vNext Planning

Salvage The Old vNext Branch

• Multithreading overhaul
• Iterator debugging overhaul (doubly-linked iterator chains, no proxy allocations)
  • Still need to overhaul deque and vector<bool>
  • Also simplifies checked/unchecked and modifiable/const, avoiding inheritance (which is non-standard)
• Remove /clr:pure
• Flat C exports
• Remove <experimental/filesystem>
  • Has a large separately compiled component.
• Remove <hash_map>, <hash_set>
  • This could be done without breaking ABI.
• Remove tr1
  • This could be done without breaking ABI.
• Other changes?
• Don't salvage "vcplatform" - it has been superseded

Major Overhauls

• <regex>
• <deque>
• <unordered_map>, <unordered_set>
• iostreams floating-point
• <tuple>
  • Specifically, implementing compression and eliminating the recursive inheritance

More Work

• vNext Issues
  • Especially this #169 comment.
  • #938 P0408R7 Efficient Access To basic_stringbuf's Buffer can be significantly simplified
• TRANSITION, ABI comments
• Consolidate satellite DLL and import LIB code back into the main DLL (and API loading code like for ICU); make sure to keep vector/string etc. out of the DLL/import LIB.
• Eradicate the last traces of XP/Vista/etc. support
• Remove _app DLLs
• Drop IDL=1
• <variant> changes:
  • Store index() + 1 instead of index() to so the range of possible stored values is contiguous for minor codegen improvements.
  • The hash value of a variant should depend on its index() as well as the stored value so variant<int, int>{in_place_index<0>, 42} and variant<int, int>{in_place_index<1>, 42} don't collide.
• [[no_unique_address]] will be unconditionally available, allowing us to replace _Compressed_pair and improve the representations of more types.
• iomanip.cpp doesn't need to be separately compiled
• Unnecessary padding in _Ctypevec (was #426)
• Likely unnecessary padding in basic_istream
• Ensure that future<const T> and promise<const T> work with the overhauled representation (was #2599)
• We should regularly add exports to prevent people from assuming they can use old DLLs
• The /ALTERNATENAME machinery is a trainwreck of complexity; consider removing it
• Figure out a "no exceptions" story and clean out the weirdness of _HAS_EXCEPTIONS=0.
• Number-suffixed types (_Meow2, _Meow3) should be cleaned up. The suffixes allowed us to change representations while avoiding ABI mismatch with the original names.

Compiler Work

• Remove /Za (effectively superseded by /permissive-)
• Remove /Zc:wchar_t-
• Increase minimum Standard version to C++17 or C++20?
• RTTI overhaul?
• EH overhaul possible? (Table-based x86?)
• Fix Empty Base Class Optimization
• Avoid ABI-breaking options (e.g. /vd2, /vmg, /J)
• How do we detect/prevent mixing?

ABI-Breaking Developer Community Tickets

• Bogus std::this_thread::sleep_for implementation
• Intermittent incorrect status returned by std::condition_variable::wait_for
• After calling std::promise::set_value_at_thread_exit, std::future::wait_for is blocked until promise's thread exits
• multiline C++
• VC++: std::exception::what has wrong signature
• TLS is not initialized correctly
• Debug version of STL is not excepion safe and causes terminate() to be called under low memory
• regex engine is extremely poor, resulting in stack overflows and other runtime failures with reasonable input
• size of std::mutex class on x64_86 is 80 bytes
• C++ iostreams over 2 GB doesn't work on x64 platform
• regex complexity exception
• Std::regex_error at memory location
• Bogus std::this_thread::sleep_for implementation
• Grouping within repetition causes regex stack error