5 a b 2 3 ^D LIST( LIST( LIST( EMPTY, NUMS( TERMINAL[4], NUMS( TERMINAL[5], EMPTYNUM ) ) ), IDS( TERMINAL[a], IDS( TERMINAL[b], EMPTYID ) ) ), NUMS( TERMINAL[2], NUMS( TERMINAL[3], EMPTYNUM ) ) )
The grammar in file examples/debuggintut/pascalenumeratedvsrange.eyp:
examples/debuggintut/pascalenumeratedvsrange.eyp
~/LEyapp/examples/debuggingtut$ eyapp -c pascalenumeratedvsrange.eyp %token TYPE DOTDOT ID %left '+' '-' %left '*' '/' %% type_decl: TYPE ID '=' type ';' ; type: '(' id_list ')' | expr DOTDOT expr ; id_list: ID | id_list ',' ID ; expr: '(' expr ')' | expr '+' expr | expr '-' expr | expr '*' expr | expr '/' expr | ID ; %%
introduces a problem that arises in the declaration of enumerated and subrange types in Pascal:
type subrange = lo .. hi; type enum = (a, b, c);
The original language standard allows only numeric literals and constant identifiers for the subrange bounds (`lo' and `hi'), but Extended Pascal and many other Pascal implementations allow arbitrary expressions there. This gives rise to the following situation, containing a superfluous pair of parentheses:
type subrange = (a) .. b;
Compare this to the following declaration of an enumerated type with only one value:
type enum = (a);
These two declarations look identical until the .. token. With normal LALR(1) one-token look-ahead it is not possible to decide between the two forms when the identifier a is parsed. It is, however, desirable for a parser to decide this, since in the latter case a must become a new identifier to represent the enumeration value, while in the former case a must be evaluated with its current meaning, which may be a constant or even a function call.
..
a
The consequence is a reduce-reduce conflict, which is summarized in this state of the LALR automata:
State 10: id_list -> ID . (Rule 4) expr -> ID . (Rule 11) ')' [reduce using rule 11 (expr)] ')' reduce using rule 4 (id_list) ',' reduce using rule 4 (id_list) $default reduce using rule 11 (expr)
The grammar in file pascalenumeratedvsrangesolvedvialex.eyp solves this particular problem by looking ahead in the lexical analyzer: if the parenthesis is followed by a sequence of comma separated identifiers finished by the closing parenthesis and a semicolon we can conclude that is a enumerated type declaration. For more details, have a look at the file. Another solution using the postponed conflict resolution strategy can be found in file pascalenumeratedvsrangesolvedviadyn.eyp.
pascalenumeratedvsrangesolvedvialex.eyp
pascalenumeratedvsrangesolvedviadyn.eyp
Though not so common, it may occur that a reduce-reduce conflict is not due to ambiguity but to the limitations of the LALR(1) algorithm. The following example illustrates the point:
pl@europa:~/LEyapp/examples/debuggingtut$ cat -n rrconflictnamefirst.eyp 1 %token VAR ',' ':' 2 3 %{ 4 use base q{Tail}; 5 %} 6 7 %% 8 def: param_spec return_spec ',' 9 ; 10 param_spec: 11 type 12 | name_list ':' type 13 ; 14 return_spec: 15 type 16 | name ':' type 17 ; 18 name: VAR 19 ; 20 type: VAR 21 ; 22 name_list: 23 name 24 | name ',' name_list 25 ; 26 %% 27 28 __PACKAGE__->main unless caller();
This non ambiguous grammar generates a language of sequences like
a, b : e f : e,
The conflict is due to the final comma in:
def: param_spec return_spec ','
If you suppress such comma there is no conflict (try it). When compiling with eyapp we get the warning:
pl@europa:~/LEyapp/examples/debuggingtut$ eyapp rrconflictnamefirst.eyp 1 reduce/reduce conflict
Editing the .output file we can see the conflict is in state 2:
46 State 2: 47 48 name -> VAR . (Rule 6) 49 type -> VAR . (Rule 7) 50 51 ',' [reduce using rule 7 (type)] 52 VAR reduce using rule 7 (type) 53 $default reduce using rule 6 (name)
If we look at the grammar we can see that a reduction by
type -> VAR .
may occur with a comma as incoming token but only after the reduction by param_spec has taken place. The problem is that the automaton forgets about it. Look the automaton transitions in the .outputfile. By making explicit the difference between the first and second type we solve the conflict:
param_spec
.outputfile
type
pl@europa:~/LEyapp/examples/debuggingtut$ cat -n rrconflictnamefirst_fix1.eyp 1 %token VAR ',' ':' 2 3 %{ 4 use base q{Tail}; 5 %} 6 7 %% 8 def: param_spec return_spec ',' 9 ; 10 param_spec: 11 type 12 | name_list ':' type 13 ; 14 return_spec: 15 typeafter 16 | name ':' typeafter 17 ; 18 name: VAR 19 ; 20 type: VAR 21 ; 22 typeafter: VAR 23 ; 24 name_list: 25 name 26 | name ',' name_list 27 ; 28 %% 29 30 __PACKAGE__->main unless caller();
A reduce-reduce conflict is solved in favor of the first production found in the text. If we execute the grammar with the conflict ./rrconflictnamefirst.pm, we get the correct behavior:
./rrconflictnamefirst.pm
pl@europa:~/LEyapp/examples/debuggingtut$ eyapp -b '' rrconflictnamefirst.eyp 1 reduce/reduce conflict pl@europa:~/LEyapp/examples/debuggingtut$ ./rrconflictnamefirst.pm Expressions. Press CTRL-D (Unix) or CTRL-Z (Windows) to finish: a,b:c d:e, <CTRL-D> $
The program accepts the correct language - in spite of the conflict - due to the fact that the production
name: VAR
is listed first.
The parser rejects the correct phrases if we swap the order of the productions writing the type: VAR production first,
type: VAR
pl@europa:~/LEyapp/examples/debuggingtut$ ./reducereduceconflict.pm Expressions. Press CTRL-D (Unix) or CTRL-Z (Windows) to finish: a,b:c d:e, <CTRL-D> Syntax error near input: ',' (lin num 1). Incoming text: === b:c d === Expected one of these terminals: VAR
Files reducereduceconflict_fix1.eyp and reducereduceconflict_fix2.eyp offer other solutions to the problem.
reducereduceconflict_fix1.eyp
reducereduceconflict_fix2.eyp
Usually there is a one-to-one relation between a token and a regexp. Problems arise, however when a token's type depends upon contextual information. An example of this problem comes from PL/I, where statements like this are legal:
if then=if then if=then
In PL/I this problem arises because keywords like if are not reserved and can be used in other contexts. This simplified grammar illustrates the problem:
if
examples/debuggingtut$ eyapp -c PL_I_conflict.eyp # This grammar deals with the famous ambiguous PL/I phrase: # if then=if then if=then # The (partial) solution uses YYExpect in the lexical analyzer to predict the token # that fulfills the parser expectatives. # Compile it with: # eyapp -b '' PL_I_conflict.eyp # Run it with; # ./PL_I_conflict.pm -debug %strict %token ID %tree bypass %% stmt: ifstmt | assignstmt ; # Exercise: change this production # for 'if' expr 'then' stmt # and check with input 'if then=if then if=then'. The problem arises again ifstmt: %name IF 'if' expr 'then' expr ; assignstmt: id '=' expr ; expr: %name EQ id '=' id | id ; id: %name ID ID ; %%
If the token ambiguity depends only in the syntactic context, the problem can be alleviated using the YYExpect method. In case of doubt, the lexical analyzer calls the YYExpect method to know which of the several feasible tokens is expected by the parser:
YYExpect
examples/debuggingtut$ sed -ne '/sub lex/,/^}/p' PL_I_conflict.eyp sub lexer { my $parser = shift; for ($parser->{input}) { # contextualize m{\G\s*(\#.*)?}gc; m{\G([a-zA-Z_]\w*)}gc and do { my $id = $1; return ('if', 'if') if ($id eq 'if') && is_in('if', $parser->YYExpect); return ('then', 'then') if ($id eq 'then') && is_in('then', $parser->YYExpect); return ('ID', $id); }; m{\G(.)}gc and return ($1, $1); return('',undef); } }
Here follows an example of execution:
examples/debuggingtut$ eyapp -b '' PL_I_conflict.eyp examples/debuggingtut$ ./PL_I_conflict.pm Expressions. Press CTRL-D (Unix) or CTRL-Z (Windows) to finish: if then=if then if=then IF(EQ(ID,ID),EQ(ID,ID))
A lexical tie-in is a flag which is set to alter the behavior of the lexical analyzer. It is a way to handle context-dependency.
C
The C language has a context dependency: the way an identifier is used depends on what its current meaning is. For example, consider this:
T(x);
This looks like a function call statement, but if T is a typedef name, then this is actually a declaration of x. How can a parser for C decide how to parse this input?
T
Here is another example:
{ T * x; ... }
What is this, a declaration of x as a pointer to T, or a void multiplication of the variables T and x?
The usual method to solve this problem is to have two different token types, ID and TYPENAME. When the lexical analyzer finds an identifier, it looks up in the symbol table the current declaration of the identifier in order to decide which token type to return: TYPENAME if the identifier is declared as a typedef, ID otherwise. See the ANSI C parser example in the directory examples/languages/C/ansic.eyp
TYPENAME
examples/languages/C/ansic.eyp
In the "Calc"-like example in examples/debuggintut/SemanticInfoInTokens.eyp we have a language with a special construct hex (hex-expr). After the keyword hex comes an expression in parentheses in which all integers are hexadecimal. In particular, strings in /[A-F0-9]+/ like A1B must be treated as an hex integer unless they were previously declared as variables. Let us see an example of execution:
examples/debuggintut/SemanticInfoInTokens.eyp
hex (hex-expr)
hex
expression
/[A-F0-9]+/
A1B
$ eyapp -b '' SemanticInfoInTokens.eyp $ cat inputforsemanticinfo2.txt int A2 A2 = HEX(A23); A2 = HEX(A2) $ ./SemanticInfoInTokens.pm -f inputforsemanticinfo2.txt -t EXPS(ASSIGN(TERMINAL[A2],NUM[2595]),ASSIGN(TERMINAL[A2],ID[A2]))
The first hex expression HEX(A23) is interpreted as the number 2595 while the second HEX(A2) refers to previously declared variable A2.
HEX(A23)
2595
HEX(A2)
A2
An alternative solution to this problem that does not make use of lexical tie-ins - but still uses an attribute HEXFLAG for communication between different semantic actions - can be found in the file examples/debuggintut/Tieins.eyp.
HEXFLAG
examples/debuggintut/Tieins.eyp
pl@nereida:~/LEyapp/examples/debuggingtut$ sed -ne '5,91p' SemanticInfoInTokens.eyp | cat -n 1 %strict 2 3 %token ID INT INTEGER 4 %syntactic token HEX 5 6 %right '=' 7 %left '+' 8 9 %{ 10 use base q{DebugTail}; 11 my %st; 12 %} 13 14 %tree bypass alias 15 16 %% 17 stmt: 18 decl <* ';'> expr <%name EXPS + ';'> 19 { 20 # make the symbol table an attribute 21 # of the root node 22 $_[2]->{st} = { %st }; 23 $_[2]; 24 } 25 ; 26 27 decl: 28 INT ID <+ ','> 29 { 30 # insert identifiers in the symbol table 31 $st{$_->{attr}} = 1 for $_[2]->children(); 32 } 33 ; 34 35 expr: 36 %name ID 37 ID 38 | %name NUM 39 INTEGER 40 | %name HEX 41 HEX '(' { $_[0]->{HEXFLAG} = 1; } $expr ')' 42 { 43 $_[0]->{HEXFLAG} = 0; 44 $expr; 45 } 46 | %name ASSIGN 47 id '=' expr 48 | %name PLUS 49 expr '+' expr 50 ; 51 52 id : ID 53 ; 54 55 %% 56 57 # Context-dependant lexer 58 __PACKAGE__->lexer( sub { 59 my $parser = shift; 60 my $hexflag = $parser->{HEXFLAG}; 61 62 for (${$parser->input}) { # contextualize 63 m{\G\s*(\#.*)?}gc; 64 65 m{\G(HEX\b|INT\b)}igc and return (uc($1), $1); 66 67 m{(\G\d+)}gc and return ('INTEGER', $hexflag? hex($1) : $1); 68 69 70 m{\G([a-zA-Z_]\w*)}gc and do { 71 my $match = $1; 72 $hexflag and !exists($st{$match}) and $match =~ m{^([A-F0-9]+$)}gc and return ('INTEGER', hex($match)); 73 return ('ID', $1); 74 }; 75 76 m{\G(.)}gc and return ($1, $1); 77 78 return('',undef); 79 } 80 } 81 ); 82 83 *TERMINAL::info = *NUM::info = *ID::info = sub { 84 $_[0]->{attr} 85 }; 86 87 __PACKAGE__->main() unless caller();
Here the lexical analyzer looks at the value of the attribute HEXFLAG; when it is nonzero, all integers are parsed in hexadecimal, and tokens starting with letters are parsed as integers if possible.
For more about lexical tie-ins see also
http://www.gnu.org/software/bison/manual/html_mono/bison.html#Lexical-Tie_002dins
http://en.wikipedia.org/wiki/The_lexer_hack
http://eli.thegreenplace.net/2007/11/24/the-context-sensitivity-of-cs-grammar/
Yacc-like parser generators provide ways to solve shift-reduce mechanims based on token precedence. No mechanisms are provided for the resolution of reduce-reduce conflicts. The solution for such kind of conflicts is to modify the grammar. The strategy We present here provides a way to broach conflicts that can't be solved using static precedences.
The postponed conflict strategy presented here can be used whenever there is a shift-reduce or reduce-reduce conflict that can not be solved using static precedences.
Let us assume we have a reduce-reduce conflict between to productions
A -> alpha . B -> beta .
for some token @. Let also assume that production
A -> alpha
has name ruleA and production
ruleA
B -> beta
has name ruleB.
ruleB
The postponed conflict resolution strategy consists in modifying the conflictive grammar by marking the points where the conflict occurs with the new %PREC directive. In this case at then end of the involved productions:
%PREC
A -> alpha %PREC IsAorB B -> beta $PREC IsAorB
The IsAorB identifier is called the conflict name.
IsAorB
Inside the head section, the programmer associates with the conflict name a code whose mission is to solve the conflict by dynamically changing the parsing table like this:
%conflict IsAorB { my $self = shift; if (looks_like_A($self)) { $self->YYSetReduce('@', 'ruleA' ); } else { $self->YYSetReduce('@', 'ruleB' ); } }
The code associated with the conflict name receives the name of conflict handler. The code of looks_like_A stands for some form of nested parsing which will decide which production applies.
looks_like_A
In file pascalenumeratedvsrangesolvedviadyn.eyp we apply the postponed conflict resolution strategy to the reduce reduce conflict that arises in Extended Pascal between the declaration of ranges and the declaration of enumerated types (see section "Reduce-Reduce conflict: Enumerated versus Range declarations in Extended Pascal"). Here is the solution:
~/LEyapp/examples/debuggingtut$ cat -n pascalenumeratedvsrangesolvedviadyn.eyp 1 %{ 2 =head1 SYNOPSIS 3 4 See 5 6 =over 2 7 8 =item * File pascalenumeratedvsrange.eyp in examples/debuggintut/ 9 10 =item * The Bison manual L<http://www.gnu.org/software/bison/manual/html_mono/bison.html> 11 12 =back 13 14 Compile it with: 15 16 eyapp -b '' pascalenumeratedvsrangesolvedviadyn.eyp 17 18 run it with this options: 19 20 ./pascalenumeratedvsrangesolvedviadyn.pm -t 21 22 Try these inputs: 23 24 type r = (x) .. y ; 25 type r = (x+2)*3 .. y/2 ; 26 type e = (x, y, z); 27 type e = (x); 28 29 =cut 30 31 use base q{DebugTail}; 32 33 my $ID = qr{[A-Za-z][A-Za-z0-9_]*}; 34 # Identifiers separated by commas 35 my $IDLIST = qr{ \s*(?:\s*,\s* $ID)* \s* }x; 36 # list followed by a closing par and a semicolon 37 my $RESTOFLIST = qr{$IDLIST \) \s* ; }x; 38 %} 39 40 %namingscheme { 41 #Receives a Parse::Eyapp object describing the grammar 42 my $self = shift; 43 44 $self->tokennames( 45 '(' => 'LP', 46 '..' => 'DOTDOT', 47 ',' => 'COMMA', 48 ')' => 'RP', 49 '+' => 'PLUS', 50 '-' => 'MINUS', 51 '*' => 'TIMES', 52 '/' => 'DIV', 53 ); 54 55 # returns the handler that will give names 56 # to the right hand sides 57 \&give_rhs_name; 58 } 59 60 %strict 61 62 %token ID NUM DOTDOT TYPE 63 %left '-' '+' 64 %left '*' '/' 65 66 %tree 67 68 %% 69 70 type_decl : TYPE ID '=' type ';' 71 ; 72 73 type : 74 %name ENUM 75 '(' id_list ')' 76 | %name RANGE 77 expr DOTDOT expr 78 ; 79 80 id_list : 81 %name EnumID 82 ID rangeORenum 83 | id_list ',' ID 84 ; 85 86 expr : '(' expr ')' 87 | expr '+' expr 88 | expr '-' expr 89 | expr '*' expr 90 | expr '/' expr 91 | %name RangeID 92 ID rangeORenum 93 | NUM 94 ; 95 96 rangeORenum: /* empty: postponed conflict resolution */ 97 { 98 my $parser = shift; 99 if (${$parser->input()} =~ m{\G(?= $RESTOFLIST)}gcx) { 100 $parser->YYSetReduce(')', 'EnumID' ); 101 } 102 else { 103 $parser->YYSetReduce(')', 'RangeID' ); 104 } 105 } 106 ; 107 108 %% 109 110 __PACKAGE__->lexer( 111 sub { 112 my $parser = shift; 113 114 for (${$parser->input()}) { # contextualize 115 m{\G(\s*)}gc; 116 $parser->tokenline($1 =~ tr{\n}{}); 117 118 m{\Gtype\b}gic and return ('TYPE', 'TYPE'); 119 120 m{\G($ID)}gc and return ('ID', $1); 121 122 m{\G([0-9]+)}gc and return ('NUM', $1); 123 124 m{\G\.\.}gc and return ('DOTDOT', '..'); 125 126 m{\G(.)}gc and return ($1, $1); 127 128 return('',undef); 129 } 130 } 131 ); 132 133 unless (caller()) { 134 $Parse::Eyapp::Node::INDENT = 1; 135 my $prompt = << 'EOP'; 136 Try this input: 137 type 138 r 139 = 140 (x) 141 .. 142 y 143 ; 144 145 Here other inputs you can try: 146 147 type r = (x+2)*3 .. y/2 ; 148 type e = (x, y, z); 149 type e = (x); 150 151 Press CTRL-D (CTRL-W in windows) to produce the end-of-file 152 EOP 153 __PACKAGE__->main($prompt); 154 }
This example also illustrates how to modify the default production naming schema. Follows the result of several executions:
~/LEyapp/examples/debuggingtut$ ./pascalenumeratedvsrangesolvedviadyn.pm -t Try this input: type r = (x) .. y ; Here other inputs you can try: type r = (x+2)*3 .. y/2 ; type e = (x, y, z); type e = (x); Press CTRL-D (CTRL-W in windows) to produce the end-of-file type r = (x+2)*3 .. y/2 ; ^D type_decl_is_TYPE_ID_type( TERMINAL[TYPE], TERMINAL[r], RANGE( expr_is_expr_TIMES_expr( expr_is_LP_expr_RP( expr_is_expr_PLUS_expr( RangeID( TERMINAL[x] ), expr_is_NUM( TERMINAL[2] ) ) ), expr_is_NUM( TERMINAL[3] ) ), TERMINAL[..], expr_is_expr_DIV_expr( RangeID( TERMINAL[y] ), expr_is_NUM( TERMINAL[2] ) ) ) ) ~/LEyapp/examples/debuggingtut$ ./pascalenumeratedvsrangesolvedviadyn.pm -t Try this input: type r = (x) .. y ; Here other inputs you can try: type r = (x+2)*3 .. y/2 ; type e = (x, y, z); type e = (x); Press CTRL-D (CTRL-W in windows) to produce the end-of-file type e = (x); ^D type_decl_is_TYPE_ID_type( TERMINAL[TYPE], TERMINAL[e], ENUM( EnumID( TERMINAL[x] ) ) )
The program in examples/debuggingtut/DynamicallyChangingTheParser2.eyp illustrates how the postponed conflict strategy is used for shift-reduce conflicts. This is an extension of the grammar in examples/debuggingtut/Debug.eyp. The generated language is constituted by sequences like:
examples/debuggingtut/DynamicallyChangingTheParser2.eyp
examples/debuggingtut/Debug.eyp
{ D; D; S; S; S; } {D; S} { S }
As you remember the conflict was:
~/LEyapp/examples/debuggingtut$ sed -ne '/^St.*13:/,/^St.*14/p' DynamicallyChangingTheParser.output State 13: ds -> D conflict . ';' ds (Rule 6) ds -> D conflict . (Rule 7) ';' shift, and go to state 16 ';' [reduce using rule 7 (ds)] State 14:
The conflict handler below sets the LR action to reduce by the production with name D1
conflict
D1
in the presence of token ';' if indeed is the last 'D', that is, if:
'D'
${$self->input()} =~ m{^\s*;\s*S}
Otherwise we set the shift action via a call to the YYSetShift method.
shift
YYSetShift
~/LEyapp/examples/debuggingtut$ sed -ne '30,$p' DynamicallyChangingTheParser.eyp | cat -n 1 %token D S 2 3 %tree bypass 4 5 # Expect just 1 shift-reduce conflict 6 %expect 1 7 8 %% 9 p: %name PROG 10 block + 11 ; 12 13 block: 14 %name BLOCK_DS 15 '{' ds ';' ss '}' 16 | %name BLOCK_S 17 '{' ss '}' 18 ; 19 20 ds: 21 %name D2 22 D conflict ';' ds 23 | %name D1 24 D conflict 25 ; 26 27 ss: 28 %name S2 29 S ';' ss 30 | %name S1 31 S 32 ; 33 34 conflict: 35 /* empty. This action solves the conflict using dynamic precedence */ 36 { 37 my $self = shift; 38 39 if (${$self->input()} =~ m{^\s*;\s*S}) { 40 $self->YYSetReduce(';', 'D1' ) 41 } 42 else { 43 $self->YYSetShift(';') 44 } 45 46 undef; # skip this node in the AST 47 } 48 ; 49 50 %% 51 52 my $prompt = 'Provide a statement like "{D; S} {D; D; S}" and press <CR><CTRL-D>: '; 53 __PACKAGE__->main($prompt) unless caller;
The more the time invested writing tests the less the time spent debugging. This section deals with the Parse::Eyapp::Node method equal which can be used to test that the trees have the shape we expect.
equal
A call $tree1->equal($tree2) compare the two trees $tree1 and $tree2. Two trees are considered equal if their root nodes belong to the same class, they have the same number of children and the children are (recursively) equal.
$tree1->equal($tree2)
$tree1
$tree2
In Addition to the two trees the programmer can specify pairs attribute_key => equality_handler:
attribute_key => equality_handler
$tree1->equal($tree2, attr1 => \&handler1, attr2 => \&handler2, ...)
In such case the definition of equality is more restrictive: Two trees are considered equal if
Their root nodes belong to the same class,
They have the same number of children
For each of the specified attributes occur that for both nodes the existence and definition of the key is the same
Assuming the key exists and is defined for both nodes, the equality handlers return true for each of its attributes and
The respective children are (recursively) equal.
An attribute handler receives as arguments the values of the attributes of the two nodes being compared and must return true if, and only if, these two attributes are considered equal. Follows an example:
examples/Node$ cat -n equal.pl 1 #!/usr/bin/perl -w 2 use strict; 3 use Parse::Eyapp::Node; 4 5 my $string1 = shift || 'ASSIGN(VAR(TERMINAL))'; 6 my $string2 = shift || 'ASSIGN(VAR(TERMINAL))'; 7 my $t1 = Parse::Eyapp::Node->new($string1, sub { my $i = 0; $_->{n} = $i++ for @_ }); 8 my $t2 = Parse::Eyapp::Node->new($string2); 9 10 # Without attributes 11 if ($t1->equal($t2)) { 12 print "\nNot considering attributes: Equal\n"; 13 } 14 else { 15 print "\nNot considering attributes: Not Equal\n"; 16 } 17 18 # Equality with attributes 19 if ($t1->equal($t2, n => sub { return $_[0] == $_[1] })) { 20 print "\nConsidering attributes: Equal\n"; 21 } 22 else { 23 print "\nConsidering attributes: Not Equal\n"; 24 }
When the former program is run without arguments produces the following output:
examples/Node$ equal.pl Not considering attributes: Equal Considering attributes: Not Equal
During the development of your compiler you add new stages to the existing ones. The consequence is that the AST is decorated with new attributes. Unfortunately, this implies that tests you wrote using is_deeply and comparisons against formerly correct abstract syntax trees are no longer valid. This is due to the fact that is_deeply requires both tree structures to be equivalent in every detail and that our new code produces a tree with new attributes.
is_deeply
Instead of is_deeply use the equal method to check for partial equivalence between abstract syntax trees. You can follow these steps:
Dump the tree for the source inserting Data::Dumper statements
Data::Dumper
Carefully check that the tree is really correct
Decide which attributes will be used for comparison
Write the code for the expected value editing the output produced by Data::Dumper
Write the handlers for the attributes you decided. Write the comparison using equal.
Tests using this methodology will not fail even if later code decorating the AST with new attributes is introduced.
See an example that checks an abstract syntax tree produced by the simple compiler (see examples/typechecking/Simple-Types-XXX.tar.gz) for a really simple source:
examples/typechecking/Simple-Types-XXX.tar.gz
Simple-Types/script$ cat prueba27.c int f() { }
The first thing is to obtain a description of the tree, that can be done executing the compiler under the control of the Perl debugger, stopping just after the tree has been built and dumping the tree with Data::Dumper:
pl@nereida:~/Lbook/code/Simple-Types/script$ perl -wd usetypes.pl prueba27.c main::(usetypes.pl:5): my $filename = shift || die "Usage:\n$0 file.c\n"; DB<1> c 12 main::(usetypes.pl:12): Simple::Types::show_trees($t, $debug); DB<2> use Data::Dumper DB<3> $Data::Dumper::Purity = 1 DB<4> p Dumper($t) $VAR1 = bless( { .............................................. }, 'PROGRAM' ); ...............................................................
Once we have the shape of a correct tree we can write our tests:
examples/Node$ cat -n testequal.pl 1 #!/usr/bin/perl -w 2 use strict; 3 use Parse::Eyapp::Node; 4 use Data::Dumper; 5 use Data::Compare; 6 7 my $debugging = 0; 8 9 my $handler = sub { 10 print Dumper($_[0], $_[1]) if $debugging; 11 Compare($_[0], $_[1]) 12 }; 13 14 my $t1 = bless( { 15 'types' => { 16 'CHAR' => bless( { 'children' => [] }, 'CHAR' ), 17 'VOID' => bless( { 'children' => [] }, 'VOID' ), 18 'INT' => bless( { 'children' => [] }, 'INT' ), 19 'F(X_0(),INT)' => bless( { 20 'children' => [ 21 bless( { 'children' => [] }, 'X_0' ), 22 bless( { 'children' => [] }, 'INT' ) ] 23 }, 'F' ) 24 }, 25 'symboltable' => { 'f' => { 'type' => 'F(X_0(),INT)', 'line' => 1 } }, 26 'lines' => 2, 27 'children' => [ 28 bless( { 29 'symboltable' => {}, 30 'fatherblock' => {}, 31 'children' => [], 32 'depth' => 1, 33 'parameters' => [], 34 'function_name' => [ 'f', 1 ], 35 'symboltableLabel' => {}, 36 'line' => 1 37 }, 'FUNCTION' ) 38 ], 39 'depth' => 0, 40 'line' => 1 41 }, 'PROGRAM' ); 42 $t1->{'children'}[0]{'fatherblock'} = $t1; 43 44 # Tree similar to $t1 but without some attributes (line, depth, etc.) 45 my $t2 = bless( { 46 'types' => { 47 'CHAR' => bless( { 'children' => [] }, 'CHAR' ), 48 'VOID' => bless( { 'children' => [] }, 'VOID' ), 49 'INT' => bless( { 'children' => [] }, 'INT' ), 50 'F(X_0(),INT)' => bless( { 51 'children' => [ 52 bless( { 'children' => [] }, 'X_0' ), 53 bless( { 'children' => [] }, 'INT' ) ] 54 }, 'F' ) 55 }, 56 'symboltable' => { 'f' => { 'type' => 'F(X_0(),INT)', 'line' => 1 } }, 57 'children' => [ 58 bless( { 59 'symboltable' => {}, 60 'fatherblock' => {}, 61 'children' => [], 62 'parameters' => [], 63 'function_name' => [ 'f', 1 ], 64 }, 'FUNCTION' ) 65 ], 66 }, 'PROGRAM' ); 67 $t2->{'children'}[0]{'fatherblock'} = $t2; 68 69 # Tree similar to $t1 but without some attributes (line, depth, etc.) 70 # and without the symboltable and types attributes used in the comparison 71 my $t3 = bless( { 72 'types' => { 73 'CHAR' => bless( { 'children' => [] }, 'CHAR' ), 74 'VOID' => bless( { 'children' => [] }, 'VOID' ), 75 'INT' => bless( { 'children' => [] }, 'INT' ), 76 'F(X_0(),INT)' => bless( { 77 'children' => [ 78 bless( { 'children' => [] }, 'X_0' ), 79 bless( { 'children' => [] }, 'INT' ) ] 80 }, 'F' ) 81 }, 82 'children' => [ 83 bless( { 84 'symboltable' => {}, 85 'fatherblock' => {}, 86 'children' => [], 87 'parameters' => [], 88 'function_name' => [ 'f', 1 ], 89 }, 'FUNCTION' ) 90 ], 91 }, 'PROGRAM' ); 92 93 $t3->{'children'}[0]{'fatherblock'} = $t2; 94 95 # Without attributes 96 if (Parse::Eyapp::Node::equal($t1, $t2)) { 97 print "\nNot considering attributes: Equal\n"; 98 } 99 else { 100 print "\nNot considering attributes: Not Equal\n"; 101 } 102 103 # Equality with attributes 104 if (Parse::Eyapp::Node::equal( 105 $t1, $t2, 106 symboltable => $handler, 107 types => $handler, 108 ) 109 ) { 110 print "\nConsidering attributes: Equal\n"; 111 } 112 else { 113 print "\nConsidering attributes: Not Equal\n"; 114 } 115 116 # Equality with attributes 117 if (Parse::Eyapp::Node::equal( 118 $t1, $t3, 119 symboltable => $handler, 120 types => $handler, 121 ) 122 ) { 123 print "\nConsidering attributes: Equal\n"; 124 } 125 else { 126 print "\nConsidering attributes: Not Equal\n"; 127 }
The code defining tree $t1 was obtained from an output using Data::Dumper. The code for trees $t2 and $t3 was written using cut-and-paste from $t1. They have the same shape than $t1 but differ in their attributes. Tree $t2 shares with $t1 the attributes symboltable and types used in the comparison and so equal returns true when compared. Since $t3 differs from $t1 in the attributes symboltable and types the call to equal returns false.
$t1
$t2
$t3
symboltable
types
true
false
I use these rules for indenting Parse::Eyapp programs:
Use uppercase identifiers for tokens, lowercase identifiers for syntactic variables
The syntactic variable that defines the rule must be at in a single line at the leftmost position:
synvar: 'a' othervar 'c' | 'b' anothervar SOMETOKEN ;
The separation bar | goes indented relative to the left side of the rule. Each production starts two spaces from the bar. The first right hand side is aligned with the rest.
|
The semicolon ; must also be in its own line at column 0
If there is an empty production it must be the first one and must be commented
syntacvar: /* empty */ | 'a' othervar 'c' | 'b' anothervar ;
Only very short semantic actions can go in the same line than the production. Semantic actions requiring more than one line must go in its own indented block like in:
exp: $NUM { $NUM->[0] } | $VAR { my $id = $VAR->[0]; my $val = $s{$id}; $_[0]->semantic_error("Accessing undefined variable $id at line $VAR->[1].\n") unless defined($val); return $val; } | $VAR '=' $exp { $s{$VAR->[0]} = $exp } | exp.x '+' exp.y { $x + $y } | exp.x '-' exp.y { $x - $y } | exp.x '*' exp.y { $x * $y } | exp.x '/'.barr exp.y { return($x/$y) if $y; $_[0]->semantic_error("Illegal division by zero at line $barr->[1].\n"); undef } | '-' $exp %prec NEG { -$exp } | exp.x '^' exp.y { $x ** $y } | '(' $exp ')' { $exp } ;
The project home is at http://code.google.com/p/parse-eyapp/. Use a subversion client to anonymously check out the latest project source code:
svn checkout http://parse-eyapp.googlecode.com/svn/trunk/ parse-eyapp-read-only
The tutorial Parsing Strings and Trees with Parse::Eyapp (An Introduction to Compiler Construction in seven pages) in http://nereida.deioc.ull.es/~pl/eyapsimple/
Parse::Eyapp, Parse::Eyapp::eyapplanguageref, Parse::Eyapp::debuggingtut, Parse::Eyapp::defaultactionsintro, Parse::Eyapp::translationschemestut, Parse::Eyapp::Driver, Parse::Eyapp::Node, Parse::Eyapp::YATW, Parse::Eyapp::Treeregexp, Parse::Eyapp::Scope, Parse::Eyapp::Base, Parse::Eyapp::datagenerationtut
The pdf file in http://nereida.deioc.ull.es/~pl/perlexamples/languageintro.pdf
The pdf file in http://nereida.deioc.ull.es/~pl/perlexamples/debuggingtut.pdf
The pdf file in http://nereida.deioc.ull.es/~pl/perlexamples/eyapplanguageref.pdf
The pdf file in http://nereida.deioc.ull.es/~pl/perlexamples/Treeregexp.pdf
The pdf file in http://nereida.deioc.ull.es/~pl/perlexamples/Node.pdf
The pdf file in http://nereida.deioc.ull.es/~pl/perlexamples/YATW.pdf
The pdf file in http://nereida.deioc.ull.es/~pl/perlexamples/Eyapp.pdf
The pdf file in http://nereida.deioc.ull.es/~pl/perlexamples/Base.pdf
The pdf file in http://nereida.deioc.ull.es/~pl/perlexamples/translationschemestut.pdf
The pdf file in http://nereida.deioc.ull.es/~pl/perlexamples/treematchingtut.pdf
perldoc eyapp,
perldoc treereg,
perldoc vgg,
The Syntax Highlight file for vim at http://www.vim.org/scripts/script.php?script_id=2453 and http://nereida.deioc.ull.es/~vim/
Analisis Lexico y Sintactico, (Notes for a course in compiler construction) by Casiano Rodriguez-Leon. Available at http://nereida.deioc.ull.es/~pl/perlexamples/ Is the more complete and reliable source for Parse::Eyapp. However is in Spanish.
Parse::Yapp,
Man pages of yacc(1) and bison(1), http://www.delorie.com/gnu/docs/bison/bison.html
Language::AttributeGrammar
Parse::RecDescent.
HOP::Parser
HOP::Lexer
ocamlyacc tutorial at http://plus.kaist.ac.kr/~shoh/ocaml/ocamllex-ocamlyacc/ocamlyacc-tutorial/ocamlyacc-tutorial.html
The classic Dragon's book Compilers: Principles, Techniques, and Tools by Alfred V. Aho, Ravi Sethi and Jeffrey D. Ullman (Addison-Wesley 1986)
CS2121: The Implementation and Power of Programming Languages (See http://www.cs.man.ac.uk/~pjj, http://www.cs.man.ac.uk/~pjj/complang/g2lr.html and http://www.cs.man.ac.uk/~pjj/cs2121/ho/ho.html) by Pete Jinks
Hal Finkel http://www.halssoftware.com/
G. Williams http://kasei.us/
Thomas L. Shinnick http://search.cpan.org/~tshinnic/
Frank Leray
Casiano Rodriguez-Leon (casiano@ull.es)
This work has been supported by CEE (FEDER) and the Spanish Ministry of Educacion y Ciencia through Plan Nacional I+D+I number TIN2005-08818-C04-04 (ULL::OPLINK project http://www.oplink.ull.es/). Support from Gobierno de Canarias was through GC02210601 (Grupos Consolidados). The University of La Laguna has also supported my work in many ways and for many years.
A large percentage of code is verbatim taken from Parse::Yapp 1.05. The author of Parse::Yapp is Francois Desarmenien.
I wish to thank Francois Desarmenien for his Parse::Yapp module, to my students at La Laguna and to the Perl Community. Thanks to the people who have contributed to improve the module (see "CONTRIBUTORS" in Parse::Eyapp). Thanks to Larry Wall for giving us Perl. Special thanks to Juana.
Copyright (c) 2006-2008 Casiano Rodriguez-Leon (casiano@ull.es). All rights reserved.
Parse::Yapp copyright is of Francois Desarmenien, all rights reserved. 1998-2001
These modules are free software; you can redistribute it and/or modify it under the same terms as Perl itself. See perlartistic.
This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
1 POD Error
The following errors were encountered while parsing the POD:
Non-ASCII character seen before =encoding in '¿Why?.'. Assuming CP1252
To install Parse::Eyapp, copy and paste the appropriate command in to your terminal.
cpanm
cpanm Parse::Eyapp
CPAN shell
perl -MCPAN -e shell install Parse::Eyapp
For more information on module installation, please visit the detailed CPAN module installation guide.