0.43  Added Hisanobu Okuda's rlwrapfilter.py python3 module
      and example filters. Filters can now be written in python
      as well as perl.

      If a filter was used, rlwrap would pass all input, output,
      history items, prompts, ... through the filter, even if it
      wouldn't change them. Now, at startup, filters (even filter
      pipelines) tell rlwrap which messages they handle, after which
      rlwrap won't bother them with anything else.

      Added bindable readline command rlwrap-direct-keypress
      that bypasses readline editing and sends its keypress directly
      to the rlwrapped command (like CTRL-G for the Erlang shell)
      
      Added bindable readline command rlwrap-hotkey that passes
      the current input buffer and history to the filter (or
      filter pipeline) specified with the '-z' option. This can
      be used e.g. to insert the current selection at the cursor
      position, or to edit (re-write) the history.

       This uncovered quite a few bugs and inconsistencies:
         - My ncurses' termcap emulation misses some codes (like
           term_cursor_hpos) that its terminfo has. rlwrap now
           always searches terminfo before termcap.
         - rlwrap was confused about the role of history_offset,
           resulting in muddled and unclear (although correct) code.
         - rlwrap --history-size -0 would clobber the history file
           (as per the manual - that has been updated as well)
         - rlwrap's ad hoc way of sending lists of strings to filters
           by interspersing them with TABS or spaces is becoming
           unwieldy, it has been replaced by a standard encoding
           <length1><string1><length2><string2>.... (where the
           <lengths> are fixed length hexadecimal numbers (this is a
           contribution by Hisanobu Okuda)
      
      Playing back a readline macro consisting of more than one line
      would crash with SIGSEGV

      rlwrap with negative --histsize would fail when there is no
      history file yet.

      An empty prompt would make $filter->{cumulative_output} miss
      its last line 

      Pre-given (-P) input would only be put in input line after
      cooking timeout (usually 40 msec)

      One-shot (-o) rlwrap could accept more than one line when input
      in quick succession.

      rlwrap didn't delete the tempfiles used in a multi-line
      edit

      configure.ac now works even when cross-compiling (skipping some
      tests, but providing sensible defaults)

      --enable-pedantic-ansi is a new configure option separate from
      --enable-debug (it used to be implied by --enable-debug)

      --complete-filenames (-c) will now work on OS X and FreeBSD as well

0.42  Added --mirror-arguments (-U) option 

      On SunOS tcgetattr(slave pty) failed with "Invalid argument" 

      If the completion list contained two words, one of which a
      prefix of the other (e.g. "sea" and "seagull") the shorter one would
      be skipped when displaying a list of completions

      reading completion files (with the -f option, or from
      $RLWRAP_HOME/<command>_completions) could fail with an incorrect
      ENOENT ("No such file or directory")

      rlwrap -z listing wouldn't list filters
      
      When both master and slave pty are unfit for sensing slave
      commands terminal settings, rlwrap now bails out with an error
      "cannot determine terminal mode of slave command"


0.41  Slightly late SIGCHLD could cause an I/O error on master pty

      Added -W (-polling) option to poll slave pty for changes in its 
      interrupt character and ISIG flag.

      if $TERM is not found in termcap/terminfo database, use vt100

0.40  configure will now find tgetent() in libtinfo 

      compiling with gcc -O2 made rlwrap hang after CTRL-D on empty line
      
      HP-UX 11 has weird tgetent() return values, confusing rlwrap 

      On Solaris, rlwrap would sometimes fail with "TIOCSWINSZ failed on 
      slave pty"
   
      Single quote ' is now word-breaking by default 

      multi-line edit would mangle upper ASCII and UTF-8 (and still does 
      that to UTF16 :( ) 

      added --extra-char-after-completion and -multi-line-ext options

      rlwrap now recognises the 'rmcup' and 'rmkx' character sequences used by 
      programs that use an alternate screen (like editors and pagers) to return 
      from it.
      
      configure will now correctly determine pty type on SCO OpenServer

      rlwrap --no-children would leak file descriptors to /proc/<command_pid>/wchan
      
      non-ASCII characters in multi-line input are no longer replaced
      by spaces after calling an external editor
  
      running rlwrap within emacs would crash (division by zero). 
      rlwrap now bails out whenever terminal width == 0

      added --enable-proc-mountpoint config option to use alternate linux-like 
      proc filesystems (like in FreeBSD) for finding <command>s working dir (-c option)
      and kernel function in which <command> is sleeping (-N option)

      added prototype for copy_without_ignore_markers, fixing a segfault on NetBSD/amd64
      
      commands final output before dying was lost on FreeBSD

      Filters now get complete echo lines even if the echo comes back in chunks


0.37  Commands that emit "status lines" using backspaces and carriage 
      returns could confuse rlwrap

      removed test program kaboom.c that triggered an internal
      gcc error on armel platforms.

      rlwrap uses C strings internally, and thus cannot cope with
      command output that contains zero bytes (padding). It used to
      replace these with spaces, now the zero bytes are removed.

      if the RLWRAP_HOME is set, but $RLWRAP_HOME doesn't exist, rlwrap
      will create it

      typo: SIGERR instead of SIG_ERR in signals.c
      
0.36  Entering a line from vi command mode would echo the input twice 

      Output from very busy commands would not always be printed on time

      When rlwrap kills itself after a command crash it will not dump 
      core, in order to avoid clobbering command's much more interesting 
      core dump.

      Premature filter death is now reported properly (it used to only 
      say: "EOF reading from filter" or "Broken pipe writing to filter") 

0.35  config.{guess,sub} have been updated to version 2009-12-13
      
      Corrected array bounds error in my_putstr("") (which could make
      rlwrap write an extra newline when exiting, but might even crash
      on some systems)

      Many small improvements and fixes for multi-line input: 

        Multi-line inputs are now written to the inferior command one
        line at a time, so that command's response (e.g. a continuation
        prompt) can be interleaved with the echo'ed (multi-line) input.
      
        Calling an external editor will no longer obliterate the prompt,
        and line/column positions are now correct.

        After a multi-line edit in vi-mode, the cursor will no longer
        end up one line too high.
      
      CTRL-D on an empty line was handed directly to command, but also 
      (erroneously) put in readline's input buffer

      Many small fixes and improvements in signal handling: 
        
         SIGSEGV, and other "error" signals like SIGFPE, are now unblocked
         all of the time, so that rlwrap can always clean up after a crash.

         Since version 0.25 rlrwap's transparency extends to signals: if
         the inferior command segfaults, rlwrap will kill itself with a
         SIGSEGV. In order to get the bug reports where they belong,
         rlwrap now reports explicitly that it has not crashed itself.

         rlwrap's call to sigaction forgot to set the signal mask (!)

         Continuing after CTRL-Z on QNX now wakes up command  

      Added --one-shot (-o) and --only-cook (-O) options

      debug log is now in a format that works well with emacs' grep-mode
    
      rlwrap's bindable readline function names (like rlwrap-call-editor) are
      now in hyphen-style instead of underscore_style (use of the 
      old_style_names will now be flagged as an error)

      Filters can now prevent a prompt from being cooked by "rejecting" it.
      
      Rlwrapfilter.pm would set $_ incorrectly in echo and output handlers.

      RlwrapFilter.pm manpage is now created by newer (and less buggy) 
      version of pod2man

      Added EXAMPLES section and -t option to rlwrap manpage

0.34  Binding wide (e.g. utf-8) chars in .inputrc now works 
      
      prefix arguments are now correctly reset (M-5 a b now 
      yields aaaaab instead of aaaaabbbbb)
      

0.33  rlwrap incorrectly fed terminfo-style capnames ("dl1") instead of
      termcap codes ("dl") into tgetstr(). On newer Debian systems this
      exposed a bug where random garbage would be printed by rlwrap

      Hyphens in rlwrap manpage are now all properly escaped

      RlwrapFilter.pm now only re-sets $filter->cumulative_output when an 
      INPUT message is received


0.32  Major new feature: filtering. Filters sit between rlwrap and the
      wrapped command, re-writing command output, input, prompts,
      history, and completion word lists. 

      System-wide filters live in DATADIR/rlwrap/filters (where DATADIR = 
      /usr/local/share by default, installation-dependent) Because of this, 
      completions now live in DATADIR/rlwrap/completions (until now: 
      DATADIR/rlwrap)  

      To make filter writing easy, a perl module RlwrapFilter.pm has
      been added. It doesn't become part of your perl installation, but lives
      in DATADIR/rlwrap/filters

      rlwrap didn't properly check errno after reading from inferior pty. 
      This could lead to a spurious "read error on master pty"

      Instead of using crusty old signal(), signal handlers are now set
      by sigaction() without SA_RESTART (BSD semantics) Different
      syscall-restarting behaviour among systems caused hard-to-trace
      bugs
     
      Now copies inferior pty's c_oflags to stdout before
      output. (some editors like joe would mess up the screen)
   
      prompt handling logic has been streamlined. Coloured prompt handling is 
      reliable now, even for long prompts.
    
      At program exit, rlwrap now outputs a newline only when the client 
      didn't.       

      Added -g, -I, -N, -S, -w and -z options

      Removed -F option (and added a filter to replace it)

      -p option now takes colour names (-pYellow)

      rlwrap (and readline) uses C strings internally, which could cause problems
      with commands that output '\0' bytes. In direct mode, such
      characters are left untouched, but in readline mode they are
      replaced by spaces.

      the tools directory has been updated, so that configure will 
      feel at home even on newer systems

      tested on SunOS, AIX, FreeBSD, HP/UX, QNX (thanks to polarhome.com), as 
      well as cygwin and linux

0.30  rlwrap can now use putenv() on machines that don't have 
      setenv() (like some Solaris systems)

      EOF on stdin (e.g. by pressing CTRL-D) would end prompt colouring.

      added -q option 

0.29  added -A option to handle (ANSI-)coloured prompts

      added -p option to colourise uncoloured prompts 

      added -t option to set terminal type for client command

      rlwrap now copies its terminal settings from the client
      even when this has put its terminal in single-keypress
      (uncooked) mode. A rlwrapped emacs will now respond to CTRL-C 
      and CTRL-G as it should.

      fixed a long-standing bug where long output lines with the -r option 
      would put mutilated words into the completion list. Drawback: 
      prompts are not put into the completion list anymore (bug or feature?).
           
      rlwrap now handles output to the client before input from the
      client, and only handles keyboard input when all client I/O has
      been handled. This will make rlwrap a little better behaved when
      handling large chunks of (pasted) input, especially with colourised
      prompts
      
      error messages and warnings now include the rlwrap version number 
      
      rlwrap now prints a warning when started in vi mode when the 
      terminal is too dumb to support upwards cursor movement
      
      added a very simple custom malloc/free debugger for use with the 
      --debug option. Rather fragile and not well tested, use with care.

      Whenever TERM is not set, rlwrap assumes vt100. Set TERM=dumb if
      you really have a dumb terminal.

      rlwrap now leaves the handling of multi-line prompts and edit
      buffers to readline (except when in horizontal-scroll mode or when 
      configured with --enable-homegrown-redisplay).  

      whenever --always-readline is set, SIGWINCH not passed to 
      client command immediately, but only *after* accepting a line.

      multi-byte prompts and input no longer confuse rlwrap (provided 
      your readline lib supports multi-byte characters)

      --spy-on-readline now enabled by default
     
      the configure script will now find term.h on cygwin

      dropped the assumption that tgetent() and friends are only ever
      declared in term.h; rlwrap now has proper terminal handling also
      under SunOS.
	
      the rlwrap source should again be fit for consumption by older
      (C91 compliant) compilers
      
      
0.28  fixed rlwrap bug that caused termcap problems on Fedora Core 6 
      (thanks Tung Nguyen)

0.27  when stdin is not a terminal, rlwrap will now exec() the
      specified command instead of complaining

      when stdout or stderr is not a terminal, rlwrap will re-open
      it to /dev/tty (the users terminal) after forking off the
      specified command (so "rlwrap cat > file" will work as a 
      quick and dirty line editor)

      rlwrap now remembers inputs of length 1

      -D option to tell rlwrap how agressively it should weed out 
      duplicate history entries 

      -H option added (history format string)

      Added temporary fix for termcap weirdness on Fedora Core 6

      new -t option for a quick check of terminal capabilities 
      (requires configuration with --enable-debug)

      rlwrap -s 0 <command> will now zap <commands> history

      fixed broken reporting of unrecognised options

0.26  configure on FreeBSD 6.0 still didn't find libutil

      some files were unnecessarily kept open (thanks Stephan Springl)

      on each platform, rlwrap -h now accurately reflects whether
      rlwrap can use long options or optional arguments

0.25  rlwrap would print input twice when in vi-mode
     
      rlwrap under FreeBSD 6.0 now uses openpty() from libutil (config
      script fixed)

      -P option added (one-shot rlwrap with pre-given prompt)

      Until now, whem the underlying command was killed by a signal,
      rlwrap would just exit with exit code 0. Now rlwrap will
      cleanup, reset its signal handlers and then send the same signal
      to itself - so that rlwraps parent (usually a shell) doesn't see 
      the difference with an un-rlwrapped command

0.24  rlwrap -r would mess up commands output (strtok() strikes again!)         
 
      -i option added to make completion case-insensitive    

      -m option added in order to handle multi-line input 
      
      rlwrap now writes to underlying command using non-blocking writes 
      and select(). This prevents deadlocks which could occur with very 
      large inputs.

      corrected some manpage clumsiness 

0.23  Completion word lists are now kept in red-black trees 
      (cf. http://libredblack.sourceforge.net/), dramatically speeding up 
      startup with large completion lists.
     
      rlwrap copies terminal settings from client, so that wrapping 
      programs that manipulate their terminal (like emacs and vim) 
      should be transparent.

      rlwrap -C1 <command_without_arguments> is now accepted
      
0.22  Added key binding to enter a line while keeping it out of the 
      history list (Control+O by default)

0.21  Added --history-filename option.

      Negative history size now means: don't write or truncate history 
      
      Updated helper scripts (like config.sub) in ./tools

      multiple -f options again work correctly

      --enable-homegrown-redisplay configuration option added 
      (kludge to circumvent display problems like sometimes 
      reported on Solaris)      
      
      All unsafe string handling functions (strcpy, sprintf,..) replaced 
      by their safe equivalents - using strlcat and consorts when available.

      --enable-spy-on-readline configuration option to keep display tidy 
      when resizing terminal

0.19  Fixed pty type finding code in configure.ac (newer FreeBSD's 
      were recognised as cygwin)

      Helper scripts moved to separate ./tools directory

0.18  rlwrap could hang when trying to run a non-existent command.

0.17  EOF on stdin would send rlwrap into infinite loop.

      Small bugfixes in testclient, which now works with perl 5.8 and
      cygwin.

0.16  ptys can be found and correctly opend on many more systems
      (thanks to code taken from rxvt).
      
      Makefiles now generated by automake.

      Much beter debugging, -d option now takes optional bitmask
      to report only certain events.

      System-wide completion files now in $datadir/rlwrap 
      (normally /usr/local/share/rlwrap).

      -C and -n options added, -a option can take an argument (to 
      prevent password logging when using this option).

      assert() macro used for run-time consistency checks.

      CTRL-D on empty line now sends EOF to client even on SunOS.
 
      manpage displays correctly even when using troff instead of groff.

      Long options now in --dash-style instead of --underscore_style.

0.15  Fixed a bug where rlwrap would segfault with -f option.

0.14  Fixed a few portability problems.

0.13  Duplicate history entries are avoided at all times, even across 
      invocations.

      Tries to chdir into the slave command's working directory before
      attempting filename completion (only on OS'es where this is found 
      under /proc/<pid>/cwd, like linux and SunOS).

      Now honours 'set horizontal-scroll-mode off' in .inputrc.

      Slave pty is never closed in parent (even after slave's death), 
      preventing long timeouts in OSes with streams-based pty IO.
 
      Lots of small fixes to adapt to gcc 3.x's more finicky behaviour
      (e.g. avoiding multi-line strings, not automaticaly including system 
      includes in Makefile.in). 
  
      configure rewrites manpage to reflect rlwraps capabilities on each 
      platform.

      history searching with CTRL-R (backwards-search-history) now works,
      !-completion is cleaned up (could even segfault in previous versions).

      SIGSEGV is now caught in order to reset terminal.

0.12  When slave pty's ECHO flag is unset (e.g. when entering a password)
      rlwrap now displays asterisks for every input character like this: 
      Password: ******

      Better handling of very long prompts, or very narrow terminal windows.
      If the prompt is wider than the current terminal, rlwrap assumes
      that it has wrapped around and uses the last (wrapped) line of 
      it as the new prompt.

      Slave pty is opened (and remains open) in parent, allowing slave side 
      to be monitored instead of master.

      testclient (a perl script) has been added, uncovering quite a few 
      embarassing bugs.

      system-wide completion word lists (in $sysconfdir/rlwrap) can be used.
      

0.11  If the tcgetattr() call to determine the pty's echo mode fails at
      startup, rlwrap now sleeps for 1 second before trying again
      once. (on FreeBSD, the first call will normally fail, but (most of 
      the time) not the second, due to a race condition that  
      ought to be fixed by some form of synchronisation between parent 
      and child)

      --libdir and --includedir for configure now work, as well as 
      LDFLAGS=xxxx ./configure. 

      Filename completion now works again (when -c option is set).

      User input is now echoed correctly when the pty's echo mode cannot be 
      determined (in that case a warning is printed at startup that
      passwords will be saved in the history file).
      

0.10: logging (-l option) implemented.
	
      history_filename and completion_filename now live on the heap
      instead of in fixed-length buffers.

0.07: Readline mode is entered (by registering callback) with the 
      first keypress of a new line, not before. All command output before  that
      will just be written to stdout unchanged (long lines could become garbled
      in 0.06 and earlier).
      
      Signal handling (esp. SIGTSTP and SIGWINCH) is much improved.

      The -a option forces rlwrap to use readline, which is useful if you
      want to rlwrap a command that already uses readline.
   
      <TAB> is now bound to menu-complete by default: it will cylce through all
      possible completions, instead of listing them all, as in 0.06.

0.06: Transparent mode (immediately handing down keypresses) is now 
      automatic whenever the pty has its ICANON flag unset.
      
      readline  version 4.2 is now mandatory.

      Cleanup of code, eliminating many bugs, possibly introducing
      others (ugh!)

      Application name (used by readline) is now set to command name.

      Duplicate history entries are not remembered.

      Manpage updated: environment variable RLWRAP_HOME is now documented.

0.04: When started in transparent mode, now properly returns to it after 
      each readline edit.

      Senses the pty's ECHO flag. When this is unset, rlwrap doesn't echo
      user input. Neither is it put in the history list (which would contain 
      passwords etc. otherwise).
 
0.03: Now uses .<command>_history and .<command>_completions
      files (by default in $HOME).

0.02: A couple of #ifdefs for portability (tested on BSDI and Digital Unix).

0.01: Initial version.