Improve build instructions and cleanup code

[Stephan T. Lavavej (StephanTLavavej)]

• Change build.x64 to out\build\x64, fixing README.md: How To Consume inconsistent with How To Build With A Native Tools Command Prompt #518.
  • This significantly improves usability as .gitignore lists /out/.
• Add "build" to skipped_directories in validate.cpp.
  • This essentially replicates the information in .gitignore, where /build/ was added by Add "/build/" to the .gitignore #581.
• Consistently test #if !_HAS_EXCEPTIONS.
  • This still isn't a supported mode.
• _RAISE was never used for real exceptions.
• Change unsigned to unsigned int.
  • There were a few cases (outside of LLVM-derived tests) where we weren't following our usual convention. I'm cleaning them up, even within tr1.
  • Additionally, change a couple of C casts to static_casts, slowly working towards src: Change C casts to C++ casts #184.
• Use {} for make_index_sequence and index_sequence_for.
  • In new code, we prefer using {} to construct tags, because () resembles a function call. See STL: Consistently use empty braces to construct tags like _Meow{} #468.
• cthread.cpp: Simplify a lengthy conditional operator.
• Extract __crtLCMapStringA calls and use C++ casts.
  • xstrxfrm.cpp: dstlen can be const.
  • Remove unnecessary parentheses (equality has well-known higher precedence than logical AND).
• Change ploc == 0 to ploc == nullptr.
• dllmain.cpp: [[maybe_unused]], preprocessor comments.
• Use #ifdef _CRT_APP for simplicity.
• Avoid _underscore_lowercase parameter names.
• Extract more assignments before comparisons.
  • Also scope size_t i to a for-loop and use preincrement.
  • Immediately initialize ans.
• More C++ casts.

[Billy O'Neal (BillyONeal)]

Change unsigned to unsigned int

I will high 5 James but not my call :P

[Stephan T. Lavavej (StephanTLavavej)]

The most important principle is consistency. If enough maintainers want to change our convention to say unsigned instead of unsigned int, then let's do that - but we shouldn't randomly switch between the two.

(It's already somewhat problematic that test code sometimes uses typename instead of class - that was me enjoying my personal preference in test code despite the history of product code - and that some tests use right const when product code uses left const. Now that we're accepting contributions from the community, global consistency is somewhat more important than before, I think.)

[Casey Carter (CaseyCarter)]

(It's already somewhat problematic that test code sometimes uses typename instead of class - that was me enjoying my personal preference in test code despite the history of product code - and that some tests use right const when product code uses left const.

You misspelled "const left" here.

Now that we're accepting contributions from the community, global consistency is somewhat more important than before, I think.)

As possibly the greatest offender / proponent of east const on the team, I agree. When there were fewer maintainers it made sense for a feature's "owner" to write and maintain the attendant tests in a slightly more personalized style. As the number of maintainers grows, this practice is becoming less defensible. I don't mind forcing the Internet to contribute code in our conventional style - that's the barrier for entry to any open-source project - but asking people to learn potentially different sets of undocumented conventions for each test they touch seems like a waste of the time and goodwill of our contributors.

EDIT: That said, I will never write short int or long long int and you can't make me!

[Stephan T. Lavavej (StephanTLavavej)]

Strongly agreed 😸

[Mahmoud Saleh (MahmoudGSaleh)]

LGTM