+** Don't use filter-model
+
+ tomo (major developer of SEMI) and akr (major developer of
+FLIM-FLAM) discussed about Emacs 20.3 problem related with SEMI and
+FLIM. They found essential problem with Emacs 20.3 are:
+
+ - Emacs 20.3 separates string-type to unibyte-string and
+ multibyte-string. Emacs 20.3 has enough APIs to treat them.
+
+ - Buffer has mode about interpretation of contents. Each mode is
+ designed to save semantics as characters. Namely buffer contains
+ unibyte-characters or multibyte-characters. One buffer can not
+ contain both representations.
+
+ - {decode|encode}-coding-{region|string} run in byte world. So it
+ is not harmonized with other API.
+
+ - It seems easy to write code in one mode or one world
+ (unibyte-string or multibyte-string). However it seems not easy
+ to write inter-mode program, such as SEMI.
+
+ - Byte <-> byte conversion, such as base64, quoted-printable, must
+ be run only with unibyte-mode.
+
+ - Byte <-> character conversion, such as
+ {decode|encode}-coding-region, should not be defined in single
+ buffer. Instead of them, decoder should read from unibyte buffer
+ and output to multibyte buffer. Similarly, encoder should read
+ from multibyte buffer and output to unibyte buffer.
+ `insert-buffer-substring' like API may be suitable. Anyway Emacs
+ introduces multiple representations, so it should be redesigned
+ based on multiple representation world model.
+
+ Anyway FLIM should introduce new APIs based on inter-representation
+world model. Conventional APIs should be implemented based on new
+APIs.
+
+ SEMI should abolish filter model and introduce new methods to
+display inline data. These methods should use new FLIM APIs based on
+inter-representation world model.