This almost catches up with pre-PEP 393 performance, when decoding needed
only one pass.

raised when the wrapped raw file is non-blocking and the write would block.
Previous code assumed that the raw write() would raise BlockingIOError, but
RawIOBase.write() is defined to returned None when the call would block.
Patch by sbt.

raised when the wrapped raw file is non-blocking and the write would block.
Previous code assumed that the raw write() would raise BlockingIOError, but
RawIOBase.write() is defined to returned None when the call would block.
Patch by sbt.

Dump the wchar_t that we are going to decode and dump the locale

character is bigger than U+10FFFF and locale.localeconv() dumps the string
before decoding it.

Temporary hack to debug the issue #13441.

And PyUnicode_GetSize() => PyUnicode_GetLength()

using an assertion

PyUnicode_GET_SIZE() => PyUnicode_GET_LENGTH()

The index is a character index, not a index in a Py_UNICODE* string.

Patch written by Arfrever Frehtes Taifersar Arahesis.

Patch written by Arfrever Frehtes Taifersar Arahesis.

The index is a character index, not a index in a Py_UNICODE* string.

and check the string consistency.

_PyUnicode_CheckConsistency() doesn't check the hash anymore. It should be
possible to call this function even if hash(str) was already called.

the deprecate Py_UNICODE type

instead of calling the deprecated PyUnicode_EncodeUTF32() function

not only its minimum value

PyUnicode_AsUnicodeAndSize()