C++ target: escape ? character to prevent accidental trigraphs

[felixn]

If grammars contain something like ?? as a lexer rule (e.g. CONDITION_CONVERSION: '??';), the generated C++ code will contain something like "'??'" in _literalNames. This can cause compiler warnings or even unintended behaviour, because sequences starting with two ? characters are interpreted as trigraphs.

Using an override of shouldUseUnicodeEscapeForCodePointInDoubleQuotedString in CppTarget
shouldUseUnicodeEscapeForCodePointInDoubleQuotedString in Target was changed from private static to protected to allow overriding

(It's possible to avoid this issue by using C++17 (drops support for trigraphs) and -Wno-trigraphs, but escaping the ? character is a universal fix).

[felixn]

Ping @mike-lischke

[mike-lischke]

Can't you switch off trigraphs in every C++ compiler? This is a feature so rarely used that it seems unnecessary to waste CPU cycles to provide a fix in ANTLR4. But if you deem it important enough then we can take it in.

[felixn]

Hi Mike,
it looks like it's only possible to disable trigraphs by selecting C++17 or higher or by selecting a "std" mode instead of a "gnu" mode. See
https://clang.llvm.org/docs/UsersManual.html#differences-between-various-standard-modes
or
https://gcc.gnu.org/onlinedocs/cpp/Initial-processing.html

Either way, my suggestion (implemented in this PR) is for ANTLR to never generate trigraphs - so we don't force the user to use certain compiler settings / disable certain warning.
@parrt - what's your take on this?

[parrt]

I'll defer to @mike-lischke

[mike-lischke]

A fix for a really marginal problem, but OK.

[felixn]

Thanks for the positive feedback.
Could you please trigger the merge? I don't have write access.
(Same with #2806)