* This is the case e.g. on NetBSD 10 for ARMv6 where Sci::Position == ptrdiff_t == long int,
  but obviously for other platforms as well, where it causes "invalid conversion"
  and "undefined symbol" errors.
  Scintilla was testing for aliasability by comparing the storage size with sizeof()
  or PTRDIFF_MAX == INT_MAX at the preprocessor level.
  This was fundamentally flawed.
* In LineVector<T>::InsertLines() we are now using the C++17 construct
  std::is_convertible_v<From*,To*> instead.
* We need RunStyles<ptrdiff_t> as well on the affected platforms.
  This is impossible to test for in a constant expression that can be used
  with the preprocessor.
  Also, it's not possible to conditionally instantiate templates.
  We tried to instantiate RunStyles for all scalar types that could be behind ptrdiff_t,
  but it was causing warnings on MSVC.
  Implicitly instantiating RunStyles would be possible, but is not desired.
  Therefore as a workaround, you can now define the PTRDIFF_DOESNT_ALIAS_INT
  macro when invoking the build system, to force instantiating RunStyles<ptrdiff_t>.
  When writing portable applications, you may have to use a compile-time check
  for checking aliasability of ptrdiff_t and int in order to define PTRDIFF_DOESNT_ALIAS_INT.

* SciTECO needs to avoid automatic scrolling as an optimization.
  While this works partially by avoiding certain messages like SCI_GOTOPOS (in favor of SCI_SETEMPTYSELECTION),
  there aren't alternatives for all messages that scroll.
  For instance SCI_UNDO will always call EnsureCaretVisible().
  Also, even if we could avoid all scrolling messages, there is no guarantee that will not suddenly
  and unexpectedly break in the future.
* Instead, every scrolling is now disabled except for an explicit SCI_SCROLLCARET.
  SciTECO can therefore use the more intuitive messages like SCI_GOTOPOS.
* SetLastXChosen() (choosing the X that caret sticks to) has been found to be a major slow down
  during UNDO and serves no purpose on SciTECO anyway - we have to implement such an algorithm
  at the SciTECO language level.
  We therefore simply disable SetLastXChosen().

* This sets the default key representations on every SCI_SETDOCPOINTER
  which is very inefficient considering the way that SciTECO uses the
  Q-Register view.
  Furthermore it meant, we had to reset the representations to their
  SciTECO versions again after every SCI_SETDOCPOINTER.
* This patch only does not cause problems because we initialize
  the representations anyway for every Scintilla view.
  This patch is not meant to be upstreamed!

Avoid processing characters multiple times by relying on from_chars munching
digit characters instead of searching for delimiters.

bidirectional mode which is used for accessibility on macOS with VoiceOver.
Avoid out-of-bounds access.

Enlarge bitmap to avoid visible blank background between items.