TDataSource proposal

interface_samples.cxx

// TDataSource
//
// The problem it tries to solve:
// - accessing specific columns of any tabular data-set through a uniform interface
// - allow thread-safe parallel processing of the data (by slicing the data-set in nThreads pieces)
// - offer the simplest interface possible, ideally iterator-based

// 1. -- build the data source
TDataSource s("file.root");

// 2. -- grab a range to iterate over + handles to the desired columns
/* c++17    */ auto [r, a, b] = s.GetReader<int, double>("a", "b");
/* or also  */ auto [r, a, b] = s.GetReader({int(), "a"}, {double(), "b"});
/* c++11    */ TDS::Reader<int, double> r;
               int a;
               double b;
               std::tie<r, a, b> = s.GetReader<int,double>("a", "b");
/* w/ macro */ TDS_GET_READER(s, r, int, a, double, b);

// 3. -- loop over the dataset (n will contain the row number)
for(auto n : range)
   std::cout << n << " " << a << " " << b << "\n";

// parallel reading
// grab two iterators over separate halves of the data
// two threads can safely use one iterator each
auto [r1, x1, y1] = s.GetReader<int, double>("a", "b", 2, 0); // split in two, give me first half
auto [r2, x2, y2] = s.GetReader<int, double>("a", "b", 2, 1); // split in two, give me second half

auto loop = [](auto &r, auto &a, auto &b) {
   for (auto n : r) std::cout << n << " " << a << " " << b << "\n";
};
std::thread t1(loop, r1, a1, b1);
std::thread t2(loop, r2, a2, b2);

// parallel reading 2
// this time each thread is responsible of grabbing its own iterator
auto loop = [](unsigned slot) {
   auto[r, x, y] = s.GetReader<int, double>("a", "b", nThreads, slot);
   for (auto n : r) std::cout << n << " " << a << " " << b << "\n";
};

for(auto i = 0u; i < nThreads; ++i)
   std::thread(loop, i);