-dealing with errors:
	-> replace most return -1 with sysfatal as long as it doesn't terminate the whole program but just a thread that can be restarted
-why is the break in xfer (comm.c) needed when reading from a tracker reply?
-forgerequest():
	-better overall algo (stupid dance with announce)
	-size of buf to allocate
-better propername(), dealing with other stupid chars
-tracker struct to store values like "interval" ?
-do a getbestr equivalent to use in comm.c:getelement
-verif hton() when send request is valid for big endian arch 
-smarter handling of keepalives
-'datadir' and 'datadir dir' to ctl
-how long do we keep open the files which have already received some pieces? 
	-> for the moment, never let open. open when needed to write, close immediately after. probably very lame and slow, but simplest.
-rethink and maybe split the Torrent struct (separate everything related to the datafiles?), it's getting bloated.
-check algo ok (preppieceslist) if data is an exact number of pieces
-cat pieces -> print bitfield
-make use of complete and incomplete from tracker reply
-why does rtorrent see <index><16384><length> (invalid piece) when asking for a piece it does not have? -> hence the connection closed
-make a func for update list (element removal)
-improve clean up in the case when we just kill the current thread
-create a pool of a few threads in advance, use them as needed, and keep the pool full by creating new threads everytime they're consumed.
-check why it happens that threads ask for the same pieces. 
-algo in writedata() could be more compact...
-fix last piece case in checkpiece()
-find out why demonoid and tpb kick me out. changing webfs ua and the peerid does not help.
-comm.c:hello():chanfree(cio) ?
-seeding:
	1) scan what we already have and check hashes
	2) create bitfield (and pieceslist)
	3) send bitfield during hello()
	3) create a coroutine that listens
	4) seed
	5) update our bitfield (and pieceslist) when appropriate
	send not interested?
-case when torrent.c:checkpiece returns not good
-shared pieces in torrent.c:scanpieces
-size of bitfield is not +1 when it's a multiple of 8.
-check nothing broken since writemsg(uchar) -> writemsg(char)
-broken with rtorrent now :/
-when expecting the bitfield, should be prepared for something else
-writes -> writen(?), same for reads