Binary Serialization of std::bitset - c++

std::bitset has a to_string() method for serializing as a char-based string of 1s and 0s. Obviously, this uses a single 8 bit char for each bit in the bitset, making the serialized representation 8 times longer than necessary.
I want to store the bitset in a binary representation to save space. The to_ulong() method is relevant only when there are less than 32 bits in my bitset. I have hundreds.
I'm not sure I want to use memcpy()/std::copy() on the object (address) itself, as that assumes the object is a POD.
The API does not seem to provide a handle to the internal array representation from which I could have taken the address.
I would also like the option to deserialize the bitset from the binary representation.
How can I do this?

This is a possible approach based on explicit creation of an std::vector<unsigned char> by reading/writing one bit at a time...
template<size_t N>
std::vector<unsigned char> bitset_to_bytes(const std::bitset<N>& bs)
{
std::vector<unsigned char> result((N + 7) >> 3);
for (int j=0; j<int(N); j++)
result[j>>3] |= (bs[j] << (j & 7));
return result;
}
template<size_t N>
std::bitset<N> bitset_from_bytes(const std::vector<unsigned char>& buf)
{
assert(buf.size() == ((N + 7) >> 3));
std::bitset<N> result;
for (int j=0; j<int(N); j++)
result[j] = ((buf[j>>3] >> (j & 7)) & 1);
return result;
}
Note that to call the de-serialization template function bitset_from_bytes the bitset size N must be specified in the function call, for example
std::bitset<N> bs1;
...
std::vector<unsigned char> buffer = bitset_to_bytes(bs1);
...
std::bitset<N> bs2 = bitset_from_bytes<N>(buffer);
If you really care about speed one solution that would gain something would be doing a loop unrolling so that the packing is done for example one byte at a time, but even better is just to write your own bitset implementation that doesn't hide the internal binary representation instead of using std::bitset.

Answering my own question for completeness.
Apparently, there is no simple and portable way of doing this.
For simplicity (though not efficiency), I ended up using to_string, and then creating consecutive 32-bit bitsets from all 32-bit chunks of the string (and the remainder*), and using to_ulong on each of these to collect the bits into a binary buffer.
This approach leaves the bit-twiddling to the STL itself, though it is probably not the most efficient way to do this.
* Note that since std::bitset is templated on the total bit-count, the remainder bitset needs to use some simple template meta-programming arithmetic.

edit: The following does not work as intended. Appearently, "binary format" actually means "ASCII representation of binary".
You should be able to write them to a std::ostream using operator<<. It says here:
[Bitsets] can also be directly inserted and extracted from streams in binary format.

As suggested by guys at gamedev.net, one can try using boost::dynamic_bitset since it allows access to internal representation of bitpacked data.

I can't see an obvious way other than converting to a string and doing your own serialization of the string that groups chunks of 8 characters into a single serialized byte.
EDIT: Better is to just iterate over all the bits with operator[] and manually serialize it.

this might help you, it's a little example of various serialization types.
I added bitset and raw bit values, that can be used like the below.
(all examples at https://github.com/goblinhack/simple-c-plus-plus-serializer)
class BitsetClass {
public:
std::bitset<1> a;
std::bitset<2> b;
std::bitset<3> c;
unsigned int d:1; // need c++20 for default initializers for bitfields
unsigned int e:2;
unsigned int f:3;
BitsetClass(void) { d = 0; e = 0; f = 0; }
friend std::ostream& operator<<(std::ostream &out,
Bits<const class BitsetClass & > const m
{
out << bits(my.t.a);
out << bits(my.t.b);
out << bits(my.t.c);
std::bitset<6> s(my.t.d | my.t.e << 1 | my.t.f << 3);
out << bits(s);
return (out);
}
friend std::istream& operator>>(std::istream &in,
Bits<class BitsetClass &> my)
{
std::bitset<1> a;
in >> bits(a);
my.t.a = a;
in >> bits(my.t.b);
in >> bits(my.t.c);
std::bitset<6> s;
in >> bits(s);
unsigned long raw_bits = static_cast<unsigned long>(s.to_ulong());
my.t.d = raw_bits & 0b000001;
my.t.e = (raw_bits & 0b000110) >> 1;
my.t.f = (raw_bits & 0b111000) >> 3;
return (in);
}
};

Related

Optimize inserting std::uint32_t's into a std::vector<char>

I'm attempting to insert an array of unsigned ints into a std::vector.
Here is my current code:
auto add_chars(std::vector<char> & vec, unsigned val[]){
std::string tmp;
tmp.resize(11) // Max chars a uint can be represented by, including the '\n' for sprintf
for (auto x = 0; x< 10; x++){
auto char_count = sprintf(tmp.data(),"%u", val[x]);
vec.insert(vec.begin()+vec.size(),tmp.data(), tmp.data()+char_count);
}
}
int main(){
std::vector<char> chars;
unsigned val[10] {1,200,3,4,5,6000,7,8,9000};
add_chars(chars,val);
for (auto & item : chars){
std::cout << item;
}
}
This solution works, however I question its efficiency (and elegance).
Two questions:
Is there a more idiomatic way of doing this?
Is there a more efficient way of doing this?
*edit Fixed a bug in the code made while transferring over to here.
Also, i'm aware that '9000' can't be represented as 1 char, whats why im using the buffer and sprintf to generate multiple chars for the one uint.
Is there a more idiomatic way of doing this?
A character stream is idiomatic for this. Unfortunately, the standard only has a stream for building a string; not a vector. You can copy the string into a vector though. This is not most efficient way:
std::ostringstream ss;
unsigned val[10] {1,200,3,4,5,6000,7,8,9000};
for (auto v : val)
ss << v;
std::string str = ss.str();
// if you need a vector for some reason
std::vector<char> chars(std::begin(str), std::end(str));
Or you could write your own custom vector stream, but that will be a lot of boilerplate.
I would make a couple of changes (and fix the bug).
Firstly the number of digits in an integer is limited so there's no need to use a dynamic object like std::string, a simple char array will do. Since you are using uint32_t and decimal digits 10 characters are sufficient, 11 if you include a nul terminator.
Secondly sprintf and similar are inefficient because they have to interpret the format string, "%u" in your case. A hand written function to perform the conversion from uint32_t to digits would be more efficient.

Read uint8_t from std::stringstream as a numeric type

My understanding is that reading a uint8_t from a stringstream is a problem because the stringstream will interpret the uint8_t as a char. I would like to know how I can read a uint8_t from a stringstream as a numeric type. For instance, the following code:
#include <iostream>
#include <sstream>
using namespace std;
int main()
{
uint8_t ui;
std::stringstream ss("46");
ss >> ui;
cout << unsigned(ui);
return 0;
}
prints out 52. I would like it to print out 46.
EDIT: An alternative would to just read a string from the stringstream and then convert the solution to uint8_t, but this breaks the nice chaining properties. For example, in the actual code I have to write, I often need something like this:
void foobar(std::istream & istream){
uint8_t a,b,c;
istream >> a >> b >> c;
// TODO...
}
You can overload the input operator>> for uint8_t, such as:
std::stringstream& operator>>(std::stringstream& str, uint8_t& num) {
uint16_t temp;
str >> temp;
/* constexpr */ auto max = std::numeric_limits<uint8_t>::max();
num = std::min(temp, (uint16_t)max);
if (temp > max) str.setstate(std::ios::failbit);
return str;
}
Live demo: https://wandbox.org/permlink/cVjLXJk11Gigf5QE
To say the truth I am not sure whether such a solution is problem-free. Someone more experienced might clarify.
UPDATE
Note that this solution is not generally applicable to std::basic_istream (as well as it's instance std::istream), since there is an overloaded operator>> for unsigned char: [istream.extractors]. The behavior will then depend on how uint8_t is implemented.
Please do not use char or unsigned char(uint8_t) if you want to read in a formatted way. Your example code and its result is an expected behavior.
As we can see from https://en.cppreference.com/w/cpp/io/basic_istream/operator_gtgt2
template< class Traits >
basic_istream<char,Traits>& operator>>( basic_istream<char,Traits>& st, unsigned char& ch );
This does "Performs character input operations".
52 is an ascii code for '4'. Which means that the stringstream has read only one byte and still ready to read '6'.
So if you want work in the desired way, you should use 2-byte or bigger integer types for sstream::operator>> then cast it to uint8_t - the exact way that you self-answered.
Here's a reference for those overloads.
https://en.cppreference.com/w/cpp/io/basic_istream/operator_gtgt
After much back and forth, the answer seems to be that there is no standard way of doing this. The options are to either read off the uint8_t as either a uint16_t or std::string, and then convert those values to uint8_t:
#include <iostream>
#include <sstream>
using namespace std;
int main()
{
uint8_t ui;
uint16_t tmp;
std::stringstream ss("46");
ss >> tmp;
ui = static_cast<uint8_t>(tmp);
cout << unsigned(ui);
return 0;
}
However, such a solution disregards range checking. So you will need to implement that yourself if you need it.

Parsing a binary file. What is a modern way?

I have a binary file with some layout I know. For example let format be like this:
2 bytes (unsigned short) - length of a string
5 bytes (5 x chars) - the string - some id name
4 bytes (unsigned int) - a stride
24 bytes (6 x float - 2 strides of 3 floats each) - float data
The file should look like (I added spaces for readability):
5 hello 3 0.0 0.1 0.2 -0.3 -0.4 -0.5
Here 5 - is 2 bytes: 0x05 0x00. "hello" - 5 bytes and so on.
Now I want to read this file. Currently I do it so:
load file to ifstream
read this stream to char buffer[2]
cast it to unsigned short: unsigned short len{ *((unsigned short*)buffer) };. Now I have length of a string.
read a stream to vector<char> and create a std::string from this vector. Now I have string id.
the same way read next 4 bytes and cast them to unsigned int. Now I have a stride.
while not end of file read floats the same way - create a char bufferFloat[4] and cast *((float*)bufferFloat) for every float.
This works, but for me it looks ugly. Can I read directly to unsigned short or float or string etc. without char [x] creating? If no, what is the way to cast correctly (I read that style I'm using - is an old style)?
P.S.: while I wrote a question, the more clearer explanation raised in my head - how to cast arbitrary number of bytes from arbitrary position in char [x]?
Update: I forgot to mention explicitly that string and float data length is not known at compile time and is variable.
If it is not for learning purpose, and if you have freedom in choosing the binary format you'd better consider using something like protobuf which will handle the serialization for you and allow to interoperate with other platforms and languages.
If you cannot use a third party API, you may look at QDataStream for inspiration
Documentation
Source code
The C way, which would work fine in C++, would be to declare a struct:
#pragma pack(1)
struct contents {
// data members;
};
Note that
You need to use a pragma to make the compiler align the data as-it-looks in the struct;
This technique only works with POD types
And then cast the read buffer directly into the struct type:
std::vector<char> buf(sizeof(contents));
file.read(buf.data(), buf.size());
contents *stuff = reinterpret_cast<contents *>(buf.data());
Now if your data's size is variable, you can separate in several chunks. To read a single binary object from the buffer, a reader function comes handy:
template<typename T>
const char *read_object(const char *buffer, T& target) {
target = *reinterpret_cast<const T*>(buffer);
return buffer + sizeof(T);
}
The main advantage is that such a reader can be specialized for more advanced c++ objects:
template<typename CT>
const char *read_object(const char *buffer, std::vector<CT>& target) {
size_t size = target.size();
CT const *buf_start = reinterpret_cast<const CT*>(buffer);
std::copy(buf_start, buf_start + size, target.begin());
return buffer + size * sizeof(CT);
}
And now in your main parser:
int n_floats;
iter = read_object(iter, n_floats);
std::vector<float> my_floats(n_floats);
iter = read_object(iter, my_floats);
Note: As Tony D observed, even if you can get the alignment right via #pragma directives and manual padding (if needed), you may still encounter incompatibility with your processor's alignment, in the form of (best case) performance issues or (worst case) trap signals. This method is probably interesting only if you have control over the file's format.
Currently I do it so:
load file to ifstream
read this stream to char buffer[2]
cast it to unsigned short: unsigned short len{ *((unsigned short*)buffer) };. Now I have length of a string.
That last risks a SIGBUS (if your character array happens to start at an odd address and your CPU can only read 16-bit values that are aligned at an even address), performance (some CPUs will read misaligned values but slower; others like modern x86s are fine and fast) and/or endianness issues. I'd suggest reading the two characters then you can say (x[0] << 8) | x[1] or vice versa, using htons if needing to correct for endianness.
read a stream to vector<char> and create a std::string from this vector. Now I have string id.
No need... just read directly into the string:
std::string s(the_size, ' ');
if (input_fstream.read(&s[0], s.size()) &&
input_stream.gcount() == s.size())
...use s...
the same way read next 4 bytes and cast them to unsigned int. Now I have a stride.
while not end of file read floats the same way - create a char bufferFloat[4] and cast *((float*)bufferFloat) for every float.
Better to read the data directly over the unsigned ints and floats, as that way the compiler will ensure correct alignment.
This works, but for me it looks ugly. Can I read directly to unsigned short or float or string etc. without char [x] creating? If no, what is the way to cast correctly (I read that style I'm using - is an old style)?
struct Data
{
uint32_t x;
float y[6];
};
Data data;
if (input_stream.read((char*)&data, sizeof data) &&
input_stream.gcount() == sizeof data)
...use x and y...
Note the code above avoids reading data into potentially unaligned character arrays, wherein it's unsafe to reinterpret_cast data in a potentially unaligned char array (including inside a std::string) due to alignment issues. Again, you may need some post-read conversion with htonl if there's a chance the file content differs in endianness. If there's an unknown number of floats, you'll need to calculate and allocate sufficient storage with alignment of at least 4 bytes, then aim a Data* at it... it's legal to index past the declared array size of y as long as the memory content at the accessed addresses was part of the allocation and holds a valid float representation read in from the stream. Simpler - but with an additional read so possibly slower - read the uint32_t first then new float[n] and do a further read into there....
Practically, this type of approach can work and a lot of low level and C code does exactly this. "Cleaner" high-level libraries that might help you read the file must ultimately be doing something similar internally....
I actually implemented a quick and dirty binary format parser to read .zip files (following Wikipedia's format description) just last month, and being modern I decided to use C++ templates.
On some specific platforms, a packed struct could work, however there are things it does not handle well... such as fields of variable length. With templates, however, there is no such issue: you can get arbitrarily complex structures (and return types).
A .zip archive is relatively simple, fortunately, so I implemented something simple. Off the top of my head:
using Buffer = std::pair<unsigned char const*, size_t>;
template <typename OffsetReader>
class UInt16LEReader: private OffsetReader {
public:
UInt16LEReader() {}
explicit UInt16LEReader(OffsetReader const or): OffsetReader(or) {}
uint16_t read(Buffer const& buffer) const {
OffsetReader const& or = *this;
size_t const offset = or.read(buffer);
assert(offset <= buffer.second && "Incorrect offset");
assert(offset + 2 <= buffer.second && "Too short buffer");
unsigned char const* begin = buffer.first + offset;
// http://commandcenter.blogspot.fr/2012/04/byte-order-fallacy.html
return (uint16_t(begin[0]) << 0)
+ (uint16_t(begin[1]) << 8);
}
}; // class UInt16LEReader
// Declined for UInt[8|16|32][LE|BE]...
Of course, the basic OffsetReader actually has a constant result:
template <size_t O>
class FixedOffsetReader {
public:
size_t read(Buffer const&) const { return O; }
}; // class FixedOffsetReader
and since we are talking templates, you can switch the types at leisure (you could implement a proxy reader which delegates all reads to a shared_ptr which memoizes them).
What is interesting, though, is the end-result:
// http://en.wikipedia.org/wiki/Zip_%28file_format%29#File_headers
class LocalFileHeader {
public:
template <size_t O>
using UInt32 = UInt32LEReader<FixedOffsetReader<O>>;
template <size_t O>
using UInt16 = UInt16LEReader<FixedOffsetReader<O>>;
UInt32< 0> signature;
UInt16< 4> versionNeededToExtract;
UInt16< 6> generalPurposeBitFlag;
UInt16< 8> compressionMethod;
UInt16<10> fileLastModificationTime;
UInt16<12> fileLastModificationDate;
UInt32<14> crc32;
UInt32<18> compressedSize;
UInt32<22> uncompressedSize;
using FileNameLength = UInt16<26>;
using ExtraFieldLength = UInt16<28>;
using FileName = StringReader<FixedOffsetReader<30>, FileNameLength>;
using ExtraField = StringReader<
CombinedAdd<FixedOffsetReader<30>, FileNameLength>,
ExtraFieldLength
>;
FileName filename;
ExtraField extraField;
}; // class LocalFileHeader
This is rather simplistic, obviously, but incredibly flexible at the same time.
An obvious axis of improvement would be to improve chaining since here there is a risk of accidental overlaps. My archive reading code worked the first time I tried it though, which was evidence enough for me that this code was sufficient for the task at hand.
I had to solve this problem once. The data files were packed FORTRAN output. Alignments were all wrong. I succeeded with preprocessor tricks that did automatically what you are doing manually: unpack the raw data from a byte buffer to a struct. The idea is to describe the data in an include file:
BEGIN_STRUCT(foo)
UNSIGNED_SHORT(length)
STRING_FIELD(length, label)
UNSIGNED_INT(stride)
FLOAT_ARRAY(3 * stride)
END_STRUCT(foo)
Now you can define these macros to generate the code you need, say the struct declaration, include the above, undef and define the macros again to generate unpacking functions, followed by another include, etc.
NB I first saw this technique used in gcc for abstract syntax tree-related code generation.
If CPP is not powerful enough (or such preprocessor abuse is not for you), substitute a small lex/yacc program (or pick your favorite tool).
It's amazing to me how often it pays to think in terms of generating code rather than writing it by hand, at least in low level foundation code like this.
You should better declare a structure (with 1-byte padding - how - depends on compiler). Write using that structure, and read using same structure. Put only POD in structure, and hence no std::string etc. Use this structure only for file I/O, or other inter-process communication - use normal struct or class to hold it for further use in C++ program.
Since all of your data is variable, you can read the two blocks separately and still use casting:
struct id_contents
{
uint16_t len;
char id[];
} __attribute__((packed)); // assuming gcc, ymmv
struct data_contents
{
uint32_t stride;
float data[];
} __attribute__((packed)); // assuming gcc, ymmv
class my_row
{
const id_contents* id_;
const data_contents* data_;
size_t len;
public:
my_row(const char* buffer) {
id_= reinterpret_cast<const id_contents*>(buffer);
size_ = sizeof(*id_) + id_->len;
data_ = reinterpret_cast<const data_contents*>(buffer + size_);
size_ += sizeof(*data_) +
data_->stride * sizeof(float); // or however many, 3*float?
}
size_t size() const { return size_; }
};
That way you can use Mr. kbok's answer to parse correctly:
const char* buffer = getPointerToDataSomehow();
my_row data1(buffer);
buffer += data1.size();
my_row data2(buffer);
buffer += data2.size();
// etc.
I personally do it this way:
// some code which loads the file in memory
#pragma pack(push, 1)
struct someFile { int a, b, c; char d[0xEF]; };
#pragma pack(pop)
someFile* f = (someFile*) (file_in_memory);
int filePropertyA = f->a;
Very effective way for fixed-size structs at the start of the file.
Use a serialization library. Here are a few:
Boost serialization and Boost fusion
Cereal (my own library)
Another library called cereal (same name as mine but mine predates theirs)
Cap'n Proto
The Kaitai Struct library provides a very effective declarative approach, which has the added bonus of working across programming languages.
After installing the compiler, you will want to create a .ksy file that describes the layout of your binary file. For your case, it would look something like this:
# my_type.ksy
meta:
id: my_type
endian: be # for big-endian, or "le" for little-endian
seq: # describes the actual sequence of data one-by-one
- id: len
type: u2 # unsigned short in C++, two bytes
- id: my_string
type: str
size: 5
encoding: UTF-8
- id: stride
type: u4 # unsigned int in C++, four bytes
- id: float_data
type: f4 # a four-byte floating point number
repeat: expr
repeat-expr: 6 # repeat six times
You can then compile the .ksy file using the kaitai struct compiler ksc:
# wherever the compiler is installed
# -t specifies the target language, in this case C++
/usr/local/bin/kaitai-struct-compiler my_type.ksy -t cpp_stl
This will create a my_type.cpp file as well as a my_type.h file, which you can then include in your C++ code:
#include <fstream>
#include <kaitai/kaitaistream.h>
#include "my_type.h"
int main()
{
std::ifstream ifs("my_data.bin", std::ifstream::binary);
kaitai::kstream ks(&ifs);
my_type_t obj(&ks);
std::cout << obj.len() << '\n'; // you can now access properties of the object
return 0;
}
Hope this helped! You can find the full documentation for Kaitai Struct here. It has a load of other features and is a fantastic resource for binary parsing in general.
I use ragel tool to generate pure C procedural source code (no tables) for microcontrollers with 1-2K of RAM. It did not use any file io, buffering, and produces both easy to debug code and .dot/.pdf file with state machine diagram.
ragel can also output go, Java,.. code for parsing, but I did not use these features.
The key feature of ragel is the ability to parse any byte-build data, but you can't dig into bit fields. Other problem is ragel able to parse regular structures but has no recursion and syntax grammar parsing.

c++ append a struct into a binary file

I have my struct:
struct a
{
int x;
float f;
double d;
char c;
char s[50];
};
and I wish append each time into my timer schedule into a binary file.
// declaration
std::ofstream outFile;
// constructor:
outFile.open( "save.dat", ios::app );
// tick:
outFile << a << endl;
but inside the save.dat appears only this:
0C3A0000..0C3A0000..0C3A0000..0C3A0000..0C3A0000..0C3A0000..0C3A0000..0C3A0000..0C3A0000..
thanks in advance
What you're currently doing is writing the address of the struct definition.
What you want to do is use ostream::write
outfile.write(reinterpret_cast<char*>(&myStruct), sizeof(a));
This will work as long as your struct is a POD (Plain Old Data) type (which your example is). POD type means that all members are of fixed size.
If you on the other hand have variable sized members then you would need to write out each member one by one.
A sensible way to serialize custom objects is to overload your own output stream operator:
std::ostream & operator<<(std::ostream & o, const a & x)
{
o.write(reinterpret_cast<char*>(&x.x), sizeof(int));
o.write(reinterpret_cast<char*>(&x.f), sizeof(float));
/* ... */
return o;
}
a x;
std::ofstream ofile("myfile.bin", std::ios::binary | std::ios::app);
ofile << a;
This is still platform-dependent, so to be a bit safer, you should probably use fixed-width data types like int32_t etc.
It might also not be the best idea semantically to use << for binary output, since it's often used for formatted output. Perhaps a slightly safer method would be to write a function void serialize(const a &, std::ostream &);

How do I convert a big-endian struct to a little endian-struct?

I have a binary file that was created on a unix machine. It's just a bunch of records written one after another. The record is defined something like this:
struct RECORD {
UINT32 foo;
UINT32 bar;
CHAR fooword[11];
CHAR barword[11];
UNIT16 baz;
}
I am trying to figure out how I would read and interpret this data on a Windows machine. I have something like this:
fstream f;
f.open("file.bin", ios::in | ios::binary);
RECORD r;
f.read((char*)&detail, sizeof(RECORD));
cout << "fooword = " << r.fooword << endl;
I get a bunch of data, but it's not the data I expect. I'm suspect that my problem has to do with the endian difference of the machines, so I've come to ask about that.
I understand that multiple bytes will be stored in little-endian on windows and big-endian in a unix environment, and I get that. For two bytes, 0x1234 on windows will be 0x3412 on a unix system.
Does endianness affect the byte order of the struct as a whole, or of each individual member of the struct? What approaches would I take to convert a struct created on a unix system to one that has the same data on a windows system? Any links that are more in depth than the byte order of a couple bytes would be great, too!
As well as the endian, you need to be aware of padding differences between the two platforms. Particularly if you have odd length char arrays and 16 bit values, you may well find different numbers of pad bytes between some elements.
Edit: if the structure was written out with no packing, then it should be fairly straightforward. Something like this (untested) code should do the job:
// Functions to swap the endian of 16 and 32 bit values
inline void SwapEndian(UINT16 &val)
{
val = (val<<8) | (val>>8);
}
inline void SwapEndian(UINT32 &val)
{
val = (val<<24) | ((val<<8) & 0x00ff0000) |
((val>>8) & 0x0000ff00) | (val>>24);
}
Then, once you've loaded the struct, just swap each element:
SwapEndian(r.foo);
SwapEndian(r.bar);
SwapEndian(r.baz);
Actually, endianness is a property of the underlying hardware, not the OS.
The best solution is to convert to a standard when writing the data -- Google for "network byte order" and you should find the methods to do this.
Edit: here's the link: http://www.gnu.org/software/hello/manual/libc/Byte-Order.html
Don't read directly into struct from a file! The packing might be different, you have to fiddle with pragma pack or similar compiler specific constructs. Too unreliable. A lot of programmers get away with this since their code isn't compiled in wide number of architectures and systems, but that doesn't mean it's OK thing to do!
A good alternative approach is to read the header, whatever, into a buffer and parse from three to avoid the I/O overhead in atomic operations like reading a unsigned 32 bit integer!
char buffer[32];
char* temp = buffer;
f.read(buffer, 32);
RECORD rec;
rec.foo = parse_uint32(temp); temp += 4;
rec.bar = parse_uint32(temp); temp += 4;
memcpy(&rec.fooword, temp, 11); temp += 11;
memcpy(%red.barword, temp, 11); temp += 11;
rec.baz = parse_uint16(temp); temp += 2;
The declaration of parse_uint32 would look like this:
uint32 parse_uint32(char* buffer)
{
uint32 x;
// ...
return x;
}
This is a very simple abstraction, it doesn't cost any extra in practise to update the pointer as well:
uint32 parse_uint32(char*& buffer)
{
uint32 x;
// ...
buffer += 4;
return x;
}
The later form allows cleaner code for parsing the buffer; the pointer is automatically updated when you parse from the input.
Likewise, memcpy could have a helper, something like:
void parse_copy(void* dest, char*& buffer, size_t size)
{
memcpy(dest, buffer, size);
buffer += size;
}
The beauty of this kind of arrangement is that you can have namespace "little_endian" and "big_endian", then you can do this in your code:
using little_endian;
// do your parsing for little_endian input stream here..
Easy to switch endianess for the same code, though, rarely needed feature.. file-formats usually have a fixed endianess anyway.
DO NOT abstract this into class with virtual methods; would just add overhead, but feel free to if so inclined:
little_endian_reader reader(data, size);
uint32 x = reader.read_uint32();
uint32 y = reader.read_uint32();
The reader object would obviously just be a thin wrapper around pointer. The size parameter would be for error checking, if any. Not really mandatory for the interface per-se.
Notice how the choise of endianess here was done at COMPILATION TIME (since we create little_endian_reader object), so we invoke the virtual method overhead for no particularly good reason, so I wouldn't go with this approach. ;-)
At this stage there is no real reason to keep the "fileformat struct" around as-is, you can organize the data to your liking and not necessarily read it into any specific struct at all; after all, it's just data. When you read files like images, you don't really need the header around.. you should have your image container which is same for all file types, so the code to read a specific format should just read the file, interpret and reformat the data & store the payload. =)
I mean, does this look complicated?
uint32 xsize = buffer.read<uint32>();
uint32 ysize = buffer.read<uint32>();
float aspect = buffer.read<float>();
The code can look that nice, and be a really low-overhead! If the endianess is same for file and architecture the code is compiled for, the innerloop can look like this:
uint32 value = *reinterpret_cast<uint32*>)(ptr); ptr += 4;
return value;
That might be illegal on some architectures, so that optimization might be a Bad Idea, and use slower, but more robust approach:
uint32 value = ptr[0] | (static_cast<uint32>(ptr[1]) << 8) | ...; ptr += 4;
return value;
On a x86 that can compile into bswap or mov, which is reasonably low-overhead if the method is inlined; the compiler would insert "move" node into the intermediate code, nothing else, which is fairly efficient. If alignment is a problem the full read-shift-or sequence might get generated, outch, but still not too shabby. Compare-branch could allow the optimization, if test the address LSB's and see if can use the fast or slow version of the parsing. But this would mean penalty for the test in every read. Might not be worth the effort.
Oh, right, we are reading HEADERS and stuff, I don't think that is a bottleneck in too many applications. If some codec is doing some really TIGHT innerloop, again, reading into a temporary buffer and decoding from there is well-adviced. Same principle.. no one reads byte-at-time from file when processing a large volume of data. Well, actually, I seen that kind of code very often and the usual reply to "why you do it" is that the file systems do block reads and that the bytes come from memory anyway, true, but they go through a deep call stack which is high-overhead for getting a few bytes!
Still, write the parser code once and use zillion times -> epic win.
Reading directly into struct from a file: DON'T DO IT FOLKS!
It affects each member independently, not the whole struct. Also, it does not affect things like arrays. For instance, it just makes bytes in an ints stored in reverse order.
PS. That said, there could be a machine with weird endianness. What I just said applies to most used machines (x86, ARM, PowerPC, SPARC).
You have to correct the endianess of each member of more than one byte, individually. Strings do not need to be converted (fooword and barword), as they can be seen as sequences of bytes.
However, you must take care of another problem: aligmenent of the members in your struct. Basically, you must check if sizeof(RECORD) is the same on both unix and windows code. Compilers usually provide pragmas to define the aligment you want (for example, #pragma pack).
You also have to consider alignment differences between the two compilers. Each compiler is allowed to insert padding between members in a structure the best suits the architecture. So you really need to know:
How the UNIX prog writes to the file
If it is a binary copy of the object the exact layout of the structure.
If it is a binary copy what the endian-ness of the source architecture.
This is why most programs (That I have seen (that need to be platform neutral)) serialize the data as a text stream that can be easily read by the standard iostreams.
I like to implement a SwapBytes method for each data type that needs swapping, like this:
inline u_int ByteSwap(u_int in)
{
u_int out;
char *indata = (char *)&in;
char *outdata = (char *)&out;
outdata[0] = indata[3] ;
outdata[3] = indata[0] ;
outdata[1] = indata[2] ;
outdata[2] = indata[1] ;
return out;
}
inline u_short ByteSwap(u_short in)
{
u_short out;
char *indata = (char *)&in;
char *outdata = (char *)&out;
outdata[0] = indata[1] ;
outdata[1] = indata[0] ;
return out;
}
Then I add a function to the structure that needs swapping, like this:
struct RECORD {
UINT32 foo;
UINT32 bar;
CHAR fooword[11];
CHAR barword[11];
UNIT16 baz;
void SwapBytes()
{
foo = ByteSwap(foo);
bar = ByteSwap(bar);
baz = ByteSwap(baz);
}
}
Then you can modify your code that reads (or writes) the structure like this:
fstream f;
f.open("file.bin", ios::in | ios::binary);
RECORD r;
f.read((char*)&detail, sizeof(RECORD));
r.SwapBytes();
cout << "fooword = " << r.fooword << endl;
To support different platforms you just need to have a platform specific implementation of each ByteSwap overload.
Something like this should work:
#include <algorithm>
struct RECORD {
UINT32 foo;
UINT32 bar;
CHAR fooword[11];
CHAR barword[11];
UINT16 baz;
}
void ReverseBytes( void *start, int size )
{
char *beg = start;
char *end = beg + size;
std::reverse( beg, end );
}
int main() {
fstream f;
f.open( "file.bin", ios::in | ios::binary );
// for each entry {
RECORD r;
f.read( (char *)&r, sizeof( RECORD ) );
ReverseBytes( r.foo, sizeof( UINT32 ) );
ReverseBytes( r.bar, sizeof( UINT32 ) );
ReverseBytes( r.baz, sizeof( UINT16 )
// }
return 0;
}