While encountering such things sucks, did you specify /source-charset? If not, MSVC is well within its right of what it's doing.

The compiler is required to support the 96 characters from the basic source set (specified in §2.3), and it is required to translate each character to a character in the basic source set in an implementation-defined manner. For literals, any character not in the basic source set is translated to a universal-character-name, again implementation-defined. This means they can basically do anything they like (even fail) as long as it's documented. Documentation says you are to use /source-charset for source files that contain extended characters not in the basic source set. So... that's that.

Yes, in phase 3, the compiler is actually required to replace an entire comment with a single space character, so whether or not weird characters appear in a comment shouldn't matter. But I guess if you already invoked undefined behavior during phase 1, that's no longer important.

Never had any issue with Unicode in source files (even in string literals). And I primarily use Visual Studio. Just now, I tried pasting an U+3000 Ideographic Space "?" into a comment. And guess what? No issues at all. Both nvcc and cl compiled it without issues. So something else must be wrong on your end...

The compiler is required to support the 96 characters from the basic source set (specified in §2.3), and it is required to translate each character to a character in the basic source set in an implementation-defined manner.

[...]

But I guess if you already invoked undefined behavior during phase 1, that's no longer important.

Implementation-defined != undefined

Implementation-defined != undefined

Implementation-defined != undefined

Which is not what I said. Implementation documents (implementation-defined) that you shall do X, and you don't do it. That's undefined behavior.

Oh, I misread that, sorry.

Countless random errors that make no sense, give no direction, and make you bald by your own bare hands.

Unicode, eh?

$ cat a.php
<?php
$??????? = 'magic';
print($??????? . "\n");
[blar@blar-linux ~]$ php a.php
magic

While Unicode identifiers can be abused horribly, they can also make code that is based on mathematical equations more readable, for example:

deltaX = x[1] - x[0]
deltaY = y[1] - y[0]
alpha = atan(deltaY / deltaX)

looks much better when written as

?x = x[1] - x[0]
?y = y[1] - y[0]
? = atan(?y / ?x)

(And I could even type that without issues due to the EurKey keyboard layout that I use...)

Other examples are domain specific symbols such as LATIN SMALL LETTER F WITH HOOK (ƒ) as a symbol for aperture.

A more Coding Horros related example: https://stackoverflow.com/questions/12692067/and-other-unicode-characters-in-identifiers-not-allowed-by-g

https://godbolt.org/g/pbuCUV