Toolset update: VS 2022 17.5 Preview 1, Python 3.11.0

[Stephan T. Lavavej (StephanTLavavej)]

• New 1ES Hosted Pool, containing Patch Tuesday.
• New toolset, VS 2022 17.5 Preview 1.
  • This is not yet required in , because the MSVC-internal build is still using MSVC 19.34.
• This also contains Python 3.11.0 (with major performance improvements, although it doesn't make a massive difference for our test runs, which are dominated by compile time).
  • Not updating the minimum required version, as the VS Installer still offers Python 3.9.7.
• Use PowerShell 7.3.0 while preparing the VM.
  • Note that PowerShell 7.3.0 can't be used locally while preparing the 1ES Hosted Pool. As mentioned in the Checklist for Toolset Updates, there's an active Azure PowerShell bug about this.
• Remove EDG compiler bug workarounds:
  • Update citation from VSO-1641993 "REPORTED: EDG __builtin_memcmp misbehavior, part 2" to VSO-1675318 "EDG __builtin_memcmp misbehavior, part 3".
  • Remove _CONSTEXPR20_STRING_LITERALS workaround.
  • Remove workaround in Dev10_816787_swap_vector_bool_elements.
• Overhaul the constexpr string/vector tests, which had accumulated a mess of EDG workarounds. 🙀 This was problematic because there were overlapping ifdefs, some became unnecessary due to fixes over time, some could be debug-only, and many were working around a real problem but the reported bugs were resolved, so no action was being taken.
  • The diff looks horrible (I'd recommend focusing on the end result, and trusting that I haven't damaged things along the way; in a few cases I needed to move test code around a bit). Ultimately, this irons out everything into a highly systematic workaround, slightly refined from what P0980R1_constexpr_strings was doing. When we're (1) EDG and (2) IDL=2 and (3) compile-time, we skip test code. This activates as much test code as possible.
  • I verified that each remaining workaround is necessary, and is working around the same error message, which I've reported as VSO-1674140 "EDG rejects constexpr string/vector in debug mode with 'attempt to access expired storage' errors".
  • One workaround is unique, in that it's specific to vector and applies to release mode too. That has been reported as VSO-1674154 "EDG rejects constexpr vector with 'invalid arithmetic on non-array pointer' error".

[Stephan T. Lavavej (StephanTLavavej)]

I'm mirroring this to the MSVC-internal repo - please notify me if any further changes are pushed.

[Stephan T. Lavavej (StephanTLavavej)]

I've pushed an additional commit to restore a workaround and update the citation to VSO-1675318 "EDG __builtin_memcmp misbehavior, part 3".