- Bug: when run from /etc/ttys, the Palm might suggest a speed that
  the serial port can't handle. But ColdSync doesn't notice this in
  time to send back a reasonable speed.
- Review Alessandro Zummo's SPC.pm. Add POD. Improve some of the
  identifiers.
- Perhaps add "force_fast" and/or "force_slow" .coldsyncrc options?
- Devfs: At configure-time, check for the existence of /dev/.devfsd.
  This is created by Linux's devfs daemon. If it exists, assume
  HAVE_DEVFS.
	In libpconn/PConnection_serial.c, when opening the device, if 
- The Linux USB-to-serial driver appears to be flaky: I've seen it
  drop 3k characters at one point.
	When this happens, the symptom is that ColdSync hangs until
  the Palm times out. This is because ColdSync is expecting a netsync
  packet of a certain length, but it never gets as many bytes as it
  wants.
	It's possible to time out in netsync_read() (after 2 or 10
  seconds or something?). The Right Thing to do at that point would be
  to resend the DLP command and try again. However, the timeout would
  be noticed by dlp_recv_resp(), which doesn't know which DLP command
  was sent, and so can't resend it.
	The obvious fix would appear to be to write a
  dlp_process_cmd() function, which sends the DLP request and receives
  the response, possibly filling in the returned arguments
  appropriately. This function could then watch for a timeout and
  resend the request.
	Note that the netsync functions, when they time out, should
  return a unique error code meaning "timeout on a reliable protocol",
  since this means something different from a timeout over PADP.
- Windows appears to use an ioctl() on the USB device to get the
  serial number. Is this possible here as well? If so, how to
  communicate this to the appropriate levels?
- Would it be possible to implement a web clipping server?
- Add a ~/.palm/commands file, containing commands to execute on the
  next sync. If absent, this should default to "sync". That way, you
  don't have to kill 'usbd' or 'getty' to do a one-time backup.
	Problem: what to do if the sync is aborted for whatever
  reason? Nuke the "commands" file? Delete only those commands that
  were successfully executed, and leave the rest for the next time?
  Perhaps have one file for permanent commands (e.g., if the user
  always wants to do a backup, or always wants to run some command
  after the sync) and one for one-shots?
	Allow execution of external programs.
- Should all of the Dlp*() functions return type dlp_stat_t instead of
  int?
- For man pages that cover several functions, write short ".so
  somefunc.3" man pages for the secondary ones.
	Actually, it looks as if FreeBSD uses either the whatis
  database or hard links instead of .so files. Perhaps the Right Thing
  is to make links (hard or soft) at install time, perhaps from some
  database. Or better yet, have a script or something find the various
  .Nm lines before the first .Nd line, and make appropriate symlinks.
- PConnection_net.c uses tini_slp() et al. even when it isn't
  appropriate. Better to add a pconn->tini() or pconn->unwind()
  method.
- Perhaps initialize $CONDUIT_PATH to $PATH (from the environment).
- It'd be nice to add types to distinguish between bytes and letters,
  and between arrays of arbitrary data and strings of characters. This
  is for i18n and for the benefit of Japanese users and such.
- Use iconv(3) to convert command-line arguments to the appropriate
  encoding. Most Unix boxen use EUC for Japanese, but the Visor uses
  SJIS.
- Fix API bogosities in libpdb, libpconn.
- Man pages for libpdb functions.
- 'make install' should install libraries, headers.
- Redo error return status in {upload,download}_database() and
  friends, as per XXX comments.
- Separate i18n stuff for libpconn/libpdb into separate files. Figure
  out some way printing the appropriate messages ("coldsync" domain
  for the main program, "libpdb" for libpdb messages, etc.)
	dlopen() calls _init(). Can this be used?
- Separate libpconn and libpdb into separate packages.
	Separate i18n files.
- Tell conduits (esp. Sync conduits) whether this is a slow or fast sync.
- Consider allowing database-at-a-time syncing.
  Databases (or files?) named on the command line.
- Consider "none" database type, to specify conduits to run without an
  attached database. Preferably also allow user to specify whether it
  should run before or after other conduits.
- Bug: run_mode_Daemon() apparently ignores serial number when looking
  in /usr/local/etc/palms. When syncing a visor, it recognizes

	|Andrew Arensburger|2072|arensb|Palm Pilot Pro
	10A813V96PC0-H|Andrew Arensburger|2072|arensb|Palm V
	*Visor*|Andrew Arensburger|2072|arensb|Visor

  though not

	109X7B28GC66-F|Gorko the Invincible|31337|arensb|Palm III|/folks/arensb/.coldsyncrc.test
- Rethink the Dlp*() functions.
	They return a DLPSTAT_* error code, but this makes
  error-checking rather hairy. Would it be better if they just set
  palm_errno and returned -1?
- Consider adding a "transient" keyword to listen{} blocks, to
  indicate that failed open()s mean that the Visor hasn't been plugged
  in yet, so just sleep and retry until it's okay.
- Under Solaris, defining _XPG4_2 might induce standards-compliance,
  and make things happier. There's a comment in PConnection_net.c that
  says that this causes all sorts of other problems, though.
- With a Visor, if there's no pda{} block that matches exactly,
  ColdSync defaults to ~/.palm and username/userid from /etc/passwd.
	This seems like the Right Thing to do for the naive user, but
  means that if you've misspelled something in ~/.coldsyncrc , your
  stuff will go to the wrong place.
	Not sure what the fix should be. Add an "abort" pda flag,
  meaning "if you get this far, abort 'cos you overshot the real pda
  block"?
- When accepting a connection from a remote host, first open the TCP
  socket, and only then acknowledge the UDP packet. That way, if the
  TCP socket can't be opened for some reason, we can abort before
  acknowledging the UDP packet, and not promise the client a
  connection we can't offer. Besides, this is probably closer to what
  happens with inetd, where by the time the program is launched, inetd
  has already done most of the work of accepting the connection.
	Actually, it looks as if the UDP stuff is part of a completely
  different (name resolution? Windows?) protocol. If so, then ColdSync
  should only pay attention to the TCP stuff. The UDP name
  resolution(?) can be handled by a separate program, e.g. the one in
  pilot-link.
- Init mode should set the NetSync host and mask, if they're set in
  the pda block.
- "forward:" line in pda block: this is of the form
	forward: <destination> [<name>];
  where <name> is the name to use in NetSync wakeup packets. HotSync
  doesn't appear to use this anywhere else, which is why it's
  optional. Perhaps the server can multiplex based on the name it was
  given, like Apache.
	<destination> is usually a host name, though it can also be a
  host address, e.g., "192.168.1.2". Ideally, it should be possible to
  specify a network address and mask, e.g., "192.168.0.0/16"; or a
  network name, to be looked up with getnetbyname() (in this case,
  though, how do you get the mask?).
- "System MIDI Sounds.pdb" complains that
	I have a new, deleted record that is neither archived nor expunged
	What am I supposed to do?
  Investigate.
- NetSync forwarding: there's a comment in usb_select() hinting that
  select() on /dev/ugen0 is broken and always returns ready.
- NetSync should time out after a while: if no TCP connection received
  N seconds after accepting a UDP connection, stop waiting and abort.
	Should time out:
	- After sending response to UDP wakeup packet (waiting for TCP
  connection) (the wakeup may have been sent to a broadcast address,
  and the client decided to talk to another machine that responded
  first).
	- After sending UDP wakeup packet (no servers responding). Log
  this on the Palm.
	- Trying to establish TCP connection to server (it may be
  broken, only responding to UDP wakeup packets).
	- During the main part of the sync (server or client may be
  broken, stuck in infinite loop or something).
	- Probably other cases as well.
- Does NetSync support tickle packets? Or is this all handled by TCP?
- It'd be nice to have a tool to make sure that all files that CVS
  knows about are in the distribution somewhere.
- Add a SIGTERM handler, to catch Ctrl-C.
- Add support for TCP wrappers. RTFM hosts_access().
- Bug?: If the serial number matches (and is a real serial number),
  but the username/userid doesn't, ColdSync proceeds to the next
  entry. Arguably, this should be a definitive non-match.
- There ought to be cmp_init() etc., just as for the other protocols.
- Better description of ${prefix}/etc/coldsync.conf in the man page.
- Add support for CMP 2.0 (once I find a CMP 2.0 Palm)
- It should be possible to set the hostid in a config file. If it's
  set in /etc/coldsync.conf, then obviously it shouldn't be possible
  to override it in ~/.coldsyncrc .
- It would be nice if add_to_log() took printf()-like arguments.
	Reimplement this with snprintf(), now that it's always
  available.
- Persistent "Bad CRC" problems under Linux:
	As far as I can tell, the Linux serial driver is buggy, and
  drops characters. 'pilot-link' also sees bad CRCs; it just doesn't
  report them unless you compile with -DDEBUG.
	However, this isn't likely to be fixed any time soon, so need
  to increase robustness of ColdSync. In particular, make sure all
  sorts of insanity can be dealt with.
- Problem: old Palms, and Visors, don't have serial numbers. Hence, if
  a user has two Visors,
	coldsync -mI
  will find the first Visor PDA block listed in .coldsyncrc , and will
  (transparently and incorrectly) initialize from that, possibly
  clobbering existing information.
	Perhaps ought to be able to give the name of a PDA on the
  command line, e.g.:
	coldsync -mI "Second Visor"
- It might be nice to allow a conduit block to specify a name, rather
  than a creator/type pair, since those are usually more intuitive.
	Problem:
	conduit sync {
		type: fooX/barX;
		type: todo/DATA;
		name: FooBarDB;
	}
  How do "type:" and "name:" interact? Multiple "type:" lines are
  or-ed together. Should "name:" entries be or-ed as well?
	Counterintuitively, perhaps the "type:"s should be anded with
  the "name:"s. That is,
		type: aaaa/bbbb;
		type: cccc/dddd;
		name: FooDB;
		name: BarDB;
  means
		((type aaaa/bbbb) || (type cccc/dddd)) &&
		((name "FooDB")   || (name "BarDB"))

  This way, it's possible to express "databases of type xxxx/yyyy and
  with name "FooDB", which I don't think you can express under any
  other interpretation.
- Per-mode options
	It appears possible (under FreeBSD) to allow mode-specific
  options: have the getopt() loop in "config.c" stop after a
  "-m<mode>" option has been seen. Then, in run_mode_Foo():
	int arg;

	optind = opterr = 0;
	arg = getopt(argc, argv, <optstr>);

	It remains to be seen whether this is portable or not.
- Add a Bool check_palm_owner(pda_block, palm) (or something) function
  that verifies that the Palm (which is known to have the correct
  serial number) has the proper username/userid and whatnot. For
  standalone mode: if it doesn't match, don't sync.
- Bug: should be smarter about installing files: it'd be nice to make
  sure that there's enough memory.
- Possible feature: degraded-mode NetSync: if ColdSync is supposed to
  forward a connection to another host for NetSync, but can't, it
  could go into degraded mode: do a local sync to a temporary
  (mqueue-like?) directory, then try to sync with the real remote
  host.
	Is this even desirable? Would it cause problems if the user
  manages to sync with the real remote host before the local host
  manages to do so?
- Bug: Under FreeBSD, when both Visor and Rio 600 are plugged in,
  Coldsync doesn't find the Visor correctly.
- Add default flag to listen blocks.
- KDE korganizer just uses vCalendar format for both events and todo
  list. Uses special fields "X-PILOTID" and "X-PILOTSTAT", presumably
  for Palm syncing. Also "X-ORGANIZER:MAILTO:arensb@localhost", not
  sure what for.
	Find an appropriate Perl module to parse iCalendar format.
- Add a type for those four-letter creator types and such.
- Something to parse version numbers, as per <System/NetMgr.h> and
  <System/SystemMgr.h>
- Lock files while syncing.
- The daemon ought to have some way to communicate with actual users,
  so that they can be notified when a sync starts, be asked questions,
  and other fun stuff. OTOH, since this is Unix, it ought to be able
  to run unattended.
	Possibly the best way to do this is to have the daemon listen
  on a "control" port. Clients can connect to this port and be
  notified, or control the daemon.
	Maybe a Unix domain socket is the obvious way to do it, since
  then there's no question about which machine the connection comes
  from. If it's a TCP or UDP socket, however, you can monitor syncs
  going on on another machine (useful for nosy administrators?). Here
  again, TCP is probably more secure, but UDP is less likely to cause
  a problem if something dies.
- When a new database is installed, it might contain dirty records. At
  the very least, it needs to make its way to the backup directory. Of
  course, it's silly to upload the database, then download it again
  just for the sake of syncing it. Fix the conduit API (not just
  GenericConduit) to make sense in this situation.
	Presumably this is the equivalent of a slow sync, but don't
  bother downloading the database from the Palm.
- Maybe add (optional) support for long arguments with libgnugetopt?
- FreeBSD's /bin/sh is still broken (as of 4.0):
	echo "before"
	for dir in a b c; do
		test "$dir" = "." && echo "found dot"
	done
	echo "after"
  still exits right after the "done". Presumably, what needs to happen
  is for information about whether to exit with the "-e" flag to be
  carried upwards through the parse tree.
	Until this gets fixed, I don't want to use 'automake'.
- Should all of the library .h files be lumped under #include
  <palm/foo.h> ?
- In daemon mode, when setting the UID, use setuid() to set both the
  real and effective UID: once it is determined that this is
  "arensb"'s Palm, we want to relinquish all of our root-ly powers and
  become as much like "arensb" as possible.
	What about setgid()?
	What's initgroups()? Is it useful?
- "Timeout. Attempting to resend" when installing AmsterdamMaps.pdb
  (166368 bytes) (from Heineken's BarTrek package, or the PHP manual).
- There appears to be a timing(?) problem with 'xcopilot',
  specifically when terminating. ColdSync thinks everything is fine,
  but 'xcopilot' hangs. I think it thinks that ColdSync hasn't
  received the final ACK. If you give the data time to drain, it
  appears to work fine, though. Maybe it's just a problem with the tty
  devices.
- Additional arguments on command line: if an argument is of the form
  "FOO=bar", it should set the variable FOO to the value "bar".
- Should variables set in .coldsyncrc be propagated to the
  environment? Should this be done as soon as they're set, or when the
  conduit is run?
- Should send mail to the user with a report of the sync (in daemon
  mode). Probably just add an option for this.
- Conduit for MH alias file: look for lines of the form
	alias:	<address> Firstname Lastname
  (where the "<" and ">" are literal). If $record->{firstName} . " " .
  $record->{name} eq "Firstname Lastname", then update the e-mail
  address.
- If I ever find a database that uses sort blocks, for testing, it
  might be a good idea to sync them as well.
- Perhaps status codes 6xx can be used for machine-readable
  informational messages, e.g., "NN% complete".
- Write 'mung-script' (or something) utility that takes a script
  pathname and an interpreter path, and mungs the #! path work
  properly. Make sure to use "#!/bin/sh" and not "#! /bin/sh", as per
  portability comments in 'automake' documentation.
- No-brainer configuration: make "conduit -config" part of the spec.
  Allow the user to add "CONDUITDIR=/conduit/dir" to .coldsyncrc.
  Then, when ColdSync runs, it runs "/conduit/dir/foo -config" for
  every conduit in /conduit/dir and saves the results in
  /conduit/dir/cache. /conduit/dir/cache is then read in, so that the
  configuration is the concatenation of ~/.coldsyncrc and
  /conduit/dir/cache.
	This way, if a new conduit is released, the naive user can
  just put it in /conduit/dir, and it'll work with its default
  configuration. A more sophisticated user can add a .coldsyncrc entry
  manually.
	Obviously, need the "final" flag in conduit blocks, to prevent
  unwanted conduits from running.
- The 'prompt_for_hotsync' argument to new_PConnection() is a hack.
  That whole function needs to be rethought. Would it be better to
  separate it into new_serial_PConnection(), new_usb_PConnection(),
  new_tcp_PConnection()? Or pass it a union that gives the
  connection-specific arguments?
- In config.c, when creating directories: ought to do ~-expansion on
  directory names: as in csh, only expand leading "~" or "~user".
- In .coldsyncrc, should be able to associate a name (or alias) with a
  pda block. In addition, should be able to specify that a given
  conduit block only applies to a given PDA.
- Installing new databases is suboptimal. Currently, it sends the
  database over, then later downloads it all over again. There's got
  to be a better way to do this.
	Try:
	Install the new database on the Palm.
	Write it (possibly under a new name) to ~/.palm/backup[1].
	Run fetch conduits.
	Run sync conduits.
	Run dump conduits.
  [1] What if the database already exists?

	Is it necessary or desirable to force a slow sync for this
  database?
- Improve fault-tolerance and error handling:
	Kill ColdSync with Ctrl-C. Have it send a valid cancel to the
  Palm.
	Fill the Palm up. See what happens.
- Dump conduit: read all of the completed "to do" items and append
  them to a text file, thus generating a status report. Optionally
  delete and/or expunge them.
- Fetch conduits: purge all completed "to do" items, and all
  appointments older than N days (where N is configurable).
	Delete and archive events by default. Allow an option to
  expunge them.
- If a conduit dies (e.g., "memo-kjots" with flavor "fetch"), should
  print an error message giving some indication of which conduit said
  that.
- Work on archive2pdb
	Figure out why it insists on writing the .pdb as "MemoDB" and
  not "MemoDB.pdb".
	Add lots of command-line options to set output file, select
  which records will be restored, etc.
	Ought to be able to delete records.
	A GUI front end would be nice.
- Fix ColdSync.pm so that "-config" will also print which preferences
  the conduit knows about.
- Allow something like "ForceInstall: yes;" line in .coldsyncrc, to
  have the same effect as -I command-line option.
	Actually, add a block for global options.
- Consider an optimization: if the local database hasn't been
  modified, don't write it back to disk. This reduces the size of
  daily backups by a trivial amount, but will also make a couple of
  users happy.
	The obvious way to do this is to add a 'Bool dirty' flag to
  struct pdb. It is false initially. Fix the PDB manipulation
  functions such that they set the dirty flag if they modify the
  database.
	At the end of the local sync, if the dirty flag is still
  false, don't bother writing the database to disk.
- libpdb and libpconn are useful in their own right. Turn them into
  standalone packages. At least libpdb.
- Look into generating libpconn/dlp_{cmd,rpc}.{c,h} automatically.
  'rpcgen' looked like just the thing, but DLP packets are not a
  subset of XDR, so there are too many niggling problems.
- Perhaps the job of communicating with the Palm can be done in a
  separate (Posix) thread? That way, that thread can deal more
  naturally and more single-mindedly with things (e.g., timeouts,
  unexpected data packets, etc.). In particular, it would be able to
  send tickle packets at appropriate intervals.
- ColdSync.pm should open SPC file handle as appropriate.
- Add some "pref" lines to sample.coldsync.rc
- Rewrite 'send-mail' as a Sync conduit.
- Do something with Daniel Klein's <bilderzeitung@hotmail.com> Doc
  stuff.
- The $VERSION stuff in the Perl modules doesn't handle branch
  versions: revision 1.2.3.4 will set $VERSION to "1.002", rather than
  "1.002_003_004", which might be desirable. Is it worth fixing this?
	Will Perl 5.6 make this comment obsolet?