* unicode in `..`?

* EBCDIC support?

* Discuss putting ZEND_ATTRIBUTE_FORMAT back on zend_error() or create a new
  zend_error_ex() function that supports new specifiers

* Find all instances where unicode strings are compared with memcmp() and
  replace either with u_memcmpCodePointOrder() or ucol_strcoll()

* Opening a collator may return U_USING_DEFAULT_WARNING, U_USING_FALLBACK_WARNING

* Need to finish making HTTP input work as described in the design doc. It
  is almost there, but needs to handle conversion errors and provide a way to
  explicitly re-decode raw data with specified encoding (input filter,
  perhaps?). Also check for _charset_ request field which might be present.

* Optimize T_INLINE_HTML blocks conversion by either creating a converter
  cache or remembering the last used converter in the executor globals.

* What to do with binary string literals and runtime casting? Literals are in
  script_encoding, casting uses runtime_encoding. If they are different, bad
  stuff happens. Maybe those who do that stuff should suffer anyway.

* Control of fallback mappings in conversions.

* Figure out generic approach to locale validation/fallback.

* Constant registration/fetching should do identifier normalization.

* Make zend_u_str_case_fold() do only case-folding and nothing else. The
  normalization should be done by zend_normalize_identifier().

* Look at performance implications of identifier normalization. Measure
  performance difference when doing quickCheck + normalize versus simple
  normalize.

* USTR_MAKE("") should be estrndup(EMPTY_STR)

* See if ext/pcre can ba adjusted to allow operations on pure binary
  strings. Ideal mode would be: convert all IS_UNICODE to UTF-8, assume that
  binary strings with /u modifier are UTF-8, otherwise it's pure binary.

* Optimize for zend_string_to_unicode() and zend_unicode_to_string() for
  0-length strings.