macosx specific zmalloc.c, uses malloc_size function in order to avoid to waste memory and time to put an additional header

shareobjectspoolsize implemented in reds.conf, in order to control the pool size when object sharing is on

ignore gcc warning about write() return code not checked. It is esplicitily this way since the "max number of clients reached" is a best-effort error

RubyRedis: set TCP_NODELAY TCP socket option to to disable the neagle algorithm. Makes a huge difference under some OS, notably Linux

Issue with redis-client used in scripts solved, now to check if the latest argument must come from standard input we do not check that stdin is or not a tty but the command arity

replication slave timeout when receiving the initial bulk data set to 3600 seconds, now that replication is non-blocking the server must save the db before to start the async replication and this can take a lot of time with huge datasets

critical bug about glueoutputbuffers=yes fixed. Under load and with pipelining and clients disconnecting on the middle of the chat with the server, Redis could block. Now it's ok