An attempt to fix potential deadlock issues in Windows.

[mathgeekcoder]

Removes synchronization with worker threads on shutdown. Also removes the "search" for the main executor in the worker threads.

Instead we simply pass the main executor to the thread as a parameter. We also pass the underlying shared_ptr to avoid potential edge cases where reference count drops to zero before some threads initialize.

I made the run_worker static to avoid any confusion about this vs executor->ptr, and so it uses the shared_ptr to reference the shared memory.

The last worker thread will delete the shared memory, via the shared_ptr reference count.

This might fix issues mentioned in #1886 and #1044.

[mathgeekcoder]

Sample C++ code that reproduces the error on windows.

std::array<double, 2> lb = { -kHighsInf, -kHighsInf };
std::array<double, 2> ub = {  kHighsInf,  kHighsInf };

std::array<HighsInt, 2> index = { 0, 1 };
std::array<double, 2> value = { 0, 1 };

std::array<double, 2> lower = { 2, 0 };
std::array<HighsInt, 2> starts = { 0, 2 };
std::array<HighsInt, 4> indices = { 0, 1, 0, 1 };
std::array<double, 4> values = { -1, 1, 1, 1 };

Highs h;
h.setOptionValue("output_flag", false);
h.setOptionValue("threads", 20); // setting to large number makes it more likely to deadlock

h.addVars(2, lb.data(), ub.data());
h.changeColsCost(2, index.data(), value.data());
h.addRows(2, lower.data(), ub.data(), 4, starts.data(), indices.data(), values.data());

std::cout << "starting" << std::endl;

for (int i = 0; i < 100; i++) {
   std::thread([&] { h.run(); }).join();
   std::cout << i << std::endl;
}

std::cout << "Done" << std::endl;

[galabovaa]

Closing, changes are now in https://github.com/ERGO-Code/HiGHS/pull/1948