- The ISO 2022-JP encoder is now static as with all others; this is
slightly slower, but localises the encoder logic to its class
- Indexed encoders now cache pointer tables on first use, yielding
significant performance benefits
- Encoding multiple characters now uses fewer function calls, yielding
moderate performance benefits at the expense of slight complication
The encoder currently operates only on single code points, but will later be
expanded to operate on iterables to construct complete strings. For encodings
other than ISO 2022-JP this is merely a convenience, but the algorithm for
that encoding mandates that encoded strings terminate in a switch to ASCII
mode, which a single-character encoder cannot accomplish by itself.
- array_flip() retains the last duplicate, when we need the first
- Indexes are now prepared with a list of first-duplicate code points
to search before flipping
- This affected only U+3000 in GBK
- Big5 did not use array_flip(), but its list of override code points
did not include U+2561; Big5 now flips like the others
- EUC-JP had a long list of errors, but this encoding was not
previously released
- Shift_JIS' indexes are probably not correct, still
- UTF-16 needs to restore dirtyEOF after seeking
- gb18030 now tracks errors like other non-synchronizing encodings
- gb18030 could produce null when asked for a character
