使用gzip_compressor会产生不同的文件大小

Question

I used gzip_compressor() to have compressed output file. I used two methods for this purpose. The common part is

std::ofstream traceOut;
traceOut.open("log.gz", std::ios_base::out);
struct traceRec {
  traceRec(uint64_t c) : cycle(c) {};
  uint64_t cycle;
};
void writeTrace(traceRec &rec)
{
  boost::iostreams::filtering_ostream o;
  o.push(boost::iostreams::gzip_compressor());
  o.push(traceOut);
  // METHOD 1 OR 2
}

Method 1

I use

 o.write(reinterpret_cast<const char*>(&rec.cycle), sizeof(rec.cycle));

With this implementation, the file size is 380K!!

Method 2

I use

 traceOut << rec.cycle << std::endl;

With this implementation, the file size is 78K!!

So why they have different size?? Another thing is that if I don't use the gzip_compressor and directly write to the file

std::ofstream traceOut;
traceOut.open("log.gz", std::ios_base::out);
...
traceOut << rec.cycle << std::endl;

The file size will be 78K.

So there are two problems:

1- Using or not using gzip_compressor has no effect on file size

2- Different implementations for using gzip_compressor yield different file sizes

Any idea about that?

Answer 1

operator << is likely using the textual representation of the number, while the write method take the complete variable size.

So if you have for example a cycle that's "13", in the "write" case, you'll consume 8 bytes, while you'll consume only 2 in the textual representation.

When compressed, the effect is even more dramatic, because when writing numbers as text, only 10 characters are used, (very very low entropy), so it's highly redundant and compressible.

On the other size, if your cycle counter is often very big (> 99999999), then the write method will gives better compression.