Changed default of file_io_threads to 4 and lock_wait_timeout to 50 seconds
  Added a missing break: if not specified, innodb_fast_shutdown got the same value as ..flush_log_at_trx commit

  Print progress info of the applying log records to the database phase in recovery
srv0srv.c:
  Do buffer pool flush and checkpoints more often to make recovery faster

  Increase table comment print size to 500 bytes to be able to print more foreign key constraint infos
ha_innobase.h:
  Change max key len back to 500 bytes because MySQL interpreter cannot handle more

  Allow foreign keys refer between fixed and var length strings

  Lower max key len to 3500 bytes because a secondary index record must also contain the primary key value

  Make sure no memory overrun of key buffer in range estimation

  Allow equal key values in a unique secondary index if the key value contains an SQL NULL

  Add warnings to recovery in some cases where log file might be corrupt and consequently recovery can be trusted less than normally

instead of DO RELEASE_LOCK()

  Add better instructions on how to specify innodb_data_file_path etc.

  Cleanup

  Do less statistics calculations for tables
dict0mem.h, row0mysql.c:
  Make calculation of new statistics less frequent, because the statistics in 3.23.44 involves many random disk reads

  Do not automatically calculate new statistics when a temporary table is used by MySQL in a query. The fact that some ORDER BY queries were slow may have been explained by this unnecessary statistics calculation.

  Add diagnostic code to find out why foreign key constraint creation sometimes fails

  Remove a theoretical overrun in adaptive hash index building

  Eliminate a Purify warning in a multithreaded test: the fields are not protected by a mutex, therefore they can be read uninitialized

added extra/mysql_install.c - will work on it in 4.0, but it does not hurt
to have it sit in 3.23 tree for now since it will eventually be backported 
to 3.23 anyway 

  Increase semaphore wait warning threshold to 240 seconds

  Improve table cardinality estimate if there are big BLOBs

  Make column names non-case-sensitive in referential constraints and put table and database names there in lower case in Windows

  Add diagnostic code to catch possible space id or page no errors in the log

  Add fault tolerance: when a log record contains nonsensical space id or page number, stop log scan there