Static Analysis

[sean-parent]

This PR was started to address an issue reported by Covarity regarding the exception handling in the future reduction code. It led to a rewrite of the future reduction code, the inclusion of clang-tidy and clang-tidy fixes in the CMake setup, and several other changes:

API Changes

• package() will now reduce futures (future<future> -> future). All APIs that return futures now reduce the future.
• await() and await_for() take the future by rvalue-ref. They now take it by rvalue-ref with deprecated overloads for const-ref for compatibility. I'm moving away from implicit copies. Use stlab::copy() or std::move() as appropriate to upgrade to the new interface.
• Clang-tidy was run across the interfaces and made several changes, including adding [[nodiscard]] to many interfaces. In general, these should catch misuse.
• Added get_ready() API to future that can be used in a recover() clause or when the future is known to be ready. It asserts the future is ready and would be a race to invoke on a non-ready future.

Fixes

• task(const task&&) is now deleted, which avoids a bad recursive runtime error if you attempt to copy a task.
• When using future<> with C++20 coroutines, exceptions are now correctly reported.
• The mutable size property in forest is now atomic. A const forest can be safely shared across threads.

Improvements

• future reduction was completely rewritten and is considerably more efficient with 1 (instead of 3) heap allocations and 2 (instead of 3) synchronization points.
• Added cmake presents and a .clang-tidy file (with support for fixes). The code is clean for the default clang-tidy analyzers, all the cert-*, performance-*, and modernize*- checks for C++17.

[dabrahams]

Just reading the summary it seems like a lot of changes that could've been separate PR's, for example, a clang tidy pass seems like a different topic from fixing the bug.

[sean-parent]

Just reading the summary it seems like a lot of changes that could've been separate PR's, for example, a clang tidy pass seems like a different topic from fixing the bug.

Yes - it should have been about 5 PRs. It started with "could clang-tidy find the same issue as Coverity" (answer, not that I've been able to determine) - and snowballed.

[nickpdemarco]

FYI I ran clang-tidy locally over these changes and it seems to work. The presets are nice. Haven't had time yet to review the changes themselves, though.

[sean-parent]

Going to take the heat for pushing my own oversized PR...