Fix irrString use-after-free with char-like assignment (operator=)

[swagtoy]

Thank you @appgurueu for suggesting use of resizing instead of making a temporary copy.

Technically fixes #15211, but I'd still like to greenlight that PR for merging.

This is a use-after-free. It is probably a bit serious but otherwise nothing dangerous as it's re-size dependent. This operator is used a lot throughout irrlicht.

My old code

Here's my old code which does a copy, if you need it.

template <class B>
string<T> &operator=(const B *const c)
{
  if (!c) {
	  clear();
	  return *this;
  }
  
  // no longer allowed!
  _IRR_DEBUG_BREAK_IF((void *)c == (void *)c_str());
  
  u32 len = calclen(c);
  B *copy = new B[len];
  std::copy(c, c+len, copy);
  str.resize(len);
  for (u32 l = 0; l < len; ++l)
      str[l] = (T)copy[l];
        
  delete [] copy;
  
  return *this;
}

[sfan5]

should be good

[swagtoy]

Once someone tests the early return (I did and it works), then it's ready to merge 👍

[appgurueu]

This is the logic I want to see. Bringing examples/AutomatedTest/test_string.cpp back could be considered but wouldn't be very useful given that we want to get rid of core::string anyways.

[swagtoy]

Im definitely against bringing the tests back for now. Let's move to our own string utility soon :)

[swagtoy]

It's ready for merge when anyone is ready

[SmallJoker]

Why do we need template <class B> ? Proper conversions should be done using wide_to_utf8 and utf8_to_wide. There's still string<T>::string(const T *const c, u32 length) (currently also using typename B) to copy as much data as needed - without strange trimming.

[swagtoy]

I'd say its more of a compatibility thing. It's just leftover junk from Irrlicht, so I don't want to bother wiping it over.

But I personally wouldn't have considered any of this and would've just hacked in utf8 checking

[swagtoy]

I'll fix changes when I get off work

[appgurueu]

Why do we need template <class B> ?

It's not trivial to remove (and needn't be refactored in this PR) because it's used for unsigned - signed conversions (which don't do anything though, essentially it's just a type cast). For example unsigned char to signed char. But that case is covered by checking for equal sizeofs so sfan5's code is good.

[sfan5]

because it's used for unsigned - signed conversions (which don't do anything though, essentially it's just a type cast)

I think it's used for some char -> wchar_t "conversions" too. Otherwise it would be easy to restrict.

[swagtoy]

Fixed kings. Sorry I can't git.

[swagtoy]

It's ready, for real, I hope.

[swagtoy]

It's ready, for real, I hope.

[swagtoy]

Tysm sfan for force pushing those, i was getting tired of this PR 😄

Thank you all for all the feedback and advice!