Sun May 22 17:13:16 CDT 2005 ( leakin at dfw dot nostrum dot com )

      * version 0.39: Simon Myers ( Smylers at stripey dot com ) sent
      a patch to allow Path::Class objects as source paths.  The previous
      error checking only allowed source to be a string or an array.
      This update bypasses the sanity-check if the source is a blessed
      object and now assumes the programmer knows what they are doing
      and that the object can be read as a string.

      Added support for the only-write-batch option in rsync-2.6.5-pre1

Wed Mar 30 22:06:04 CST 2005 ( leakin at dfw dot nostrum dot com )

      * version 0.38: Harald Flaucher ( Harald dot Flaucher at
      mondial dot de ) requested support for the --acls option, which
      is not part of rsync, but an included patch that some use.  His
      request inspired the idea for the 'literal' option in the module
      to allow passing arbitrary literal arguments to rsync so other
      patches or even newer versions of rsync can be used.  Also added
      support for the protocol option that was previously undocumented.

Sat Mar  5 19:59:01 CST 2005 ( leakin at dfw dot nostrum dot com )

      * version 0.37: added support for new options in 2.6.4-pre1:
      del, delete-before, delete-during, delay-updates, dirs, filter,
      fuzzy, itemize-changes, list-only, omit-dir-times, remove-sent-files,
      and max-size.  See the rsync manpage for full details on these options.

      Fixed a type-o that kept partial-dir from being handled properly.

      Bill Uhl ( buhl at web-services dot net ) provided a patch to
      split out the creation of the rsync command from the exec function.
      The new call (getcmd) returns a reference to an array of the command
      to be executed.  Bill is working on a graphical interface to rsync
      using this module and wants an alternate exec wrapper.  This version
      includes a variation on his patch.

Tue Jan 18 17:05:00 CST 2005 ( leakin at dfw dot nostrum dot com )

      * versioon 0.36: fix for "uninitialized value" warning submitted
      thru rt.cpan.org.  Thanks go to the anonymous Debian user (HTH).

Wed Dec  8 18:33:45 CST 2004 ( leakin at dfw dot nostrum dot com )

      * version 0.35: test 4 was failing due to bad pattern match in the
      test code. no functional changes (error message changed for 2.6.3).

Thu Aug 12 20:45:43 CDT 2004 ( leakin at dfw dot nostrum dot com )

      * version 0.34: added support for new options in 2.6.3: partial-dir,
      checksum-seed, keep-dirlinks, inplace, ipv4, and ipv6.

Tue Jun  1 12:59:36 CDT 2004 ( leakin at dfw dot nostrum dot com )

      * version 0.33: Fixed a type-o in parsing of quote-dst tag.
      Peter teStrake (peter dot testrake at tradingscreen dot com)
      pointed out that the exec method did not protect itself from a
      user-defined SIGCHLD handler and could miss the exit of rsync and
      report a false error.  A local reset of the handler should protect
      it and still leave the handler in place for the rest of the user
      code.

Sun Dec 21 03:18:26 CST 2003 ( leakin at dfw dot nostrum dot com )

      * version 0.32: Added support for new option in 2.6.0: --from0,
      --no-relative, --no-implied-dirs, and --files-from.  A new
      infun option has been added so a user-supplied function can print the
      list to --files-from=-.  See the rsync man page and the pod doc for
      more info.  Also fixed a bug underscores were not recognized as the
      hyphen equivalent for quote_dst, quote_src and path_to_rsync.
      An anonymous user submitted a fix where parens were needed due to
      precedence problems when srchost is defined.  The fix is included
      in this release.

Tue Jul 29 22:13:50 CDT 2003 ( leakin at dfw dot nostrum dot com )

      * version 0.31: Bill Uhl ( buhl at web-services dot net ) sent a
      patch to add the missing ignore-existing option and make write-batch
      a scalar option (both were oversights on my part).  He also
      suggested quoting the source paths.  Since this could break intended
      actions, There are now 2 options to do this if desired.  The
      quote-src and quote-dst flags will add double-quotes around each
      path in the source or the destination when set.  This should help
      avoid with pathnames containing spaces.  James CE Johnson sent a
      patch to add the no-stdin flag since calling the module under
      mod_perl was having problems.  Bill also reported problems on win32
      and I found strange behaviour myself during some tests, so I have
      backed-out the passing of stdin (from 0.27) entirely.

Tue Jun 10 14:41:54 CDT 2003 Lee Eakin  ( leakin at dfw dot nostrum dot com )

      * version 0.30: Paul Egan sent a patch that does a much better
      job of cleaning up the select loop.  I was still not taking full
      advantage of the features of IO::Select or IO::Handle.  Thanks
      go to Paul for a much simplified loop, and hopefully an end to
      the hangs (at least those due to the perl code).

Mon Jun  9 17:39:42 CDT 2003 Lee Eakin  ( leakin at dfw dot nostrum dot com )

      * version 0.29: Paul Egan had another suggestion I liked very much.
      I switched to using IO::Select rather than calling select directly.
      I've always hated the ugly vec code that must be added in order to use
      select.  It clutters the program with lots of unrelated but essential
      cruft.  The module keeps all that separate from what is going on in the
      rsync wrapper.  Paul was still seeing a hang in certain cases, and
      Ronald J Kimball ( rjk at linguist dot thayer dot dartmouth dot edu )
      was having similar problems.  Both were seeing a hang on the eof tests,
      so I removed them (eof does not play well with sysread, but we need
      sysread to make sure we don't block).  I have not been able to make it
      hang (at least not yet).

Thu May 29 21:33:07 CDT 2003 Lee Eakin  ( leakin at dfw dot nostrum dot com )

      * version 0.28: Paul Egan ( paulegan at mail dot com ) found one of the
      cases where the module could block, and provided a patch to fix it.  The
      cost is a going thru select again after each block is read, but he was
      able to prove it stops the hang.  Paul also suggested an improvement on
      the sysread.  I am now using the block size of the file descriptor
      (defaulting back to 1024 if that fails for some reason).  Many thanks to
      Paul for not only finding a way to reproduce the hang, but also finding
      a fix.

Sat Feb  1 23:03:55 CST 2003 Lee Eakin  ( leakin at dfw dot nostrum dot com )

      * version 0.27: got into too big a rush and forgot to test.
      The STDIN hack caused IPC::Open3 to spew, so I had to change
      it to glob to work properly.  0.26 was never released, but
      now I'm thinking seriously about breaking the link between
      the module version and RCS.  I'll have to do this anyway
      if I move to Subversion for revision control, since it's versions
      are handled in a different manner.

Mon Jan 13 11:25:42 CST 2003 Lee Eakin  ( leakin at dfw dot nostrum dot com )

      * version 0.26: Added link-dest option (new in 2.5.6pre1).
      I previously closed stdin of the forked rsync command since rsync
      didn't really use it.  Now the include-from and exclude-from options
      may be set to '-' and rsync will read the list from stdin.  I've
      attempted to pass STDIN directly to rsync in case someone wants
      to use this option.  In my simple tests the code works, but I have
      not done extensive tests, and do not know if you can pass other data
      such as passwords (not a good idea in terms of security).  This
      may also have other side effects since STDIN could be a tty in some
      instances and not in others (may cause the underlying rsh, ssh, or
      other transport to behave differently during testing versus cron or
      other unattended execution).  If this becomes a problem for anyone
      I'll figure out some way to make it optional.

Tue Jan  7 13:46:56 CST 2003 Lee Eakin  ( leakin at dfw dot nostrum dot com )

      * version 0.25: Tong Zhu ( tongzhu at thereinc dot com ) found an error
      where use of errfun and/or outfun exposed possible use of uninitialized
      variables (those used to break output into individual lines).  Thanks
      for pointing out my error, and the useful 'diff' output making my coding
      error easy to track down and fix.

Wed Dec 11 17:57:25 CST 2002 Lee Eakin  ( leakin at dfw dot nostrum dot com )

      * version 0.24: Heiko Jansen ( heiko_jansen at web dot de ) found an
      error in the exec function where I was accessing the errfun and outfun
      functions from the main object, and ignoring them if they were passed to
      'exec'.  Thanks for pointing out the error, hopefully this will work
      much better for you know.

Wed Dec  4 22:38:14 CST 2002 Lee Eakin  ( leakin at dfw dot nostrum dot com )

      * version 0.23: I got 3 separate reports of hangs, but cannot reproduce
      the error (yet).  Added a trap and more eof tests to hopefully catch
      and exit properly.

Thu Mar 14 23:19:35 CST 2002 Lee Eakin  ( leakin at dfw dot nostrum dot com )

      * version 0.22: added options for rsync 2.5.4, no-blocking-io, no-detach,
      and no-whole-file.

Thu Dec 20 13:52:33 CST 2001 Lee Eakin  ( leakin at dfw dot nostrum dot com )

      * version 0.21: added rsync+ options from 2.4.7pre4/2.5.0, read-batch
      and write-batch.

      Thanks to Jonathan Pelletier ( jonathan dot pelletier at matrox dot com
      ) for catching (and supply a fix for) an output message loss.  This
      version contains a variation on his patch.  Jonathan also suggested the
      new method 'lastcmd' to return the last 'rsync' command passed to the
      system for devugging and error messages.

Fri Sep  7 15:18:58 CDT 2001  Lee Eakin  ( leakin at dfw dot nostrum dot com )

      * version 0.20: made changes to test.pl so failed tests would
      return non-zero exit and actually fail the tests.  Also corrected
      several type-os in the POD documentation (thanks to Joe Smith
      ( Joe dot Smith at wcom dot com ) for pointing them out).

Wed May 16 21:11:31 CDT 2001  Lee Eakin  ( leakin at dfw dot nostrum dot com )

      * version 0.19: discovered the new makefile was installing
      into the wrong path.  It should follow the proper naming convention
      now.

Thu May 10 22:28:28 CDT 2001  Lee Eakin  ( leakin at dfw dot nostrum dot com )

      * version 0.18: Many thanks to Andreas Koenig ( andreas dot koenig at
      dubravka dot kbx dot de ) for setting me straight on the architecture
      specific versus the arch-independent directory usage, and for explaining
      why this wasn't showing up in the by-module listings.  The tar file now
      has a recognizable .pm file in it for proper CPAN distribution.  There
      is also a File::Rsync::Config.pm in the arch-specific dir where the
      build process stores the path to the rsync executable.

Mon May  7 01:07:44 CDT 2001  Lee Eakin  ( leakin at dfw dot nostrum dot com )

      * version 0.17: I decided it is better for the out and err methods
      to return any collected output regardless of whther or not an output
      function has been defined.  This way you can process each line yourself
      and still print the entire output in bulk if desired.  This also
      keeps me from having to work out what these methods should return
      if called when an output function has been defined.

Thu Feb 15 21:25:23 CST 2001  Lee Eakin  ( leakin at dfw dot nostrum dot com )

      * version 0.16: James Mello ( kingjamm at colltech dot com ) submitted a
      patch to add a 'list' function, and educated me on the interesting
      aspect of calling rsync without a destination in order to obtain an 'ls
      -l' or 'ls -lr' (with -r or -a) style output.  Rather than duplicating
      the exec function I made modifications to it and built a wrapper to get
      the list functionality.  This has all kinds of possibilities.  Many
      thanks to James for the great idea and the education on an option I had
      never explored.

      use File::Rsync;$rs=File::Rsync->new(srchost => 'localhost:');
      print $rs->list; will list available modules on the local daemon.
      
Tue Nov 28 10:22:50 CST 2000  Lee Eakin  ( leakin at dfw dot nostrum dot com )

      * version 0.15: After more thought and discussion with Boris we decided
      a second arg of either 'out' or 'err' should be passed to the functions
      so a single user-defined function could be used and it will be able to
      determine the source of the text in arg 1.  Simple functions can ignore
      the second arg.  The examples below would need to change to 'outfun =>
      sub {print shift}' and 'errfun => sub {print shift}' to keep from
      printing the source string. (version 0.14 was not released to CPAN)

Fri Nov 17 23:03:00 CST 2000  Lee Eakin  ( leakin at dfw dot nostrum dot com )

      * version 0.14: Per suggestion from Boris Goldowsky ( boris at range dot
      infoplease dot com ), added 'outfun' and 'errfun' options.  If you
      supply a function (CODE) reference to these the stderr and stdout from
      rsync is passed to these functions one output line at a time and is not
      collected internally for post examination.  Providing a function like
      'outfun => sub {print @_}' will cause each line to be printed as they
      arrive.  A function like 'outfun => sub {print "."}' will simply shows
      progress dots like cpio's -V option.  Thanks to Boris for the
      suggestion.  Sorry it took me so long to get it added.

Sat Sep 16 23:32:59 CDT 2000  Lee Eakin  ( leakin at dfw dot nostrum dot com )

      * version 0.13: added suport for new flags blocking-io and ignore-errors,
      as well as scalars bwlimit, max-delete, and modify-window.  I also
      rearranged the options hashes into a vertical sort similar to 'ls'.
      This makes additions a bit more difficult, but makes reading and
      verifying the options much easier (at least for me).

Tue Feb 29 17:41:14 CST 2000  Lee Eakin  ( leakin at dfw dot nostrum dot com )

      * version 0.12: pervious version fixed blocking, but was not clean
      and caused the tests to fail in odd ways.  This should be better.

      * also added code to treat underscore the same as hyphen in option keys
      similar to Tk.  This allows the use of bareword keys in hashes without
      being mis-interpreted as subraction.

Fri Feb 18 17:45:13 CST 2000  Lee Eakin  ( leakin at dfw dot nostrum dot com )

      * version 0.11: fixed a blocking problem in the exec function
      If one of the output pipes filled up it hung forever (oops).

Mon Feb  7 14:05:37 CST 2000  Lee Eakin  ( leakin at dfw dot nostrum dot com )

      * version 0.10: removed comments about rsync not considering some things
      to be errors.  This appears to be a bug in 2.3.2 on Linux.  It has been
      submitted to the rsync team.  On Solaris rsync properly returns a non-zero
      exit status if the source path does not exist.  The bug was fixed in the
      CVS tree, and in the release 2.4.1.

      * added support for --backup-dir and --existing options (new in 2.4.1)

      * Since I've not heard from anyone about keeping the command-line style
      options, I've dropped them.  The hash-style affords more flexibility
      anyway.  I also removed some duplicate code in parseopts since I can now
      treat all args as a hash.

      * After some discussion on the rsync mailing list I was told about a
      method of passing multiple source arguments for a remote server.  I've
      added a new hash key 'srchost' to be used in combination with the 'source'
      array to make it easier to use this capability.  If the source host is
      remote, and you have multiple source paths, you can put the source
      hostname in 'srchost', and the paths in the 'source' array.  The module
      will assemble them as a single argument to rsync.  If 'srchost' is null
      then the 'source' array is passed to exec as seperate arguments.

Wed Dec  1 21:55:44 CST 1999  Lee Eakin  ( leakin at dfw dot nostrum dot com )

      * version 0.09: removed bareword keys in hashes (5.00404 doesn't like
      my use of 'scalar' as a key)

      * version 0.08: changed test 5 in test.pl to account for different
      return status between Solaris and Linux.  I'll need to look into why
      one detects the error and the other doesn't (rsync itself, not perl).

Tue Nov 30 20:46:00 CST 1999  Lee Eakin  ( leakin at dfw dot nostrum dot com )

      * version 0.07:

      * removed global $pkgname, now uses 'my $pkgname=ref $self' instead.
      I also had to bless $self earlier in 'new' so defopts, and parseopts can
      use 'ref $self' to determine the package name.  It was not too bright
      to hardcode the package name (vacations can have odd effects on the brain)

      * changed saveopts and parseopts to begin with underscore since they
      are for internal use only.

Sun Nov 28 19:41:31 CST 1999  Lee Eakin  ( leakin at dfw dot nostrum dot com )

      * version 0.06:

      * added $pkgname for 'carp' messages to make it clearer where they came
      from.

      * added missing check for 'include' in exec method

      * other misc cleanup and clarification of code and documentation

Sun Nov 28 02:47:34 CST 1999  Lee Eakin  ( leakin at dfw dot nostrum dot com )

      * version 0.05: major re-organization and re-writing of internal
      functions to add the ability to pass a hash reference for perl-style
      options handling.  (Thanks to Greg Ward for the suggestion)  This gives
      increased flexibility including the ability to turn off boolean flags,
      and to build your own include or exclude list as an array and pass a
      reference to the array to the module.  The same can be done for source
      paths.  

      * removed leading double-dashes from the option keys.  It is easier to
      add them in the parser for command-line style than to remove them for
      perl-style hashes.

      * changed 'err' and 'out' methods to return an empty string or an empty
      list if no output was generated.  If there was output they return it as
      an array or a reference to an array as before.  This should make testing
      for the existance of output more efficient.  Before you had to call them
      in a scalar context and dereference the return in order to check.  Now
      you can use a simple truth test.  Also added a note about conditions
      where rsync itself can return a true (0) exit but the programmer may
      consider it an error.  In these situations you need to check stderr for
      error messages (such as 'no such file or directory').  See 'exec' method
      in the manpage.

      * removed newlines from carp messages so the 'at <prog> line ##' is on
      the same line as the rest of the error message (bad habit from die).

      * changed all return values except 'new' to zero instead of undef so
      -w won't complain about use of un-initialized values.  If anyone has
      a good reason to put this back (like good/standard coding practices) let
      me know.

      * added test.pl script for 'make test'.

Sun Nov 21 02:18:56 CST 1999  Lee Eakin  ( leakin at dfw dot nostrum dot com )

      * version 0.04: added missing 'next' that would cause a parsing
      error if -P was used.

      * Renamed module to File::Rsync (Thanks to Tim Bunce for the suggestion)
      It really fits better here than in Net::, but once I thought of it
      as a network-based program I had a mental block when considering other
      groups.

Fri Nov 19 13:25:18 CST 1999  Lee Eakin  ( leakin at dfw dot nostrum dot com )

      * version 0.03: --address should be a scalar, not a flag (the
      rsync man page is unclear on this).

Wed Nov 17 21:20:16 CST 1999  Lee Eakin  ( leakin at dfw dot nostrum dot com )

      * version 0.02:

      * wrote first draft of Makefile.PL and makepm.PL to wrap it all up
      as a real perl module.

      * added new options for rsync 2.3.2 (-P, --delete-after, and
      --address)
      -P is short for --partial and --progress. --address is only
      used in server mode.  I don't know what good these are inside
      this module, but you should still be able to pass them if you
      want.
      --delete-after allows the user to delay any --delete actions
      until after all other files are synced.  The default is to
      perform any deletes before in order help insure sufficient disk
      space is available.  See the rsync (1) man page for more
      details.

      * remove any chance of clobbering $_ (mostly paranoia I hope)

Tue Jun 15 15:27:20 CDT 1999  Lee Eakin  ( leakin at dfw dot nostrum dot com )

      * version 0.01: (initial) based on rsync 2.3.1