sunterlib/s48/binary-parse/binary-parse.scm

463 lines
15 KiB
Scheme

; Binary parsing
;----------------------------------------
; Apologia
;
; Binary parsing and unparsing are transformations between primitive or
; composite Scheme values and their external binary representations.
;
; Examples include reading and writing JPEG, TIFF, MP3, ELF file
; formats, communicating with DNS, Kerberos, LDAP, SLP internet
; services, participating in Sun RPC and CORBA/IIOP distributed systems,
; storing and retrieving (arrays of) floating-point numbers in a
; portable and efficient way. This project will propose a set of low- and
; intermediate- level procedures that make binary parsing possible.
; Scheme is a good language to do research in text compression. Text
; compression involves a great deal of building and traversing
; dictionaries, trees and similar data structures, where Scheme
; excels. Performance doesn't matter in research, but the size of
; compressed files does (to figure out the bpc for the common
; benchmarks). Variable-bit i/o is a necessity. It is implemented
; in the present file.
; ASN.1 corresponds to a higher-level parsing (LR parser
; vs. lexer). Information in LDAP responses and X.509 certificates is
; structural and recursive, and so lends itself to be processed in
; Scheme. Variable bit i/o is necessary, and so is a binary lexer for
; a LR parser. Parsing of ASN.1 is a highly profitable enterprise
;----------------------------------------
; The outline of the project
;
; Primitives and streams
;
; - read-byte
; - read-u8vector (cf. read-string)
; - with-input-from-u8vector, with-input-from-encoded-u8vector 'base64,...
; building binary i/o streams from a sequence of bytes. Streams over
; u8vector, u16vector, etc. provide a serial access to memory. See SRFI-4
;
; - read-bit, read-bits via overlayed streams given read-byte
; implemented in the present file.
;
; - mmap-u8vector, munmap-u8vector
;
; Conversions
; - u8vector->integer u8vector endianness,
; u8vector->sinteger u8vector endianness
; These conversion procedures turn a sequence of bytes to an unsigned or
; signed integer, minding the byte order. The u8vector in question can
; have size 1,2,4,8, 3 etc. bytes. These two functions therefore can be
; used to read shorts, longs, extra longs, etc. numbers.
; - u8vector-reverse and other useful u8vector operations
;
; - modf, frexp, ldexp
; The above primitives can be emulated in R5RS, yet they are quite handy
; (for portable FP manipulation) and can be executed very efficiently by
; an FPU.
;
; Higher-level parsing and combinators
; These are combinators that can compose primitives above for more
; complex (possibly iterative) actions.
;
; - skip-bits, next-u8token,...
; - IIOP, RPC/XDR, RMI
; - binary lexer for existing LR/LL-parsers
;
; The composition of primitives and combinators will represent binary
; parsing language in a _full_ notation. This is similar to XPath
; expressions in full notation. Later we need to find out the
; most-frequently used patterns of the binary parsing language and
; design an abbreviated notation. The latter will need a special
; "interpreter". The abbreviated notation may turn out to look like
; Olin's regular expressions.
; $Id: binary-parse.scm,v 1.1 2003/04/14 06:38:47 mainzelM Exp $
;----------------------------------------
; Test harness
;
; The following macro runs built-in test cases -- or does not run,
; depending on which of the two lines below you commented out
;(define-syntax run-test
; (syntax-rules ()
; ((run-test body ...)
; (begin (newline)
; (display "-->Test")
; (newline)
; body ...))))
(define-syntax run-test
(syntax-rules ()
((run-test body ...)
(begin #f))))
;(define-macro (run-test . body) '(begin #f))
;(defmacro run-test body `(begin (display "\n-->Test\n") ,@body))
;;========================================================================
;; Configuration section
;;
; Performance is very important for binary parsing. We have to get all
; help from a particular Scheme system we can get. If a Scheme function
; can support the following primitives faster, we should take
; advantage of that fact.
;---
; Configuration for Gambit. See below for other systems, as well as R5RS
; implementations
;(define-macro (logior x y) `(##fixnum.logior ,x ,y))
;(define-macro (logand x y) `(##fixnum.logand ,x ,y))
;(define-macro (lsh-left x n) `(##fixnum.shl ,x ,n))
;(define-macro (lsh-right x n) `(##fixnum.lshr ,x ,n))
;(define-macro (lsh-left-one x) `(##fixnum.shl ,x 1))
;(define-macro (lsh-right-one x) `(##fixnum.lshr ,x 1))
;(define-macro (-- x) `(##fixnum.- ,x 1))
;(define-macro (++ x) `(##fixnum.+ ,x 1))
;(define-macro (bit-set? x mask) ; return x & mask != 0
; `(##not (##fixnum.zero? (logand ,x ,mask)))
;)
; End of the Gambit-specific configuration section
;---
; ---
; Configuration for S48/scsh
(define logior bitwise-ior)
(define logand bitwise-and)
(define (lsh-right x n) (arithmetic-shift x (- n)))
(define lsh-left arithmetic-shift)
(define (lsh-left-one x) (arithmetic-shift x 1))
(define (lsh-right-one x) (arithmetic-shift x -1))
(define (bit-set? x mask)
(not (zero? (bitwise-and x mask))))
; -- and ++ are not valid R5RS names:
; ++ is not used in this file anyway
; -- is replaced by decrement-one:
(define (decrement-one n)
(- n 1))
(define-syntax time
(syntax-rules ()
((time expr)
expr)))
; End of S48/scsh configuration
; ---
; combine bytes in the MSB order. A byte may be #f
(define (combine-two b1 b2) ; The result is for sure a fixnum
(and b1 b2 (logior (lsh-left b1 8) b2)))
(define (combine-three b1 b2 b3) ; The result is for sure a fixnum
(and b1 b2 b3 (logior (lsh-left (logior (lsh-left b1 8) b2) 8) b3)))
; Here the result may be a BIGNUM
(define (combine-bytes . bytes)
(cond
((null? bytes) 0)
((not (car bytes)) #f)
(else
(let loop ((bytes (cdr bytes)) (result (car bytes)))
(cond
((null? bytes) result)
((not (car bytes)) #f)
(else (loop (cdr bytes) (+ (car bytes) (* 256 result)))))))))
;---
; R5RS implementations of the primitives
; This is the most portable -- and the slowest implementation
; See also logical.scm from SLIB
; (define (logior x y)
; (cond ((= x y) x)
; ((zero? x) y)
; ((zero? y) x)
; (else
; (+ (* (logior (quotient x 2) (quotient y 2)) 2)
; (if (and (even? x) (even? y)) 0 1)))))
; (define (logand x y)
; (cond ((= x y) x)
; ((zero? x) 0)
; ((zero? y) 0)
; (else
; (+ (* (logand (quotient x 2) (quotient y 2)) 2)
; (if (or (even? x) (even? y)) 0 1)))))
; (define (lsh-left x n) (* x (expt 2 n)))
; (define (lsh-right x n) (quotient x (expt 2 n)))
; (define (lsh-left-one x) (* x 2))
; (define (lsh-right-one x) (quotient x 2))
; (define (-- x) (- x 1))
; (define (++ x) (+ x 1))
;(define (bit-set? x mask) ; return x & mask != 0
; (odd? (quotient x mask)) ; mask is an exact power of two
;)
;========================================================================
; Reading a byte
; Read-byte is a fundamental primitive; it needs to be
; added to the standard. Most of the other functions are library
; procedures. The following is an approximation, which clearly doesn't
; hold if the port is a Unicode (especially UTF-8) character stream.
; Return a byte as an exact integer [0,255], or the EOF object
(define (read-byte port)
(let ((c (read-char port)))
(if (eof-object? c) c (char->ascii c))))
; The same as above, but returns #f on EOF.
(define (read-byte-f port)
(let ((c (read-char port)))
(and (not (eof-object? c)) (char->ascii c))))
;========================================================================
; Bit stream
; -- Function: make-bit-reader BYTE-READER
; Given a BYTE-READER (a thunk), construct and return a function
; bit-reader N
;
; that reads N bits from a byte-stream represented by the BYTE-READER.
; The BYTE-READER is a function that takes no arguments and returns
; the current byte as an exact integer [0-255]. The byte reader
; should return #f on EOF.
; The bit reader returns N bits as an exact unsigned integer,
; 0 -... (no limit). N must be a positive integer, otherwise the bit reader
; returns #f. There is no upper limit on N -- other than the size of the
; input stream itself and the amount of (virtual) memory an OS is willing
; to give to your process. If you want to read 1M of _bits_, go ahead.
;
; It is assumed that the bit order is the most-significant bit first.
;
; Note the bit reader keeps the following condition true at all times:
; (= current-inport-pos (ceiling (/ no-bits-read 8)))
; That is, no byte is read until the very moment we really need (some of)
; its bits. The bit reader does _not_ "byte read ahead".
; Therefore, it can be used to handle a concatenation of different
; bit/byte streams *STRICTLY* sequentially, _without_ 'backing up a char',
; 'unreading-char' etc. tricks.
; For example, make-bit-reader has been used to read GRIB files of
; meteorological data, which made of several bitstreams with headers and
; tags.
; Thus careful attention to byte-buffering and optimization are the
; features of this bit reader.
;
; Usage example:
; (define bit-reader (make-bit-reader (lambda () #b11000101)))
; (bit-reader 3) ==> 6
; (bit-reader 4) ==> 2
; The test driver below is another example.
;
; Notes on the algorithm.
; The function recognizes and handles the following special cases:
; - the buffer is empty and 8, 16, 24 bits are to be read
; - reading all bits which are currently in the byte-buffer
; (and then maybe more)
; - reading only one bit
; Since the bit reader is going to be called many times, optimization is
; critical. We need all the help from the compiler/interpreter
; we can get.
(define (make-bit-reader byte-reader)
(let ((buffer 0) (mask 0) ; mask = 128 means that the buffer is full and
; the msb bit is the current (yet unread) bit
(bits-in-buffer 0))
; read the byte into the buffer and set up the counters.
; return #f on eof
(define (set-buffer)
(set! buffer (byte-reader))
(and buffer
(begin
(set! bits-in-buffer 8)
(set! mask 128)
#t)))
; Read fewer bits than there are in the buffer
(define (read-few-bits n)
(let ((value (logand buffer ; all bits in buffer
(decrement-one (lsh-left-one mask)))))
(set! bits-in-buffer (- bits-in-buffer n))
(set! mask (lsh-right mask n))
(lsh-right value bits-in-buffer))) ; remove extra bits
; read n bits given an empty buffer, and append them to value, n>=8
(define (add-more-bits value n)
(let loop ((value value) (n n))
(cond
((zero? n) value)
((< n 8)
(let ((rest (read-n-bits n)))
(and rest (+ (* value (lsh-left 1 n)) rest))))
(else
(let ((b (byte-reader)))
(and b (loop (+ (* value 256) b) (- n 8))))))))
; The main module
(define (read-n-bits n)
; Check the most common cases first
(cond
((not (positive? n)) #f)
((zero? bits-in-buffer) ; the bit-buffer is empty
(case n
((8) (byte-reader))
((16)
(let ((b (byte-reader)))
(combine-two b (byte-reader))))
((24)
(let* ((b1 (byte-reader)) (b2 (byte-reader)))
(combine-three b1 b2 (byte-reader))))
(else
(cond
((< n 8)
(and (set-buffer) (read-few-bits n)))
((< n 16)
(let ((b (byte-reader)))
(and (set-buffer)
(logior (lsh-left b (- n 8))
(read-few-bits (- n 8))))))
(else
(let ((b (byte-reader)))
(and b (add-more-bits b (- n 8)))))))))
((= n 1) ; read one bit
(let ((value (if (bit-set? buffer mask) 1 0)))
(set! mask (lsh-right-one mask))
(set! bits-in-buffer (decrement-one bits-in-buffer))
value))
((>= n bits-in-buffer) ; will empty the buffer
(let ((n-rem (- n bits-in-buffer))
(value (logand buffer ; for mask=64, it'll be &63
(decrement-one (lsh-left-one mask)))))
(set! bits-in-buffer 0)
(cond
((zero? n-rem) value)
((<= n-rem 16)
(let ((rest (read-n-bits n-rem)))
(and rest (logior (lsh-left value n-rem) rest))))
(else (add-more-bits value n-rem)))))
(else (read-few-bits n))
))
read-n-bits)
)
; Validation tests
(run-test
(define (read-bits numbers nbits)
(let* ((left-numbers numbers)
(bit-reader
(make-bit-reader
(lambda ()
(and (pair? left-numbers)
(let ((byte (car left-numbers)))
(set! left-numbers (cdr left-numbers))
byte))))))
(let loop ((result '()))
(let ((num (bit-reader nbits)))
(if num (loop (cons num result)) (reverse result))))))
(define (do-test numbers nbits expected)
(let ((result (read-bits numbers nbits)))
(for-each display
(list "Reading " numbers " by " nbits " bits" ;newline
"The result is: " result ))
(or (equal? result expected)
(error "the result differs from the expected: " expected))))
(do-test '(1 2 3 4 5 6 7) 8 '(1 2 3 4 5 6 7))
(do-test '(193 5 131 4) 1
'(1 1 0 0 0 0 0 1 0 0 0 0 0 1 0 1 1 0 0 0 0 0 1 1
0 0 0 0 0 1 0 0))
(do-test '(193 5 131 4 5) 2
'(3 0 0 1 0 0 1 1 2 0 0 3 0 0 1 0 0 0 1 1))
(do-test '(193 5 131 4) 3
'(6 0 2 0 2 6 0 3 0 1))
(do-test '(193 5 131 4 5 6 7) 4
'(12 1 0 5 8 3 0 4 0 5 0 6 0 7))
(do-test '(193 5 131 4 5 6 7) 5
'(24 4 2 24 6 1 0 5 0 24 3))
(do-test '(193 5 131 4 5 6 7 8 17 24) 8
'(193 5 131 4 5 6 7 8 17 24))
(do-test '(193 5 131 4 5 6 7 8 17 24) 9
'(386 22 24 64 160 385 388 17))
(do-test '(193 5 131 4 5 6 7 8 17) 16
'(49413 33540 1286 1800))
(do-test '(193 5 131 4 5 6 104) 17
'(98827 3088 10291))
(do-test '(193 5 131 4 5 6 104) 19
'(395308 49409))
(do-test '(193 5 131 4 5 6 104) 55
'(27165365385724724))
(do-test '(193 5 131 4 5 6 104) 56
'(54330730771449448))
)
; Timing test
; This test relies on a Gambit special form 'time' to clock
; evaluation of an expression.
; R5RS does not provide any timing facilities. So the test below
; might not run on your particular system, and probably needs
; adjustment anyway.
(run-test
(let ((fname "/tmp/a") (size 10240)
(pattern (ascii->char #x55)))
(define (read-by-bits n)
(for-each display
(list "Reading the file by " n " bits "))
(call-with-input-file fname
(lambda (port)
(let ((bit-reader (make-bit-reader
(lambda () (read-byte-f port)))))
(time
(do ((c (bit-reader n) (bit-reader n))) ((not c))))))))
(for-each display
(list "Creating a file " fname " of size " size " filled with "
pattern ;"\n"
))
(with-output-to-file fname
(lambda () (do ((i 0 (+ 1 i))) ((>= i size)) (write-char pattern))))
(newline)
(display "Reading the file by characters: the baseline ")
(call-with-input-file fname
(lambda (port)
(time
(do ((c (read-char port) (read-char port))) ((eof-object? c))))))
(newline)
(display "Reading the file by bytes: ")
(call-with-input-file fname
(lambda (port)
(time
(do ((c (read-byte-f port) (read-byte-f port))) ((not c))))))
(for-each read-by-bits
(list 1 2 3 4 5 6 7 8 9 10 11 15 16 17 23 24 25 30 32 65535
(* 8 size)))
))