It is is called from _PyUnicode_Dump() and so must not fail.

ascii=1 is no more reserved to PyASCIIObject. Use
PyUnicode_IS_COMPACT_ASCII(obj) to check if obj is a PyASCIIObject (as before).

Only use _PyUnicode_READY_REPLACE() on just created strings.

 * resize_compact() updates also wstr_len for non-ascii strings sharing wstr
 * resize_inplace() updates also utf8_len/wstr_len for strings sharing
   utf8/wstr

Fix also a compiler warning.

Simplify the test in wstr pointer in unicode_sizeof().

 * Document Unicode string states
 * Use _PyUnicode_CheckConsistency() to ensure that objects are always
   consistent.

not a unicode, instead of failing with a fatal error.

Use assertions in debug mode (provide better error messages).

 * Rename _PyUnicode_Resize() to unicode_resize()
 * unicode_resize() creates a copy if the string cannot be resized instead
   of failing
 * Optimize resize_copy() for wstr strings
 * Disable temporary resize_inplace()

real-time signals.

 * Ensure that the input string is ready
 * Raise a ValueError instead of of a fatal error