srv0start.c	InnoDB should now run with the same binary on Win95 and NT: checks OS version

os0file.c	Changed file write buffer size in os_file_set_size from 16 MB to 8 MB

ibuf0ibuf.c	Insert buffer calculated SQL NULL size wrong, causing overflow and assertion failure in some cases
log0log.c	Retrieve a dummy value from a function to prevent gcc crashing on HP-UX

ignore  	added Logs/* binary/*

into threads.polyesthetic.msg:/usr/local/src/my/3

to bdb/dist/s_rpc so that bk doesn't complain about the
rpc_server/db_server_proc.c file that gets overwritten (but needs to
be stored in BK).  Add a bdb/dist/template directory which is needed
by bdb/dist/s_recover.

Fixed bug in SELECT db1.table.* FROM db1.table,db2.table
Fixed bug in INSERT DELAYED when doing shutdown and a table was locked
Changed that tmp_table_size =4G-1 means unlimited.

into mysql.sashanet.com:/home/sasha/src/bk/mysql

    forgot to add result

    added a test for updates on user table

os0file.c	Fixed an extraneous error message at the start of db, changed from O_SYNC back to fdatasync because extremely slow on Sun

ut0dbg.h	Changed Innobase->InnoDB
os0file.c	InnoDB prints now the number of OS error in a file operation

into threads.polyesthetic.msg:/usr/local/src/my/3

Redhat's RPM's are looking for.

Submitted by Florin Andrei <florin@sgi.com>.

into mysql.sashanet.com:/home/sasha/src/bk/mysql

    fixed frame depth since we now are a frame deeper 
    on Alpha, use SP instead of FP

into threads.polyesthetic.msg:/usr/local/src/my/3

configure.in	Check if fdatasync exists
manual.texi	Updated defragmenting doc
manual.texi	Corrected fillfactor contract threashold for a page to 1/2

Updated qsort source from glibc.
Add debugging of keycache when EXTRA_DEBUG is specified

manual.texi	book errata.

    put back the things that the merge removed

limited support on Alpha - in general case, even with frame pointer,
stack trace on alpha is impossible without looking at the symbol table
frame pointer does get saved on the stack, but you have no idea where
and where the return address is saved. So the best we can do without
the symbol table is look for magic opcodes and try to guess how big 
each frame is and where the return address was hidden from the 
instruction parameters. In practice, we can usually go up 3-4 frames 
before we hit some nasty frame that the current code cannot figure
out. This is actually not too bad, especially when we already have the query

Also cleaned up messages, print more variables, tell the user of
how much memory mysqld could potentially use, and warn of
what can happen with default STACK_SIZE and a lot of connections if
coredump happens when there are more than 200 connections. 

into donna.mysql.fi:/home/heikki/mysqle

Fixes to InnoDB to compile on Windows
Fix for temporary InnoDB tables
Fixed bug in REPLACE()
Fixed sub char keys for InnoDB