mirror of
https://github.com/qpdf/qpdf.git
synced 2024-11-16 17:45:09 +00:00
5d4cad9c02
Significantly improve the code's use of off_t for file offsets, size_t for memory sizes, and integer types in cases where there has to be compatibility with external interfaces. Rework sections of the code that would have prevented qpdf from working on files larger than 2 (or maybe 4) GB in size.
229 lines
4.7 KiB
C++
229 lines
4.7 KiB
C++
#include <qpdf/Pl_LZWDecoder.hh>
|
|
|
|
#include <qpdf/QTC.hh>
|
|
#include <stdexcept>
|
|
#include <string.h>
|
|
#include <assert.h>
|
|
|
|
Pl_LZWDecoder::Pl_LZWDecoder(char const* identifier, Pipeline* next,
|
|
bool early_code_change) :
|
|
Pipeline(identifier, next),
|
|
code_size(9),
|
|
next(0),
|
|
byte_pos(0),
|
|
bit_pos(0),
|
|
bits_available(0),
|
|
code_change_delta(early_code_change ? 1 : 0),
|
|
eod(false),
|
|
last_code(256)
|
|
{
|
|
memset(buf, 0, 3);
|
|
}
|
|
|
|
Pl_LZWDecoder::~Pl_LZWDecoder()
|
|
{
|
|
}
|
|
|
|
void
|
|
Pl_LZWDecoder::write(unsigned char* bytes, size_t len)
|
|
{
|
|
for (size_t i = 0; i < len; ++i)
|
|
{
|
|
this->buf[next++] = bytes[i];
|
|
if (this->next == 3)
|
|
{
|
|
this->next = 0;
|
|
}
|
|
this->bits_available += 8;
|
|
if (this->bits_available >= this->code_size)
|
|
{
|
|
sendNextCode();
|
|
}
|
|
}
|
|
}
|
|
|
|
void
|
|
Pl_LZWDecoder::finish()
|
|
{
|
|
getNext()->finish();
|
|
}
|
|
|
|
void
|
|
Pl_LZWDecoder::sendNextCode()
|
|
{
|
|
int high = this->byte_pos;
|
|
int med = (this->byte_pos + 1) % 3;
|
|
int low = (this->byte_pos + 2) % 3;
|
|
|
|
int bits_from_high = 8 - this->bit_pos;
|
|
int bits_from_med = this->code_size - bits_from_high;
|
|
int bits_from_low = 0;
|
|
if (bits_from_med > 8)
|
|
{
|
|
bits_from_low = bits_from_med - 8;
|
|
bits_from_med = 8;
|
|
}
|
|
int high_mask = (1 << bits_from_high) - 1;
|
|
int med_mask = 0xff - ((1 << (8 - bits_from_med)) - 1);
|
|
int low_mask = 0xff - ((1 << (8 - bits_from_low)) - 1);
|
|
int code = 0;
|
|
code += (this->buf[high] & high_mask) << bits_from_med;
|
|
code += ((this->buf[med] & med_mask) >> (8 - bits_from_med));
|
|
if (bits_from_low)
|
|
{
|
|
code <<= bits_from_low;
|
|
code += ((this->buf[low] & low_mask) >> (8 - bits_from_low));
|
|
this->byte_pos = low;
|
|
this->bit_pos = bits_from_low;
|
|
}
|
|
else
|
|
{
|
|
this->byte_pos = med;
|
|
this->bit_pos = bits_from_med;
|
|
}
|
|
if (this->bit_pos == 8)
|
|
{
|
|
this->bit_pos = 0;
|
|
++this->byte_pos;
|
|
this->byte_pos %= 3;
|
|
}
|
|
this->bits_available -= this->code_size;
|
|
|
|
handleCode(code);
|
|
}
|
|
|
|
unsigned char
|
|
Pl_LZWDecoder::getFirstChar(int code)
|
|
{
|
|
unsigned char result = '\0';
|
|
if (code < 256)
|
|
{
|
|
result = (unsigned char) code;
|
|
}
|
|
else
|
|
{
|
|
assert(code > 257);
|
|
unsigned int idx = code - 258;
|
|
assert(idx < table.size());
|
|
Buffer& b = table[idx];
|
|
result = b.getBuffer()[0];
|
|
}
|
|
return result;
|
|
}
|
|
|
|
void
|
|
Pl_LZWDecoder::addToTable(unsigned char next)
|
|
{
|
|
unsigned int last_size = 0;
|
|
unsigned char const* last_data = 0;
|
|
unsigned char tmp[1];
|
|
|
|
if (this->last_code < 256)
|
|
{
|
|
tmp[0] = this->last_code;
|
|
last_data = tmp;
|
|
last_size = 1;
|
|
}
|
|
else
|
|
{
|
|
assert(this->last_code > 257);
|
|
unsigned int idx = this->last_code - 258;
|
|
assert(idx < table.size());
|
|
Buffer& b = table[idx];
|
|
last_data = b.getBuffer();
|
|
last_size = (unsigned int) b.getSize();
|
|
}
|
|
|
|
Buffer entry(1 + last_size);
|
|
unsigned char* new_data = entry.getBuffer();
|
|
memcpy(new_data, last_data, last_size);
|
|
new_data[last_size] = next;
|
|
this->table.push_back(entry);
|
|
}
|
|
|
|
void
|
|
Pl_LZWDecoder::handleCode(int code)
|
|
{
|
|
if (this->eod)
|
|
{
|
|
return;
|
|
}
|
|
|
|
if (code == 256)
|
|
{
|
|
if (! this->table.empty())
|
|
{
|
|
QTC::TC("libtests", "Pl_LZWDecoder intermediate reset");
|
|
}
|
|
this->table.clear();
|
|
this->code_size = 9;
|
|
}
|
|
else if (code == 257)
|
|
{
|
|
this->eod = true;
|
|
}
|
|
else
|
|
{
|
|
if (this->last_code != 256)
|
|
{
|
|
// Add to the table from last time. New table entry would
|
|
// be what we read last plus the first character of what
|
|
// we're reading now.
|
|
unsigned char next = '\0';
|
|
unsigned int table_size = (unsigned int) table.size();
|
|
if (code < 256)
|
|
{
|
|
// just read < 256; last time's next was code
|
|
next = code;
|
|
}
|
|
else if (code > 257)
|
|
{
|
|
size_t idx = code - 258;
|
|
if (idx > table_size)
|
|
{
|
|
throw std::runtime_error("LZWDecoder: bad code received");
|
|
}
|
|
else if (idx == table_size)
|
|
{
|
|
// The encoder would have just created this entry,
|
|
// so the first character of this entry would have
|
|
// been the same as the first character of the
|
|
// last entry.
|
|
QTC::TC("libtests", "Pl_LZWDecoder last was table size");
|
|
next = getFirstChar(this->last_code);
|
|
}
|
|
else
|
|
{
|
|
next = getFirstChar(code);
|
|
}
|
|
}
|
|
unsigned int new_idx = 258 + table_size;
|
|
if (new_idx == 4096)
|
|
{
|
|
throw std::runtime_error("LZWDecoder: table full");
|
|
}
|
|
addToTable(next);
|
|
unsigned int change_idx = new_idx + code_change_delta;
|
|
if ((change_idx == 511) ||
|
|
(change_idx == 1023) ||
|
|
(change_idx == 2047))
|
|
{
|
|
++this->code_size;
|
|
}
|
|
}
|
|
|
|
if (code < 256)
|
|
{
|
|
unsigned char ch = (unsigned char) code;
|
|
getNext()->write(&ch, 1);
|
|
}
|
|
else
|
|
{
|
|
Buffer& b = table[code - 258];
|
|
getNext()->write(b.getBuffer(), b.getSize());
|
|
}
|
|
}
|
|
|
|
this->last_code = code;
|
|
}
|