pure_parser.tex

由matlab开发的hybrid系统的描述语言

\chapter{Global Objects}
\label{globals_c}

\section{Globals}

The class \cppclass{Globals} contains all global objects, among which the symbol table and the options specified
at the command line. Almost every class has a pointer to \cppclass{Globals}, which is set by the constructor.


\section{Pure Parser}

This section describes how the parser and the lexer communicate. The way it is implemented is messy, however, this
is due to bison which does not clearly separate header from source file. Neither can it generate the parser as a
class.

In the file \file{hys.l}, we put the following two lines:
\begin{verbatim}
%option c++
%option yyclass="Pure_lexer"
\end{verbatim}
This causes flex to generate a class \cppclass{yyFlexLexer} containing all the functionality the lexer requires
except the actual lexer function \function{yylex}. That function is generated as \method{Pure\_lexer}{yylex},
where \cppclass{Pure\_lexer} is derived from \cppclass{yyFlexLexer}.

In order to pass information to and from the lexer, the following two lines go in \file{hys.y}:
\begin{verbatim}
%pure_parser
#define YYLEX_PARAM p_in
\end{verbatim}
With these two options, bison expects the following lexer declaration:
\begin{verbatim}
int yylex (YYSTYPE *lvalp, YYLTYPE *llocp, lexer_input *l_in);
\end{verbatim}
The returned value is the next token read. \code{lvalp} and \code{llocp} are used to pass the semantic value of the
token and its line number to the parser. \code{l\_in} is used to pass information to the lexer.
\code{lexer\_input} contains a pointer to \cppclass{Globals} and a pointer to the actual lexer (an instance of
\cppclass{Pure\_lexer}).

\cppclass{Pure\_lexer} has the following attributes:
\begin{verbatim}
    Globals *globals;
    YYSTYPE *lvalp;
    YYLTYPE *llocp;
\end{verbatim}
\code{int Pure\_lexer::yylex (YYSTYPE \*lvalp, YYLTYPE \*llocp, lexer\_input \*l\_in)} sets these attributes and
then calls \function{yylex} (the lexer generated by flex), which can use these values. The global function {\tt
yylex} called from within the parser is:
\begin{verbatim}
int yylex (YYSTYPE *lvalp, YYLTYPE *llocp, lexer_input *l_in) {
    return l_in->lexer->yylex(lvalp, llocp, l_in->globals);
}
\end{verbatim}

Here is the catch: \function{yylex} needs to be declared in \file{hys.y} in the C declaration section.
\code{YYSTYPE} and \code{YYLTYPE} however are declared in the grammar section of \file{hys.y} and are therefore
not yet available. It is true that bison generates a header (\file{hys.tab.h}) containing these declarations, but
this header cannot be included in \file{hys.y}, because then \code{YYSTYPE} would be defined twice resulting in a
compile error. Our solution is to put the \function{yylex} declaration immediately after \code{\%union \{ \ldots
\} } using \code{\%\{ \ldots \%\} }. The code put there goes unchanged into \file{hys.tab.c}, at a position where
\code{YYSTYPE} is already known.

In order to pass information from the main program to the parser, we did the following: In \file{hys.y}, defining
{\sl \#define YYPARSE\_PARAM p\_in } causes bison to generate the parser as
\begin{verbatim}
int yyparse(void *p_in);
\end{verbatim}
The following structure is used as the argument:
\begin{verbatim}
typedef struct {
    Globals *globals;
    ofstream *MLD_out, *simu_out;
    struct lexer_input *l_in;
} parser_input;
\end{verbatim}
Thus, the parser gets a pointer to \cppclass{Globals}, which contains the symbol table and the command line
options. \code{MLD\_out} is where the \MLD{} matrices are written to and \code{simu\_out} is the output stream for
the simulator, if it is requested. \code{l\_in} is the argument given to the lexer.