This library was motivated by the need of writing large numerical datasets with the following constraints:

• human-readable format (e.g. TSV)
• compile-time-configurable row "tuple" with different data types
• thread-safety (writing from multiple, possibly many threads in parallel)
• output efficiency (cheap caching + fast chunked output)

I hope the goals were reasonably achieved.

It can use three threading model for mutex:

• from C++11 standard (by default)
• from pthreads (set macro TSW_USE_POSIX_THREADS)
• no threading model, in case you want to use it non-threading application (set macro TSW_NO_THREADS)

In any case, the order of stored rows is preserved.

This small library is header-only. Just copy tsw.hh to your project and start using it. You need:

• a C++11-supporting compiler, tested ones include:
  • gcc 4.4.7, 4.8.3, 5.1.1 (should work for all 4.4+)
  • clang 3.5
• pthread (needed if used directly, needed in GCC on Linux)

• The I/O operation is done in the thread that arrives at the cache limit (and blocks further storing). So, in principle, don't expect each Store operation to be extremely fast - only most of them are. A better approach (planned?) would be to dedicate a separate thread to I/O.

• Strings don't yet get the proper treatment. Thus the TSVWriter does not check for the presence of delimiters in the string. This could be done, for sure...

#include "tsw.hh"

using namespace tsw;

int main() {
    tsw::TSVWriter<double, double> writer("out.tsv");
    writer.SetColumnNames("x", "x^2");
    for (int i = 1; i < 10; i++) {
        writer.Store(i, i * i);
    }
}

Output (out.tsv):

x       x^2
1       1
2       4
3       9
4       16
5       25
6       36
7       49
8       64
9       81
10      100  

See examples/std_thread.cc