man wncstr (Fonctions bibliothèques) - general parsing and writing package

NAME

wn_make_cstream, wn_free_cstream, wn_current_mark, wn_set_current_mark, wn_flush_to_mark, wn_cstream_debug - general parsing and writing package

SYNOPSIS

#include <wn/wncstr.h>

WN_PARSE
WN_WRITE

wn_make_cstream(&stream,pmove_block,move_block_ptr,
block_size,parse_or_write)
wn_cstream stream;
void (*pmove_block)(/* block,&actually_moved,request_moved,
move_block_ptr */);
ptr move_block_ptr;
int block_size,parse_or_write;

wn_free_cstream(stream)
wn_cstream stream;

int wn_current_mark(stream)
wn_cstream stream;

bool wn_set_current_mark(stream,mark)
wn_cstream stream;
int mark;

wn_flush_to_mark(stream,mark)
wn_cstream stream;
int mark;

bool wn_cstream_debug(stream)
wn_cstream stream;

DESCRIPTION

wncstr, wncstl, wnptok, wnwtok, wnwtokp, and wncinf comprise an integrated package designed to parse and write complicated files and strings. Parsing is accomplished using a generalized recursive-descent methodology [1]; writing is accomplished using a recursive-descent like methodology. Identical or very similar grammars work for parsing and for writing a language. The C code is also very similar. Very complicated grammars can be parsed efficiently using this package, including context-free grammars with long lookahead and many context-sensitive grammars. No distinction is made between lexical analysis and parsing proper; the lexical analysis is handled in the same way as the parsing.

The object to be parsed or written must first be converted to a logical character array (a cstream). wn_make_cstream accomplishes this in general; for files and strings see wncstl for shortcuts. A cstream is logically an array of characters (that is, bytes), together with a cursor or "current mark". The current mark points to the character currently being looked at. Thus, during parsing or writing, the current mark moves from 0 (beginning of stream) to larger and larger values and eventually reaches end of stream.

Parsing is accomplished using a generalized recursive-descent methodology. Generally, you need a C subroutine for each non-terminal in the grammar. The subroutine generally has 2 or more arguments and returns a boolean. The first argument is the cstream to parse; the other arguments are parse trees and other data constructed if the parse succeeds; the return value says whether the parse succeeded. When called, the subroutine attempts to parse the non-terminal from the cstream, starting at the current mark. If it fails, it returns FALSE and sets the current mark back to its start value (using wn_set_current_mark and wn_current_mark). If its succeeds, it returns TRUE, usually generates some data (such as a parse tree), and leaves the current mark after the parsed non-terminal in the cstream. These subroutines are built from other non-terminal parsing subroutines. The hierarchy bottoms out with the subroutines in wnptok. Subroutines are often combined in logical expressions.

All characters earlier in the cstream than the current mark take up dynamic memory unless flushed using wn_flush_to_mark. Use wn_flush_to_mark when it is clear that characters before mark will never be examined again.

wn_cstream_debug prints the characters from stream in the neighborhood of the current mark. This information is very useful for debugging parsers. wn_cstream_debug always returns TRUE, so it can be inserted into logical expressions.

Writing is also accomplished with a recursive-descent like methodology, with a subroutine for each non-terminal in the grammar. The subroutine generally has 2 or more arguments; no value is returned. The first argument is the cstream to write into; the other arguments contain the information to write. These subroutines are built from other non-terminal writing subroutines. The hierarchy bottoms out with the subroutines in wnwtok and wnwtokp.

Generally, the same or similar subroutine hierarchies works for parsing and writing a language.

Routines in wncinf assist in printing error messages with line numbers.

Routines in wnscan scan a cstream for patterns.

BUGS

Too many subroutine names to remember.

This package may be slower than less general parsing packages.

SEE ALSO

wncstl, wnptok, wnscan, wnwtok, wnwtokp, wncinf

REFERENCES

[1] Alfred V. Aho, Jeffrey D. Ullman: Principles of Compiler Design, April 1979, Addison-Wesley Publishing, Reading Massachusetts.

AUTHOR

Will Naylor

CETTE PAGE DOCUMENTE AUSSI :