1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
|
Unions revived in bf(C++) with the advent of unrestricted unions. In this
section we'll have a look at how to used them as semantic values in a Bisonc++
generated parser.
The just developed polymorphic semantic value had to support
itemization(
itt(int) values;
itt(textual) values of type tt(string);
)
When unrestricted unions are used, the union must be able to support tt(int)
and tt(string) values. Using the tt(pair) to allow the union to do
`introspection' (cf. section ref(UNIONS)), we now define a em(union)
tt(Semantic) having the following data members:
verb( std::pair<int, int> d_int;
std::pair<int, std::string> d_str;)
Since an unrestricted union by default deletes all its standard
constructors and operators (it must do so, as it cannot tell which data
variant is currently active) they must all be implemented explicitly by
the tt(Semantic) union. To aid the introspection a type is defined, indicating
whether the variant is an tt(int) or a tt(string), and the default constructor
simply defines an tt(INT) value 0.
Here is tt(Semantic)'s interface:
verbinclude(-a unrestricted/semantic/semantic.h)
tt(Semantic) objects are as big as required to accommodate all its
variants. However, as tt(Semantic) is a union, its constructors and destructor
cannot predefine or destroy all of its variants. Rather, it must pick the
appropriate field based on the tt(pair)'s tt(first) field. For tt(int) values
a simple assignment is OK, but for tt(string) values a constructor and
destructor must be used. For this, placement new and an explicit destructor
call is required. Here are tt(Semantic)'s tt(Semantic(Type, char const *)),
its copy constructor and its destructor:
verbinclude(-a unrestricted/semantic/semantic1.cc)
verbinclude(-a unrestricted/semantic/semantic2.cc)
verbinclude(-a unrestricted/semantic/destructor.cc)
Our little parser only needs to be able to insert semantic values. For this an
overloaded insertion operator can be used:
verbinclude(-a unrestricted/semantic/operatorinsert.cc)
The remaining members (tt(operator=, swap)) are standard, and need no further
explanation.
As tt(Semantic) objects know what type of value they represent, the parser's
grammar specification needs no modification. Minor modifications for the
scanner's specification, tt(lexer), however are required. Since tt(Semantic)
is no longer a wrapper class for a polymorphic base class, using a shared
pointer, the previously used tt(shared_ptr)'s tt(reset) calls need to be
replaced by direct assignments of tt(Semantic) values to the tt(*d_semval)
semantic value. The relevant section of the scanner's specification file
tt(lexer) now becomes:
verb( [0-9]+ {
*d_semval = Semantic(Semantic::INT, yytext);
return Parser::INT;
}
[a-zA-Z_][a-zA-Z0-9_]* {
*d_semval = Semantic(Semantic::IDENTIFIER, yytext);
return Parser::IDENTIFIER;
})
This is all that is required to change a parser using a polymorphic base
class to a parser using an unrestricted union. As memory allocation using
placement new is fast, an unrestricted union is an efficient way to store
multiple data types. Moreover, the semantic value classes implementing the
polymorphic semantic value behavior,
tt(Base, Int), and tt(Text) are no longer required, nor is the use of the
tt(shared_ptr) in a wrapper class (the previously used tt(Semantic) value)
required anymore. All in, using an unrestricted union for the parser's
semantic values seems like a good deal.
|
