Cleanup pak extractor a tiny bit more

This commit is contained in:
Lily Tsuru 2022-07-09 04:28:10 -05:00
parent 5dbf7db274
commit 535f8c0571
1 changed files with 37 additions and 38 deletions

View File

@ -25,7 +25,6 @@ T LameRead(std::istream& is) {
return t;
}
/**
* Reads package files.
*/
@ -44,9 +43,8 @@ struct PackageReader {
}
void Init() {
is.seekg(-0xc, std::istream::end);
is.seekg(-sizeof(jmmt::PackageEofHeader), std::istream::end);
// Read the eof header
eofHeader = LameRead<jmmt::PackageEofHeader>(is);
// We ideally should be at the end of file after reading the eof header.
@ -61,6 +59,8 @@ struct PackageReader {
return;
}
// Read the string table, and hash every string out into a map.
// This is used to build our crc->filename mapping for this archive.
{
@ -73,8 +73,6 @@ struct PackageReader {
while(true) {
c = is.get();
// std::printf("%c\n", c);
if(c == '\0')
return s;
@ -94,33 +92,37 @@ struct PackageReader {
l = is.tellg();
// print out the creation of the crc/filename map for debugging
// std::printf("%s -> 0x%x\n", string.c_str(), jmmt::HashString(string.c_str()));
#if 0
std::printf("%s -> 0x%x\n", string.c_str(), jmmt::HashString(string.c_str()));
#endif
}
}
// Go to the start of the first pfil (skipping the pgrp we just read)
// after we setup our map.
//std::cout << "Group name: \"" << crcToFilename[group.groupNameCrc] << "\"\n";
// Go to the start of the first file chunk, skipping the group that we just read,
// after we have finished creating our CRC->filename map.
is.seekg(static_cast<std::streamsize>(eofHeader.headerStartOffset) + sizeof(jmmt::PackageGroup), std::istream::beg);
}
/**
* Get if the package file is invalid.
*
* \return false if file isn't invalid, true otherwise.
* \return True if file is invalid; false otherwise
*/
[[nodiscard]] bool Invalid() const {
return fileInvalid;
}
// Read a file chunk.
/**
* Read a single file chunk.
*/
void ReadFileChunk() {
if(fileInvalid)
return;
currChunk = LameRead<jmmt::PackageFile>(is);
if(currChunk.magic != jmmt::PackageFile::TypeMagic) {
fileInvalid = true;
return;
std::cout << "Invalid file chunk\n";
std::exit(1);
}
// Setup some variables
@ -133,7 +135,7 @@ struct PackageReader {
currFileName = crcToFilename[currChunk.filenameCrc];
std::cout << "Reading \"" << currFileName << "\".\n";
//std::cout << "Reading \"" << currFileName << "\".\n";
chunksLeft = currChunk.chunkAmount - 1;
fileWorkBuffer.resize(currChunk.fileSize);
@ -160,43 +162,35 @@ struct PackageReader {
/**
* Read a file from this package.
* \param[in] cb Called when file is read
* \param[in] cb Called when file is finished being read.
*/
template <class DoneCallback>
void ReadFile(DoneCallback&& cb) {
if(fileInvalid)
return;
// Read first file chunk.
// It's perfectly legal for this to be all we need to do.
ReadFileChunk();
// Read additional chunks, if required.
// Read additional chunks required to complete the file,
// if we (, well) have to.
for(auto i = 0; i < chunksLeft; ++i) {
// std::printf("reading additional chunk %d/%d\n", i, chunksLeft);
//std::cout << "Reading additional chunk " << i + 1 << '/' << chunksLeft << ".\n";
ReadFileChunk();
}
std::cout << "Read file \"" << currFileName << "\"\n";
std::cout << "Read file \"" << currFileName << "\" from archive.\n";
// Call user-provided callback
cb(DecompressedFile { .filename = currFileName,
.data = fileWorkBuffer });
// write file data to stdout (debugging!)
// std::cout.write(reinterpret_cast<const char*>(fileWorkBuffer.data()), fileWorkBuffer.size());
fileWorkBuffer.clear();
}
/**
* Read all possible files from this package.
* \param[in] cb Called when file is read
* \param[in] cb Called when file is finished being read.
*/
template <class DoneCallback>
void ReadFiles(DoneCallback&& cb) {
if(fileInvalid)
return;
for(auto i = 0; i < group.fileCount; ++i)
ReadFile(cb);
}
@ -220,7 +214,9 @@ struct PackageReader {
*/
std::map<jmmt::crc32_t, std::string> crcToFilename;
// file stuff
/**
* The amount of chunks left that we need to read to complete a file.
*/
uint32_t chunksLeft {};
// The name of the file we are processing.
@ -230,14 +226,15 @@ struct PackageReader {
jmmt::PackageFile currChunk {};
// File-sized work buffer used to store the file
// we're currently working on. Freed when a file is
// finished being extracted.
// we're currently working on.
//
// Freed when a file is extracted completely (after being copied).
std::vector<std::uint8_t> fileWorkBuffer;
};
int main(int argc, char** argv) {
if(argc != 2) {
std::cout << "Usage: " << argv[0] << " [path 2 JMMT PAK file]";
std::cout << "Usage: " << argv[0] << " [path to JMMT PAK file]";
return 1;
}
@ -265,15 +262,17 @@ int main(int argc, char** argv) {
std::ofstream ofs(outpath.string(), std::ofstream::binary);
if(!ofs) {
std::cout << "Could not open \"" << outpath.string() << "\".\n";
std::exit(1);
std::cerr << "Could not open \"" << outpath.string() << "\" for writing.\n";
return;
}
ofs.write(reinterpret_cast<const char*>(file.data.data()), static_cast<std::streampos>(file.data.size()));
ofs.close();
std::cout << "Wrote \"" << outpath.string() << "\".\n";
std::cout << "Wrote \"" << outpath.string() << "\" to disk.\n";
});
std::cout << "Finished extracting successfully.\n";
return 0;
}