LWG-3464 istream::gcount() can overflow

[futuarmo]

LWG-3464 istream::gcount() can overflow, so in case, if _Chcount is negative, return std::numeric_limits<streamsize>::max()

Fixes #1480

[barcharcraz]

I wouldn't mind tests for this, but I won't demand them.

[StephanTLavavej]

Thanks for mentioning testing - while thinking about whether it would be practical to test this, I realized that I was confused by reading LWG-3464, which said:

On a 32-bit platform files larger than 2GB can overflow the counter

This might not apply to our implementation. We always use long long:

STL/stl/inc/iosfwd

Line 49 in c2ab522

using streamsize = long long;

STL/stl/inc/istream

Line 664 in c2ab522

streamsize _Chcount; // the character count

STL/stl/inc/istream

Lines 602 to 604 in c2ab522

	_NODISCARD streamsize __CLR_OR_THIS_CALL gcount() const { // get count from last extraction
	return _Chcount;
	}

There's no practical way to exceed 2⁶³ - 1, and thus no practical way to test this.

However, we are affected by #388 - I'm not sure if that's relevant here. I would guess not, since this is directly incrementing an integer, but I'm not certain.

So, I wonder whether we need to make any changes here at all. That said, I believe that the PR clearly doesn't harm correctness, and I believe that it won't harm performance either (the function can be inlined, and the branch is trivially predictable).

[StephanTLavavej]

Thanks for implementing this LWG issue and making the Status Chart happier (when it's updated next week)! 😸