mirror of
https://github.com/qpdf/qpdf.git
synced 2024-11-10 23:20:58 +00:00
1362 lines
43 KiB
C++
1362 lines
43 KiB
C++
// This program tests miscellaneous functionality in the qpdf library
|
|
// that we don't want to pollute the qpdf program with.
|
|
|
|
#include <qpdf/QPDF.hh>
|
|
|
|
#include <qpdf/QUtil.hh>
|
|
#include <qpdf/QTC.hh>
|
|
#include <qpdf/Pl_StdioFile.hh>
|
|
#include <qpdf/Pl_Buffer.hh>
|
|
#include <qpdf/Pl_Flate.hh>
|
|
#include <qpdf/QPDFWriter.hh>
|
|
#include <iostream>
|
|
#include <sstream>
|
|
#include <algorithm>
|
|
#include <stdio.h>
|
|
#include <string.h>
|
|
#include <stdlib.h>
|
|
#include <assert.h>
|
|
#include <map>
|
|
|
|
static char const* whoami = 0;
|
|
|
|
void usage()
|
|
{
|
|
std::cerr << "Usage: " << whoami << " n filename1 [arg2]"
|
|
<< std::endl;
|
|
exit(2);
|
|
}
|
|
|
|
class Provider: public QPDFObjectHandle::StreamDataProvider
|
|
{
|
|
public:
|
|
Provider(PointerHolder<Buffer> b) :
|
|
b(b),
|
|
bad_length(false)
|
|
{
|
|
}
|
|
virtual ~Provider()
|
|
{
|
|
}
|
|
virtual void provideStreamData(int objid, int generation,
|
|
Pipeline* p)
|
|
{
|
|
p->write(b->getBuffer(), b->getSize());
|
|
if (this->bad_length)
|
|
{
|
|
unsigned char ch = ' ';
|
|
p->write(&ch, 1);
|
|
}
|
|
p->finish();
|
|
}
|
|
void badLength(bool v)
|
|
{
|
|
this->bad_length = v;
|
|
}
|
|
|
|
private:
|
|
PointerHolder<Buffer> b;
|
|
bool bad_length;
|
|
};
|
|
|
|
class ParserCallbacks: public QPDFObjectHandle::ParserCallbacks
|
|
{
|
|
public:
|
|
virtual ~ParserCallbacks()
|
|
{
|
|
}
|
|
|
|
virtual void handleObject(QPDFObjectHandle);
|
|
virtual void handleEOF();
|
|
};
|
|
|
|
void
|
|
ParserCallbacks::handleObject(QPDFObjectHandle obj)
|
|
{
|
|
if (obj.isName() && (obj.getName() == "/Abort"))
|
|
{
|
|
std::cout << "test suite: terminating parsing" << std::endl;
|
|
terminateParsing();
|
|
}
|
|
std::cout << obj.getTypeName() << ": ";
|
|
if (obj.isInlineImage())
|
|
{
|
|
// Exercise getTypeCode
|
|
assert(obj.getTypeCode() == QPDFObject::ot_inlineimage);
|
|
std::cout << QUtil::hex_encode(obj.getInlineImageValue()) << std::endl;
|
|
}
|
|
else
|
|
{
|
|
std::cout << obj.unparse() << std::endl;
|
|
}
|
|
}
|
|
|
|
void
|
|
ParserCallbacks::handleEOF()
|
|
{
|
|
std::cout << "-EOF-" << std::endl;
|
|
}
|
|
|
|
static std::string getPageContents(QPDFObjectHandle page)
|
|
{
|
|
PointerHolder<Buffer> b1 =
|
|
page.getKey("/Contents").getStreamData();
|
|
return std::string(
|
|
reinterpret_cast<char *>(b1->getBuffer()), b1->getSize()) + "\0";
|
|
}
|
|
|
|
static void checkPageContents(QPDFObjectHandle page,
|
|
std::string const& wanted_string)
|
|
{
|
|
std::string contents = getPageContents(page);
|
|
if (contents.find(wanted_string) == std::string::npos)
|
|
{
|
|
std::cout << "didn't find " << wanted_string << " in "
|
|
<< contents << std::endl;
|
|
}
|
|
}
|
|
|
|
static QPDFObjectHandle createPageContents(QPDF& pdf, std::string const& text)
|
|
{
|
|
std::string contents = "BT /F1 15 Tf 72 720 Td (" + text + ") Tj ET\n";
|
|
return QPDFObjectHandle::newStream(&pdf, contents);
|
|
}
|
|
|
|
void runtest(int n, char const* filename1, char const* arg2)
|
|
{
|
|
// Most tests here are crafted to work on specific files. Look at
|
|
// the test suite to see how the test is invoked to find the file
|
|
// that the test is supposed to operate on.
|
|
|
|
if (n == 0)
|
|
{
|
|
// Throw in some random test cases that don't fit anywhere
|
|
// else. This is in addition to whatever else is going on in
|
|
// test 0.
|
|
|
|
// The code to trim user passwords looks for 0x28 (which is
|
|
// "(") since it marks the beginning of the padding. Exercise
|
|
// the code to make sure it skips over 0x28 characters that
|
|
// aren't part of padding.
|
|
std::string password(
|
|
"1234567890123456789012(45678\x28\xbf\x4e\x5e");
|
|
assert(password.length() == 32);
|
|
QPDF::trim_user_password(password);
|
|
assert(password == "1234567890123456789012(45678");
|
|
|
|
QPDFObjectHandle uninitialized;
|
|
assert(uninitialized.getTypeCode() == QPDFObject::ot_uninitialized);
|
|
assert(strcmp(uninitialized.getTypeName(), "uninitialized") == 0);
|
|
}
|
|
|
|
QPDF pdf;
|
|
PointerHolder<char> file_buf;
|
|
FILE* filep = 0;
|
|
if (n == 0)
|
|
{
|
|
pdf.setAttemptRecovery(false);
|
|
}
|
|
if (((n == 35) || (n == 36)) && (arg2 != 0))
|
|
{
|
|
// arg2 is password
|
|
pdf.processFile(filename1, arg2);
|
|
}
|
|
else if (n % 2 == 0)
|
|
{
|
|
if (n % 4 == 0)
|
|
{
|
|
QTC::TC("qpdf", "exercise processFile(name)");
|
|
pdf.processFile(filename1);
|
|
}
|
|
else
|
|
{
|
|
QTC::TC("qpdf", "exercise processFile(FILE*)");
|
|
filep = QUtil::safe_fopen(filename1, "rb");
|
|
pdf.processFile(filename1, filep, false);
|
|
}
|
|
}
|
|
else
|
|
{
|
|
QTC::TC("qpdf", "exercise processMemoryFile");
|
|
FILE* f = QUtil::safe_fopen(filename1, "rb");
|
|
fseek(f, 0, SEEK_END);
|
|
size_t size = QUtil::tell(f);
|
|
fseek(f, 0, SEEK_SET);
|
|
file_buf = PointerHolder<char>(true, new char[size]);
|
|
char* buf_p = file_buf.getPointer();
|
|
size_t bytes_read = 0;
|
|
size_t len = 0;
|
|
while ((len = fread(buf_p + bytes_read, 1, size - bytes_read, f)) > 0)
|
|
{
|
|
bytes_read += len;
|
|
}
|
|
if (bytes_read != size)
|
|
{
|
|
if (ferror(f))
|
|
{
|
|
throw std::runtime_error(
|
|
std::string("failure reading file ") + filename1 +
|
|
" into memory: read " +
|
|
QUtil::int_to_string(bytes_read) + "; wanted " +
|
|
QUtil::int_to_string(size));
|
|
}
|
|
else
|
|
{
|
|
throw std::logic_error(
|
|
std::string("premature eof reading file ") + filename1 +
|
|
" into memory: read " +
|
|
QUtil::int_to_string(bytes_read) + "; wanted " +
|
|
QUtil::int_to_string(size));
|
|
}
|
|
}
|
|
fclose(f);
|
|
pdf.processMemoryFile(filename1, buf_p, size);
|
|
}
|
|
|
|
if ((n == 0) || (n == 1))
|
|
{
|
|
QPDFObjectHandle trailer = pdf.getTrailer();
|
|
QPDFObjectHandle qtest = trailer.getKey("/QTest");
|
|
|
|
if (! trailer.hasKey("/QTest"))
|
|
{
|
|
// This will always happen when /QTest is null because
|
|
// hasKey returns false for null keys regardless of
|
|
// whether the key exists or not. That way there's never
|
|
// any difference between a key that is present and null
|
|
// and a key that is absent.
|
|
QTC::TC("qpdf", "main QTest implicit");
|
|
std::cout << "/QTest is implicit" << std::endl;
|
|
}
|
|
|
|
QTC::TC("qpdf", "main QTest indirect",
|
|
qtest.isIndirect() ? 1 : 0);
|
|
std::cout << "/QTest is "
|
|
<< (qtest.isIndirect() ? "in" : "")
|
|
<< "direct and has type "
|
|
<< qtest.getTypeName()
|
|
<< " (" << qtest.getTypeCode() << ")" << std::endl;
|
|
|
|
if (qtest.isNull())
|
|
{
|
|
QTC::TC("qpdf", "main QTest null");
|
|
std::cout << "/QTest is null" << std::endl;
|
|
}
|
|
else if (qtest.isBool())
|
|
{
|
|
QTC::TC("qpdf", "main QTest bool",
|
|
qtest.getBoolValue() ? 1 : 0);
|
|
std::cout << "/QTest is Boolean with value "
|
|
<< (qtest.getBoolValue() ? "true" : "false")
|
|
<< std::endl;
|
|
}
|
|
else if (qtest.isInteger())
|
|
{
|
|
QTC::TC("qpdf", "main QTest int");
|
|
std::cout << "/QTest is an integer with value "
|
|
<< qtest.getIntValue() << std::endl;
|
|
}
|
|
else if (qtest.isReal())
|
|
{
|
|
QTC::TC("qpdf", "main QTest real");
|
|
std::cout << "/QTest is a real number with value "
|
|
<< qtest.getRealValue() << std::endl;
|
|
}
|
|
else if (qtest.isName())
|
|
{
|
|
QTC::TC("qpdf", "main QTest name");
|
|
std::cout << "/QTest is a name with value "
|
|
<< qtest.getName() << std::endl;
|
|
}
|
|
else if (qtest.isString())
|
|
{
|
|
QTC::TC("qpdf", "main QTest string");
|
|
std::cout << "/QTest is a string with value "
|
|
<< qtest.getStringValue() << std::endl;
|
|
}
|
|
else if (qtest.isArray())
|
|
{
|
|
QTC::TC("qpdf", "main QTest array");
|
|
int n = qtest.getArrayNItems();
|
|
std::cout << "/QTest is an array with "
|
|
<< n << " items" << std::endl;
|
|
for (int i = 0; i < n; ++i)
|
|
{
|
|
QTC::TC("qpdf", "main QTest array indirect",
|
|
qtest.getArrayItem(i).isIndirect() ? 1 : 0);
|
|
std::cout << " item " << i << " is "
|
|
<< (qtest.getArrayItem(i).isIndirect() ? "in" : "")
|
|
<< "direct" << std::endl;
|
|
}
|
|
}
|
|
else if (qtest.isDictionary())
|
|
{
|
|
QTC::TC("qpdf", "main QTest dictionary");
|
|
std::cout << "/QTest is a dictionary" << std::endl;
|
|
std::set<std::string> keys = qtest.getKeys();
|
|
for (std::set<std::string>::iterator iter = keys.begin();
|
|
iter != keys.end(); ++iter)
|
|
{
|
|
QTC::TC("qpdf", "main QTest dictionary indirect",
|
|
(qtest.getKey(*iter).isIndirect() ? 1 : 0));
|
|
std::cout << " " << *iter << " is "
|
|
<< (qtest.getKey(*iter).isIndirect() ? "in" : "")
|
|
<< "direct" << std::endl;
|
|
}
|
|
}
|
|
else if (qtest.isStream())
|
|
{
|
|
QTC::TC("qpdf", "main QTest stream");
|
|
std::cout << "/QTest is a stream. Dictionary: "
|
|
<< qtest.getDict().unparse() << std::endl;
|
|
|
|
std::cout << "Raw stream data:" << std::endl;
|
|
std::cout.flush();
|
|
QUtil::binary_stdout();
|
|
PointerHolder<Pl_StdioFile> out = new Pl_StdioFile("raw", stdout);
|
|
qtest.pipeStreamData(out.getPointer(), false, false, false);
|
|
|
|
std::cout << std::endl << "Uncompressed stream data:" << std::endl;
|
|
if (qtest.pipeStreamData(0, true, false, false))
|
|
{
|
|
std::cout.flush();
|
|
QUtil::binary_stdout();
|
|
out = new Pl_StdioFile("filtered", stdout);
|
|
qtest.pipeStreamData(out.getPointer(), true, false, false);
|
|
std::cout << std::endl << "End of stream data" << std::endl;
|
|
}
|
|
else
|
|
{
|
|
std::cout << "Stream data is not filterable." << std::endl;
|
|
}
|
|
}
|
|
else
|
|
{
|
|
// Should not happen!
|
|
std::cout << "/QTest is an unknown object" << std::endl;
|
|
}
|
|
|
|
std::cout << "unparse: " << qtest.unparse() << std::endl
|
|
<< "unparseResolved: " << qtest.unparseResolved()
|
|
<< std::endl;
|
|
}
|
|
else if (n == 2)
|
|
{
|
|
// Encrypted file. This test case is designed for a specific
|
|
// PDF file.
|
|
|
|
QPDFObjectHandle trailer = pdf.getTrailer();
|
|
std::cout << trailer.getKey("/Info").
|
|
getKey("/CreationDate").getStringValue() << std::endl;
|
|
std::cout << trailer.getKey("/Info").
|
|
getKey("/Producer").getStringValue() << std::endl;
|
|
|
|
QPDFObjectHandle encrypt = trailer.getKey("/Encrypt");
|
|
std::cout << encrypt.getKey("/O").unparse() << std::endl;
|
|
std::cout << encrypt.getKey("/U").unparse() << std::endl;
|
|
|
|
QPDFObjectHandle root = pdf.getRoot();
|
|
QPDFObjectHandle pages = root.getKey("/Pages");
|
|
QPDFObjectHandle kids = pages.getKey("/Kids");
|
|
QPDFObjectHandle page = kids.getArrayItem(1); // second page
|
|
QPDFObjectHandle contents = page.getKey("/Contents");
|
|
QUtil::binary_stdout();
|
|
PointerHolder<Pl_StdioFile> out = new Pl_StdioFile("filtered", stdout);
|
|
contents.pipeStreamData(out.getPointer(), true, false, false);
|
|
}
|
|
else if (n == 3)
|
|
{
|
|
QPDFObjectHandle streams = pdf.getTrailer().getKey("/QStreams");
|
|
for (int i = 0; i < streams.getArrayNItems(); ++i)
|
|
{
|
|
QPDFObjectHandle stream = streams.getArrayItem(i);
|
|
std::cout << "-- stream " << i << " --" << std::endl;
|
|
std::cout.flush();
|
|
QUtil::binary_stdout();
|
|
PointerHolder<Pl_StdioFile> out =
|
|
new Pl_StdioFile("tokenized stream", stdout);
|
|
stream.pipeStreamData(out.getPointer(), true, true, false);
|
|
}
|
|
}
|
|
else if (n == 4)
|
|
{
|
|
// Mutability testing: Make /QTest direct recursively, then
|
|
// copy to /Info. Also make some other mutations so we can
|
|
// tell the difference and ensure that the original /QTest
|
|
// isn't effected.
|
|
QPDFObjectHandle trailer = pdf.getTrailer();
|
|
QPDFObjectHandle qtest = trailer.getKey("/QTest");
|
|
qtest.makeDirect();
|
|
qtest.removeKey("/Subject");
|
|
qtest.replaceKey("/Author",
|
|
QPDFObjectHandle::newString("Mr. Potato Head"));
|
|
// qtest.A and qtest.B.A were originally the same object.
|
|
// They no longer are after makeDirect(). Mutate one of them
|
|
// and ensure the other is not changed. These test cases are
|
|
// crafted around a specific set of input files.
|
|
QPDFObjectHandle A = qtest.getKey("/A");
|
|
if (A.getArrayItem(0).getIntValue() == 1)
|
|
{
|
|
// Test mutators
|
|
A.setArrayItem(1, QPDFObjectHandle::newInteger(5)); // 1 5 3
|
|
A.insertItem(2, QPDFObjectHandle::newInteger(10)); // 1 5 10 3
|
|
A.appendItem(QPDFObjectHandle::newInteger(12)); // 1 5 10 3 12
|
|
A.eraseItem(3); // 1 5 10 12
|
|
A.insertItem(4, QPDFObjectHandle::newInteger(6)); // 1 5 10 12 6
|
|
A.insertItem(0, QPDFObjectHandle::newInteger(9)); // 9 1 5 10 12 6
|
|
}
|
|
else
|
|
{
|
|
std::vector<QPDFObjectHandle> items;
|
|
items.push_back(QPDFObjectHandle::newInteger(14));
|
|
items.push_back(QPDFObjectHandle::newInteger(15));
|
|
items.push_back(QPDFObjectHandle::newInteger(9));
|
|
A.setArrayFromVector(items);
|
|
}
|
|
|
|
trailer.replaceKey("/Info", pdf.makeIndirectObject(qtest));
|
|
QPDFWriter w(pdf, 0);
|
|
w.setQDFMode(true);
|
|
w.setStaticID(true);
|
|
w.write();
|
|
|
|
// Prevent "done" message from getting appended
|
|
exit(0);
|
|
}
|
|
else if (n == 5)
|
|
{
|
|
std::vector<QPDFObjectHandle> pages = pdf.getAllPages();
|
|
int pageno = 0;
|
|
for (std::vector<QPDFObjectHandle>::iterator iter = pages.begin();
|
|
iter != pages.end(); ++iter)
|
|
{
|
|
QPDFObjectHandle& page = *iter;
|
|
++pageno;
|
|
|
|
std::cout << "page " << pageno << ":" << std::endl;
|
|
|
|
std::cout << " images:" << std::endl;
|
|
std::map<std::string, QPDFObjectHandle> images =
|
|
page.getPageImages();
|
|
for (std::map<std::string, QPDFObjectHandle>::iterator iter =
|
|
images.begin(); iter != images.end(); ++iter)
|
|
{
|
|
std::string const& name = (*iter).first;
|
|
QPDFObjectHandle image = (*iter).second;
|
|
QPDFObjectHandle dict = image.getDict();
|
|
int width = dict.getKey("/Width").getIntValue();
|
|
int height = dict.getKey("/Height").getIntValue();
|
|
std::cout << " " << name
|
|
<< ": " << width << " x " << height
|
|
<< std::endl;
|
|
}
|
|
|
|
std::cout << " content:" << std::endl;
|
|
std::vector<QPDFObjectHandle> content = page.getPageContents();
|
|
for (std::vector<QPDFObjectHandle>::iterator iter = content.begin();
|
|
iter != content.end(); ++iter)
|
|
{
|
|
std::cout << " " << (*iter).unparse() << std::endl;
|
|
}
|
|
|
|
std::cout << "end page " << pageno << std::endl;
|
|
}
|
|
|
|
QPDFObjectHandle root = pdf.getRoot();
|
|
QPDFObjectHandle qstrings = root.getKey("/QStrings");
|
|
if (qstrings.isArray())
|
|
{
|
|
std::cout << "QStrings:" << std::endl;
|
|
int n = qstrings.getArrayNItems();
|
|
for (int i = 0; i < n; ++i)
|
|
{
|
|
std::cout << qstrings.getArrayItem(i).getUTF8Value()
|
|
<< std::endl;
|
|
}
|
|
}
|
|
|
|
QPDFObjectHandle qnumbers = root.getKey("/QNumbers");
|
|
if (qnumbers.isArray())
|
|
{
|
|
std::cout << "QNumbers:" << std::endl;
|
|
int n = qnumbers.getArrayNItems();
|
|
for (int i = 0; i < n; ++i)
|
|
{
|
|
std::cout << QUtil::double_to_string(
|
|
qnumbers.getArrayItem(i).getNumericValue(), 3)
|
|
<< std::endl;
|
|
}
|
|
}
|
|
}
|
|
else if (n == 6)
|
|
{
|
|
QPDFObjectHandle root = pdf.getRoot();
|
|
QPDFObjectHandle metadata = root.getKey("/Metadata");
|
|
if (! metadata.isStream())
|
|
{
|
|
throw std::logic_error("test 6 run on file with no metadata");
|
|
}
|
|
Pl_Buffer bufpl("buffer");
|
|
metadata.pipeStreamData(&bufpl, false, false, false);
|
|
Buffer* buf = bufpl.getBuffer();
|
|
unsigned char const* data = buf->getBuffer();
|
|
bool cleartext = false;
|
|
if ((buf->getSize() > 9) &&
|
|
(strncmp(reinterpret_cast<char const*>(data),
|
|
"<?xpacket", 9) == 0))
|
|
{
|
|
cleartext = true;
|
|
}
|
|
delete buf;
|
|
std::cout << "encrypted="
|
|
<< (pdf.isEncrypted() ? 1 : 0)
|
|
<< "; cleartext="
|
|
<< (cleartext ? 1 : 0)
|
|
<< std::endl;
|
|
}
|
|
else if (n == 7)
|
|
{
|
|
QPDFObjectHandle root = pdf.getRoot();
|
|
QPDFObjectHandle qstream = root.getKey("/QStream");
|
|
if (! qstream.isStream())
|
|
{
|
|
throw std::logic_error("test 7 run on file with no QStream");
|
|
}
|
|
qstream.replaceStreamData(
|
|
"new data for stream\n",
|
|
QPDFObjectHandle::newNull(), QPDFObjectHandle::newNull());
|
|
QPDFWriter w(pdf, "a.pdf");
|
|
w.setStaticID(true);
|
|
w.setStreamDataMode(qpdf_s_preserve);
|
|
w.write();
|
|
}
|
|
else if (n == 8)
|
|
{
|
|
QPDFObjectHandle root = pdf.getRoot();
|
|
QPDFObjectHandle qstream = root.getKey("/QStream");
|
|
if (! qstream.isStream())
|
|
{
|
|
throw std::logic_error("test 7 run on file with no QStream");
|
|
}
|
|
Pl_Buffer p1("buffer");
|
|
Pl_Flate p2("compress", &p1, Pl_Flate::a_deflate);
|
|
p2.write(QUtil::unsigned_char_pointer("new data for stream\n"),
|
|
20); // no null!
|
|
p2.finish();
|
|
PointerHolder<Buffer> b = p1.getBuffer();
|
|
// This is a bogus way to use StreamDataProvider, but it does
|
|
// adequately test its functionality.
|
|
Provider* provider = new Provider(b);
|
|
PointerHolder<QPDFObjectHandle::StreamDataProvider> p = provider;
|
|
qstream.replaceStreamData(
|
|
p, QPDFObjectHandle::newName("/FlateDecode"),
|
|
QPDFObjectHandle::newNull());
|
|
provider->badLength(false);
|
|
QPDFWriter w(pdf, "a.pdf");
|
|
w.setStaticID(true);
|
|
// Linearize to force the provider to be called multiple times.
|
|
w.setLinearization(true);
|
|
w.setStreamDataMode(qpdf_s_preserve);
|
|
w.write();
|
|
|
|
// Every time a provider pipes stream data, it has to provide
|
|
// the same amount of data.
|
|
provider->badLength(true);
|
|
try
|
|
{
|
|
qstream.getStreamData();
|
|
std::cout << "oops -- getStreamData didn't throw" << std::endl;
|
|
}
|
|
catch (std::exception const& e)
|
|
{
|
|
std::cout << "exception: " << e.what() << std::endl;
|
|
}
|
|
}
|
|
else if (n == 9)
|
|
{
|
|
QPDFObjectHandle root = pdf.getRoot();
|
|
// Explicitly exercise the Buffer version of newStream
|
|
PointerHolder<Buffer> buf = new Buffer(20);
|
|
unsigned char* bp = buf->getBuffer();
|
|
memcpy(bp, "data for new stream\n", 20); // no null!
|
|
QPDFObjectHandle qstream = QPDFObjectHandle::newStream(
|
|
&pdf, buf);
|
|
QPDFObjectHandle rstream = QPDFObjectHandle::newStream(&pdf);
|
|
try
|
|
{
|
|
rstream.getStreamData();
|
|
std::cout << "oops -- getStreamData didn't throw" << std::endl;
|
|
}
|
|
catch (std::logic_error const& e)
|
|
{
|
|
std::cout << "exception: " << e.what() << std::endl;
|
|
}
|
|
rstream.replaceStreamData(
|
|
"data for other stream\n",
|
|
QPDFObjectHandle::newNull(), QPDFObjectHandle::newNull());
|
|
root.replaceKey("/QStream", qstream);
|
|
root.replaceKey("/RStream", rstream);
|
|
QPDFWriter w(pdf, "a.pdf");
|
|
w.setStaticID(true);
|
|
w.setStreamDataMode(qpdf_s_preserve);
|
|
w.write();
|
|
}
|
|
else if (n == 10)
|
|
{
|
|
std::vector<QPDFObjectHandle> pages = pdf.getAllPages();
|
|
pages.at(0).addPageContents(
|
|
QPDFObjectHandle::newStream(
|
|
&pdf, "BT /F1 12 Tf 72 620 Td (Baked) Tj ET\n"), true);
|
|
pages.at(0).addPageContents(
|
|
QPDFObjectHandle::newStream(
|
|
&pdf, "BT /F1 18 Tf 72 520 Td (Mashed) Tj ET\n"), false);
|
|
|
|
QPDFWriter w(pdf, "a.pdf");
|
|
w.setStaticID(true);
|
|
w.setStreamDataMode(qpdf_s_preserve);
|
|
w.write();
|
|
}
|
|
else if (n == 11)
|
|
{
|
|
QPDFObjectHandle root = pdf.getRoot();
|
|
QPDFObjectHandle qstream = root.getKey("/QStream");
|
|
PointerHolder<Buffer> b1 = qstream.getStreamData();
|
|
PointerHolder<Buffer> b2 = qstream.getRawStreamData();
|
|
if ((b1->getSize() == 7) &&
|
|
(memcmp(b1->getBuffer(), "potato\n", 7) == 0))
|
|
{
|
|
std::cout << "filtered stream data okay" << std::endl;
|
|
}
|
|
if ((b2->getSize() == 15) &&
|
|
(memcmp(b2->getBuffer(), "706F7461746F0A\n", 15) == 0))
|
|
{
|
|
std::cout << "raw stream data okay" << std::endl;
|
|
}
|
|
}
|
|
else if (n == 12)
|
|
{
|
|
pdf.setOutputStreams(0, 0);
|
|
pdf.showLinearizationData();
|
|
}
|
|
else if (n == 13)
|
|
{
|
|
std::ostringstream out;
|
|
std::ostringstream err;
|
|
pdf.setOutputStreams(&out, &err);
|
|
pdf.showLinearizationData();
|
|
std::cout << "---output---" << std::endl
|
|
<< out.str()
|
|
<< "---error---" << std::endl
|
|
<< err.str();
|
|
}
|
|
else if (n == 14)
|
|
{
|
|
// Exercise swap and replace. This test case is designed for
|
|
// a specific file.
|
|
std::vector<QPDFObjectHandle> pages = pdf.getAllPages();
|
|
if (pages.size() != 4)
|
|
{
|
|
throw std::logic_error("test " + QUtil::int_to_string(n) +
|
|
" not called 4-page file");
|
|
}
|
|
// Swap pages 2 and 3
|
|
pdf.swapObjects(pages.at(1).getObjGen(), pages.at(2).getObjGen());
|
|
// Replace object and swap objects
|
|
QPDFObjectHandle trailer = pdf.getTrailer();
|
|
QPDFObjectHandle qdict = trailer.getKey("/QDict");
|
|
QPDFObjectHandle qarray = trailer.getKey("/QArray");
|
|
// Force qdict but not qarray to resolve
|
|
qdict.isDictionary();
|
|
QPDFObjectHandle new_dict = QPDFObjectHandle::newDictionary();
|
|
new_dict.replaceKey("/NewDict", QPDFObjectHandle::newInteger(2));
|
|
try
|
|
{
|
|
// Do it wrong first...
|
|
pdf.replaceObject(qdict.getObjGen(), qdict);
|
|
}
|
|
catch (std::logic_error)
|
|
{
|
|
std::cout << "caught logic error as expected" << std::endl;
|
|
}
|
|
pdf.replaceObject(qdict.getObjGen(), new_dict);
|
|
// Now qdict still points to the old dictionary
|
|
std::cout << "old dict: " << qdict.getKey("/Dict").getIntValue()
|
|
<< std::endl;
|
|
// Swap dict and array
|
|
pdf.swapObjects(qdict.getObjGen(), qarray.getObjGen());
|
|
// Now qarray will resolve to new object but qdict is still
|
|
// the old object
|
|
std::cout << "old dict: " << qdict.getKey("/Dict").getIntValue()
|
|
<< std::endl;
|
|
std::cout << "new dict: " << qarray.getKey("/NewDict").getIntValue()
|
|
<< std::endl;
|
|
// Reread qdict, now pointing to an array
|
|
qdict = pdf.getObjectByObjGen(qdict.getObjGen());
|
|
std::cout << "swapped array: " << qdict.getArrayItem(0).getName()
|
|
<< std::endl;
|
|
|
|
// Exercise getAsMap and getAsArray
|
|
std::vector<QPDFObjectHandle> array_elements =
|
|
qdict.getArrayAsVector();
|
|
std::map<std::string, QPDFObjectHandle> dict_items =
|
|
qarray.getDictAsMap();
|
|
if ((array_elements.size() == 1) &&
|
|
(array_elements.at(0).getName() == "/Array") &&
|
|
(dict_items.size() == 1) &&
|
|
(dict_items["/NewDict"].getIntValue() == 2))
|
|
{
|
|
std::cout << "array and dictionary contents are correct"
|
|
<< std::endl;
|
|
}
|
|
|
|
// Exercise writing to memory buffer
|
|
for (int i = 0; i < 2; ++i)
|
|
{
|
|
QPDFWriter w(pdf);
|
|
w.setOutputMemory();
|
|
// Exercise setOutputMemory with and without static ID
|
|
w.setStaticID(i == 0);
|
|
w.setStreamDataMode(qpdf_s_preserve);
|
|
w.write();
|
|
Buffer* b = w.getBuffer();
|
|
std::string const filename = (i == 0 ? "a.pdf" : "b.pdf");
|
|
FILE* f = QUtil::safe_fopen(filename.c_str(), "wb");
|
|
fwrite(b->getBuffer(), b->getSize(), 1, f);
|
|
fclose(f);
|
|
delete b;
|
|
}
|
|
}
|
|
else if (n == 15)
|
|
{
|
|
std::vector<QPDFObjectHandle> const& pages = pdf.getAllPages();
|
|
// Reference to original page numbers for this test case are
|
|
// numbered from 0.
|
|
|
|
// Remove pages from various places, checking to make sure
|
|
// that our pages reference is getting updated.
|
|
assert(pages.size() == 10);
|
|
pdf.removePage(pages.back()); // original page 9
|
|
assert(pages.size() == 9);
|
|
pdf.removePage(*pages.begin()); // original page 0
|
|
assert(pages.size() == 8);
|
|
checkPageContents(pages.at(4), "Original page 5");
|
|
pdf.removePage(pages.at(4)); // original page 5
|
|
assert(pages.size() == 7);
|
|
checkPageContents(pages.at(4), "Original page 6");
|
|
checkPageContents(pages.at(0), "Original page 1");
|
|
checkPageContents(pages.at(6), "Original page 8");
|
|
|
|
// Insert pages
|
|
|
|
// Create some content streams.
|
|
std::vector<QPDFObjectHandle> contents;
|
|
contents.push_back(createPageContents(pdf, "New page 1"));
|
|
contents.push_back(createPageContents(pdf, "New page 0"));
|
|
contents.push_back(createPageContents(pdf, "New page 5"));
|
|
contents.push_back(createPageContents(pdf, "New page 6"));
|
|
contents.push_back(createPageContents(pdf, "New page 11"));
|
|
contents.push_back(createPageContents(pdf, "New page 12"));
|
|
|
|
// Create some page objects. Start with an existing
|
|
// dictionary and modify it. Using the results of
|
|
// getDictAsMap to create a new dictionary effectively creates
|
|
// a shallow copy.
|
|
QPDFObjectHandle page_template = pages.at(0);
|
|
std::vector<QPDFObjectHandle> new_pages;
|
|
for (std::vector<QPDFObjectHandle>::iterator iter = contents.begin();
|
|
iter != contents.end(); ++iter)
|
|
{
|
|
// We will retain indirect object references to other
|
|
// indirect objects other than page content.
|
|
QPDFObjectHandle page = page_template.shallowCopy();
|
|
page.replaceKey("/Contents", *iter);
|
|
if (iter == contents.begin())
|
|
{
|
|
// leave direct
|
|
new_pages.push_back(page);
|
|
}
|
|
else
|
|
{
|
|
new_pages.push_back(pdf.makeIndirectObject(page));
|
|
}
|
|
}
|
|
|
|
// Now insert the pages
|
|
pdf.addPage(new_pages.at(0), true);
|
|
checkPageContents(pages.at(0), "New page 1");
|
|
pdf.addPageAt(new_pages.at(1), true, pages.at(0));
|
|
assert(pages.at(0).getObjGen() == new_pages.at(1).getObjGen());
|
|
pdf.addPageAt(new_pages.at(2), true, pages.at(5));
|
|
assert(pages.at(5).getObjGen() == new_pages.at(2).getObjGen());
|
|
pdf.addPageAt(new_pages.at(3), false, pages.at(5));
|
|
assert(pages.at(6).getObjGen() == new_pages.at(3).getObjGen());
|
|
assert(pages.size() == 11);
|
|
pdf.addPage(new_pages.at(4), false);
|
|
assert(pages.at(11).getObjGen() == new_pages.at(4).getObjGen());
|
|
pdf.addPageAt(new_pages.at(5), false, pages.back());
|
|
assert(pages.size() == 13);
|
|
checkPageContents(pages.at(0), "New page 0");
|
|
checkPageContents(pages.at(1), "New page 1");
|
|
checkPageContents(pages.at(5), "New page 5");
|
|
checkPageContents(pages.at(6), "New page 6");
|
|
checkPageContents(pages.at(11), "New page 11");
|
|
checkPageContents(pages.at(12), "New page 12");
|
|
|
|
// Exercise writing to FILE*
|
|
FILE* out = QUtil::safe_fopen("a.pdf", "wb");
|
|
QPDFWriter w(pdf, "FILE* a.pdf", out, true);
|
|
w.setStaticID(true);
|
|
w.setStreamDataMode(qpdf_s_preserve);
|
|
w.write();
|
|
}
|
|
else if (n == 16)
|
|
{
|
|
// Insert a page manually and then update the cache.
|
|
std::vector<QPDFObjectHandle> const& all_pages = pdf.getAllPages();
|
|
|
|
QPDFObjectHandle contents = createPageContents(pdf, "New page 10");
|
|
QPDFObjectHandle page =
|
|
pdf.makeIndirectObject(
|
|
QPDFObjectHandle(all_pages.at(0)).shallowCopy());
|
|
page.replaceKey("/Contents", contents);
|
|
|
|
// Insert the page manually.
|
|
QPDFObjectHandle root = pdf.getRoot();
|
|
QPDFObjectHandle pages = root.getKey("/Pages");
|
|
QPDFObjectHandle kids = pages.getKey("/Kids");
|
|
page.replaceKey("/Parent", pages);
|
|
pages.replaceKey(
|
|
"/Count",
|
|
QPDFObjectHandle::newInteger(1 + all_pages.size()));
|
|
kids.appendItem(page);
|
|
assert(all_pages.size() == 10);
|
|
pdf.updateAllPagesCache();
|
|
assert(all_pages.size() == 11);
|
|
assert(all_pages.back().getObjGen() == page.getObjGen());
|
|
|
|
QPDFWriter w(pdf, "a.pdf");
|
|
w.setStaticID(true);
|
|
w.setStreamDataMode(qpdf_s_preserve);
|
|
w.write();
|
|
}
|
|
else if (n == 17)
|
|
{
|
|
// The input file to this test case is broken to exercise an
|
|
// error condition.
|
|
std::vector<QPDFObjectHandle> const& pages = pdf.getAllPages();
|
|
pdf.removePage(pages.at(0));
|
|
std::cout << "you can't see this" << std::endl;
|
|
}
|
|
else if (n == 18)
|
|
{
|
|
// Remove a page and re-insert it in the same file.
|
|
std::vector<QPDFObjectHandle> const& pages = pdf.getAllPages();
|
|
|
|
// Remove pages from various places, checking to make sure
|
|
// that our pages reference is getting updated.
|
|
assert(pages.size() == 10);
|
|
QPDFObjectHandle page5 = pages.at(5);
|
|
pdf.removePage(page5);
|
|
pdf.addPage(page5, false);
|
|
assert(pages.size() == 10);
|
|
assert(pages.back().getObjGen() == page5.getObjGen());
|
|
|
|
QPDFWriter w(pdf, "a.pdf");
|
|
w.setStaticID(true);
|
|
w.setStreamDataMode(qpdf_s_preserve);
|
|
w.write();
|
|
}
|
|
else if (n == 19)
|
|
{
|
|
// Remove a page and re-insert it in the same file.
|
|
std::vector<QPDFObjectHandle> const& pages = pdf.getAllPages();
|
|
|
|
// Try to insert a page that's already there.
|
|
pdf.addPage(pages.at(5), false);
|
|
std::cout << "you can't see this" << std::endl;
|
|
}
|
|
else if (n == 20)
|
|
{
|
|
// Shallow copy an array
|
|
QPDFObjectHandle trailer = pdf.getTrailer();
|
|
QPDFObjectHandle qtest = trailer.getKey("/QTest");
|
|
QPDFObjectHandle copy = qtest.shallowCopy();
|
|
// Append shallow copy of a scalar
|
|
copy.appendItem(trailer.getKey("/Size").shallowCopy());
|
|
trailer.replaceKey("/QTest2", copy);
|
|
|
|
QPDFWriter w(pdf, "a.pdf");
|
|
w.setStaticID(true);
|
|
w.setStreamDataMode(qpdf_s_preserve);
|
|
w.write();
|
|
}
|
|
else if (n == 21)
|
|
{
|
|
// Try to shallow copy a stream
|
|
std::vector<QPDFObjectHandle> const& pages = pdf.getAllPages();
|
|
QPDFObjectHandle page = pages.at(0);
|
|
QPDFObjectHandle contents = page.getKey("/Contents");
|
|
contents.shallowCopy();
|
|
std::cout << "you can't see this" << std::endl;
|
|
}
|
|
else if (n == 22)
|
|
{
|
|
// Try to remove a page we don't have
|
|
std::vector<QPDFObjectHandle> const& pages = pdf.getAllPages();
|
|
QPDFObjectHandle page = pages.at(0);
|
|
pdf.removePage(page);
|
|
pdf.removePage(page);
|
|
std::cout << "you can't see this" << std::endl;
|
|
}
|
|
else if (n == 23)
|
|
{
|
|
std::vector<QPDFObjectHandle> const& pages = pdf.getAllPages();
|
|
pdf.removePage(pages.back());
|
|
}
|
|
else if (n == 24)
|
|
{
|
|
// Test behavior of reserved objects
|
|
QPDFObjectHandle res1 = QPDFObjectHandle::newReserved(&pdf);
|
|
QPDFObjectHandle res2 = QPDFObjectHandle::newReserved(&pdf);
|
|
QPDFObjectHandle trailer = pdf.getTrailer();
|
|
trailer.replaceKey("Array1", res1);
|
|
trailer.replaceKey("Array2", res2);
|
|
|
|
QPDFObjectHandle array1 = QPDFObjectHandle::newArray();
|
|
QPDFObjectHandle array2 = QPDFObjectHandle::newArray();
|
|
array1.appendItem(res2);
|
|
array1.appendItem(QPDFObjectHandle::newInteger(1));
|
|
array2.appendItem(res1);
|
|
array2.appendItem(QPDFObjectHandle::newInteger(2));
|
|
// Make sure trying to ask questions about a reserved object
|
|
// doesn't break it.
|
|
if (res1.isArray())
|
|
{
|
|
std::cout << "oops -- res1 is an array" << std::endl;
|
|
}
|
|
if (res1.isReserved())
|
|
{
|
|
std::cout << "res1 is still reserved after checking if array"
|
|
<< std::endl;
|
|
}
|
|
pdf.replaceReserved(res1, array1);
|
|
if (res1.isReserved())
|
|
{
|
|
std::cout << "oops -- res1 is still reserved" << std::endl;
|
|
}
|
|
else
|
|
{
|
|
std::cout << "res1 is no longer reserved" << std::endl;
|
|
}
|
|
res1.assertArray();
|
|
std::cout << "res1 is an array" << std::endl;
|
|
|
|
try
|
|
{
|
|
res2.unparseResolved();
|
|
std::cout << "oops -- didn't throw" << std::endl;
|
|
}
|
|
catch (std::logic_error e)
|
|
{
|
|
std::cout << "logic error: " << e.what() << std::endl;
|
|
}
|
|
try
|
|
{
|
|
res2.makeDirect();
|
|
std::cout << "oops -- didn't throw" << std::endl;
|
|
}
|
|
catch (std::logic_error e)
|
|
{
|
|
std::cout << "logic error: " << e.what() << std::endl;
|
|
}
|
|
|
|
pdf.replaceReserved(res2, array2);
|
|
|
|
res2.assertArray();
|
|
std::cout << "res2 is an array" << std::endl;
|
|
|
|
// Verify that the previously added reserved keys can be
|
|
// dereferenced properly now
|
|
int i1 = res1.getArrayItem(0).getArrayItem(1).getIntValue();
|
|
int i2 = res2.getArrayItem(0).getArrayItem(1).getIntValue();
|
|
if ((i1 == 2) && (i2 == 1))
|
|
{
|
|
std::cout << "circular access and lazy resolution worked" << std::endl;
|
|
}
|
|
|
|
QPDFWriter w(pdf, "a.pdf");
|
|
w.setStaticID(true);
|
|
w.setStreamDataMode(qpdf_s_preserve);
|
|
w.write();
|
|
}
|
|
else if (n == 25)
|
|
{
|
|
// The copy object tests are designed to work with a specific
|
|
// file. Look at the test suite for the file, and look at the
|
|
// file for comments about the file's structure.
|
|
|
|
// Copy qtest without crossing page boundaries. Should get O1
|
|
// and O2 and their streams but not O3 or any other pages.
|
|
|
|
assert(arg2 != 0);
|
|
QPDF newpdf;
|
|
newpdf.processFile(arg2);
|
|
QPDFObjectHandle qtest = pdf.getTrailer().getKey("/QTest");
|
|
newpdf.getTrailer().replaceKey(
|
|
"/QTest", newpdf.copyForeignObject(qtest));
|
|
|
|
QPDFWriter w(newpdf, "a.pdf");
|
|
w.setStaticID(true);
|
|
w.setStreamDataMode(qpdf_s_preserve);
|
|
w.write();
|
|
}
|
|
else if (n == 26)
|
|
{
|
|
// Copy the O3 page using addPage. Copy qtest without
|
|
// crossing page boundaries. In addition to previous results,
|
|
// should get page O3 but no other pages including the page
|
|
// that O3 points to. Also, inherited object will have been
|
|
// pushed down and will be preserved.
|
|
|
|
assert(arg2 != 0);
|
|
QPDF newpdf;
|
|
newpdf.processFile(arg2);
|
|
QPDFObjectHandle qtest = pdf.getTrailer().getKey("/QTest");
|
|
QPDFObjectHandle O3 = qtest.getKey("/O3");
|
|
newpdf.addPage(O3, false);
|
|
newpdf.getTrailer().replaceKey(
|
|
"/QTest", newpdf.copyForeignObject(qtest));
|
|
|
|
QPDFWriter w(newpdf, "a.pdf");
|
|
w.setStaticID(true);
|
|
w.setStreamDataMode(qpdf_s_preserve);
|
|
w.write();
|
|
}
|
|
else if (n == 27)
|
|
{
|
|
// Copy O3 and the page O3 refers to before copying qtest.
|
|
// Should get qtest plus only the O3 page and the page that O3
|
|
// points to. Inherited objects should be preserved.
|
|
|
|
assert(arg2 != 0);
|
|
QPDF newpdf;
|
|
newpdf.processFile(arg2);
|
|
QPDFObjectHandle qtest = pdf.getTrailer().getKey("/QTest");
|
|
QPDFObjectHandle O3 = qtest.getKey("/O3");
|
|
newpdf.addPage(O3.getKey("/OtherPage"), false);
|
|
newpdf.addPage(O3, false);
|
|
newpdf.getTrailer().replaceKey(
|
|
"/QTest", newpdf.copyForeignObject(qtest));
|
|
|
|
QPDFWriter w(newpdf, "a.pdf");
|
|
w.setStaticID(true);
|
|
w.setStreamDataMode(qpdf_s_preserve);
|
|
w.write();
|
|
}
|
|
else if (n == 28)
|
|
{
|
|
// Copy foreign object errors
|
|
try
|
|
{
|
|
pdf.copyForeignObject(pdf.getTrailer().getKey("/QTest"));
|
|
std::cout << "oops -- didn't throw" << std::endl;
|
|
}
|
|
catch (std::logic_error e)
|
|
{
|
|
std::cout << "logic error: " << e.what() << std::endl;
|
|
}
|
|
try
|
|
{
|
|
pdf.copyForeignObject(QPDFObjectHandle::newInteger(1));
|
|
std::cout << "oops -- didn't throw" << std::endl;
|
|
}
|
|
catch (std::logic_error e)
|
|
{
|
|
std::cout << "logic error: " << e.what() << std::endl;
|
|
}
|
|
}
|
|
else if (n == 29)
|
|
{
|
|
// Detect mixed objects in QPDFWriter
|
|
assert(arg2 != 0);
|
|
QPDF other;
|
|
other.processFile(arg2);
|
|
// Should use copyForeignObject instead
|
|
other.getTrailer().replaceKey(
|
|
"/QTest", pdf.getTrailer().getKey("/QTest"));
|
|
|
|
try
|
|
{
|
|
QPDFWriter w(other, "a.pdf");
|
|
w.write();
|
|
std::cout << "oops -- didn't throw" << std::endl;
|
|
}
|
|
catch (std::logic_error e)
|
|
{
|
|
std::cout << "logic error: " << e.what() << std::endl;
|
|
}
|
|
}
|
|
else if (n == 30)
|
|
{
|
|
assert(arg2 != 0);
|
|
QPDF encrypted;
|
|
encrypted.processFile(arg2, "user");
|
|
QPDFWriter w(pdf, "b.pdf");
|
|
w.setStreamDataMode(qpdf_s_preserve);
|
|
w.copyEncryptionParameters(encrypted);
|
|
w.write();
|
|
|
|
// Make sure the contents are actually the same
|
|
QPDF final;
|
|
final.processFile("b.pdf", "user");
|
|
std::vector<QPDFObjectHandle> pages = pdf.getAllPages();
|
|
std::string orig_contents = getPageContents(pages.at(0));
|
|
pages = final.getAllPages();
|
|
std::string new_contents = getPageContents(pages.at(0));
|
|
if (orig_contents != new_contents)
|
|
{
|
|
std::cout << "oops -- page contents don't match" << std::endl
|
|
<< "original:\n" << orig_contents
|
|
<< "new:\n" << new_contents
|
|
<< std::endl;
|
|
}
|
|
}
|
|
else if (n == 31)
|
|
{
|
|
// Test object parsing from a string. The input file is not used.
|
|
|
|
QPDFObjectHandle o1 =
|
|
QPDFObjectHandle::parse(
|
|
"[/name 16059 3.14159 false\n"
|
|
" << /key true /other [ (string1) (string2) ] >> null]");
|
|
std::cout << o1.unparse() << std::endl;
|
|
QPDFObjectHandle o2 = QPDFObjectHandle::parse(" 12345 \f ");
|
|
assert(o2.isInteger() && (o2.getIntValue() == 12345));
|
|
try
|
|
{
|
|
QPDFObjectHandle::parse("[1 0 R]", "indirect test");
|
|
std::cout << "oops -- didn't throw" << std::endl;
|
|
}
|
|
catch (std::logic_error e)
|
|
{
|
|
std::cout << "logic error parsing indirect: " << e.what()
|
|
<< std::endl;
|
|
}
|
|
try
|
|
{
|
|
QPDFObjectHandle::parse("0 trailing", "trailing test");
|
|
std::cout << "oops -- didn't throw" << std::endl;
|
|
}
|
|
catch (std::runtime_error e)
|
|
{
|
|
std::cout << "trailing data: " << e.what()
|
|
<< std::endl;
|
|
}
|
|
}
|
|
else if (n == 32)
|
|
{
|
|
// Extra header text
|
|
char const* filenames[] = {"a.pdf", "b.pdf", "c.pdf", "d.pdf"};
|
|
for (int i = 0; i < 4; ++i)
|
|
{
|
|
bool linearized = ((i & 1) != 0);
|
|
bool newline = ((i & 2) != 0);
|
|
QPDFWriter w(pdf, filenames[i]);
|
|
w.setStaticID(true);
|
|
std::cout
|
|
<< "file: " << filenames[i] << std::endl
|
|
<< "linearized: " << (linearized ? "yes" : "no") << std::endl
|
|
<< "newline: " << (newline ? "yes" : "no") << std::endl;
|
|
w.setLinearization(linearized);
|
|
w.setExtraHeaderText(newline
|
|
? "%% Comment with newline\n"
|
|
: "%% Comment\n% No newline");
|
|
w.write();
|
|
}
|
|
}
|
|
else if (n == 33)
|
|
{
|
|
// Test writing to a custom pipeline
|
|
Pl_Buffer p("buffer");
|
|
QPDFWriter w(pdf);
|
|
w.setStaticID(true);
|
|
w.setOutputPipeline(&p);
|
|
w.write();
|
|
PointerHolder<Buffer> b = p.getBuffer();
|
|
FILE* f = QUtil::safe_fopen("a.pdf", "wb");
|
|
fwrite(b->getBuffer(), b->getSize(), 1, f);
|
|
fclose(f);
|
|
}
|
|
else if (n == 34)
|
|
{
|
|
// Look at Extensions dictionary
|
|
std::cout << "version: " << pdf.getPDFVersion() << std::endl
|
|
<< "extension level: " << pdf.getExtensionLevel() << std::endl
|
|
<< pdf.getRoot().getKey("/Extensions").unparse() << std::endl;
|
|
}
|
|
else if (n == 35)
|
|
{
|
|
// Extract attachments
|
|
|
|
std::map<std::string, PointerHolder<Buffer> > attachments;
|
|
QPDFObjectHandle root = pdf.getRoot();
|
|
QPDFObjectHandle names = root.getKey("/Names");
|
|
QPDFObjectHandle embeddedFiles = names.getKey("/EmbeddedFiles");
|
|
names = embeddedFiles.getKey("/Names");
|
|
for (int i = 0; i < names.getArrayNItems(); ++i)
|
|
{
|
|
QPDFObjectHandle item = names.getArrayItem(i);
|
|
if (item.isDictionary() &&
|
|
item.getKey("/Type").isName() &&
|
|
(item.getKey("/Type").getName() == "/Filespec") &&
|
|
item.getKey("/EF").isDictionary() &&
|
|
item.getKey("/EF").getKey("/F").isStream())
|
|
{
|
|
std::string filename = item.getKey("/F").getStringValue();
|
|
QPDFObjectHandle stream = item.getKey("/EF").getKey("/F");
|
|
attachments[filename] = stream.getStreamData();
|
|
}
|
|
}
|
|
for (std::map<std::string, PointerHolder<Buffer> >::iterator iter =
|
|
attachments.begin(); iter != attachments.end(); ++iter)
|
|
{
|
|
std::string const& filename = (*iter).first;
|
|
std::string data = std::string(
|
|
reinterpret_cast<char const*>((*iter).second->getBuffer()),
|
|
(*iter).second->getSize());
|
|
bool is_binary = false;
|
|
for (size_t i = 0; i < data.size(); ++i)
|
|
{
|
|
if ((data.at(i) < 0) || (data.at(i) > 126))
|
|
{
|
|
is_binary = true;
|
|
break;
|
|
}
|
|
}
|
|
if (is_binary)
|
|
{
|
|
std::string t;
|
|
for (size_t i = 0;
|
|
i < std::min(data.size(), static_cast<size_t>(20));
|
|
++i)
|
|
{
|
|
if ((data.at(i) >= 32) && (data.at(i) <= 126))
|
|
{
|
|
t += data.at(i);
|
|
}
|
|
else
|
|
{
|
|
t += ".";
|
|
}
|
|
}
|
|
t += " (" + QUtil::int_to_string(data.size()) + " bytes)";
|
|
data = t;
|
|
}
|
|
std::cout << filename << ":\n" << data << "--END--\n";
|
|
}
|
|
}
|
|
else if (n == 36)
|
|
{
|
|
// Extract raw unfilterable attachment
|
|
|
|
QPDFObjectHandle root = pdf.getRoot();
|
|
QPDFObjectHandle names = root.getKey("/Names");
|
|
QPDFObjectHandle embeddedFiles = names.getKey("/EmbeddedFiles");
|
|
names = embeddedFiles.getKey("/Names");
|
|
for (int i = 0; i < names.getArrayNItems(); ++i)
|
|
{
|
|
QPDFObjectHandle item = names.getArrayItem(i);
|
|
if (item.isDictionary() &&
|
|
item.getKey("/Type").isName() &&
|
|
(item.getKey("/Type").getName() == "/Filespec") &&
|
|
item.getKey("/EF").isDictionary() &&
|
|
item.getKey("/EF").getKey("/F").isStream() &&
|
|
(item.getKey("/F").getStringValue() == "attachment1.txt"))
|
|
{
|
|
std::string filename = item.getKey("/F").getStringValue();
|
|
QPDFObjectHandle stream = item.getKey("/EF").getKey("/F");
|
|
Pl_Buffer p1("buffer");
|
|
Pl_Flate p2("compress", &p1, Pl_Flate::a_inflate);
|
|
stream.pipeStreamData(&p2, false, false, false);
|
|
PointerHolder<Buffer> buf = p1.getBuffer();
|
|
std::string data = std::string(
|
|
reinterpret_cast<char const*>(buf->getBuffer()),
|
|
buf->getSize());
|
|
std::cout << stream.getDict().unparse()
|
|
<< filename << ":\n" << data << "--END--\n";
|
|
}
|
|
}
|
|
}
|
|
else if (n == 37)
|
|
{
|
|
// Parse content streams of all pages
|
|
std::vector<QPDFObjectHandle> pages = pdf.getAllPages();
|
|
for (std::vector<QPDFObjectHandle>::iterator iter = pages.begin();
|
|
iter != pages.end(); ++iter)
|
|
{
|
|
QPDFObjectHandle page = *iter;
|
|
QPDFObjectHandle contents = page.getKey("/Contents");
|
|
ParserCallbacks cb;
|
|
QPDFObjectHandle::parseContentStream(contents, &cb);
|
|
}
|
|
}
|
|
else if (n == 38)
|
|
{
|
|
// Designed for override-compressed-object.pdf
|
|
QPDFObjectHandle qtest = pdf.getRoot().getKey("/QTest");
|
|
for (int i = 0; i < qtest.getArrayNItems(); ++i)
|
|
{
|
|
std::cout << qtest.getArrayItem(i).unparseResolved() << std::endl;
|
|
}
|
|
}
|
|
else
|
|
{
|
|
throw std::runtime_error(std::string("invalid test ") +
|
|
QUtil::int_to_string(n));
|
|
}
|
|
|
|
if (filep)
|
|
{
|
|
fclose(filep);
|
|
}
|
|
std::cout << "test " << n << " done" << std::endl;
|
|
}
|
|
|
|
int main(int argc, char* argv[])
|
|
{
|
|
QUtil::setLineBuf(stdout);
|
|
if ((whoami = strrchr(argv[0], '/')) == NULL)
|
|
{
|
|
whoami = argv[0];
|
|
}
|
|
else
|
|
{
|
|
++whoami;
|
|
}
|
|
// For libtool's sake....
|
|
if (strncmp(whoami, "lt-", 3) == 0)
|
|
{
|
|
whoami += 3;
|
|
}
|
|
|
|
if ((argc < 3) || (argc > 4))
|
|
{
|
|
usage();
|
|
}
|
|
|
|
try
|
|
{
|
|
int n = atoi(argv[1]);
|
|
char const* filename1 = argv[2];
|
|
char const* arg2 = argv[3];
|
|
runtest(n, filename1, arg2);
|
|
}
|
|
catch (std::exception& e)
|
|
{
|
|
std::cerr << e.what() << std::endl;
|
|
exit(2);
|
|
}
|
|
|
|
return 0;
|
|
}
|