OS/2 Shareware BBS: 5 Edit/05-Edit.zip/elvis22g.zip/BUGS

Text File | 2002-04-10 | 156KB | 3,368 lines

REPORT BUGS TO [e-mail] (STEVE KIRKENDALL)

NOTE: Each bug in this file is marked with either "*", "/", or "?".
The "*" means the bug is still pending.  The "/" means that I believe it has
been solved.  A "?" means that I've done some work on it, and *hope* it is
solved because it seems to work better but I never really understood what
was wrong in the first place.  A "?" will become a "/" if I don't receive
more reports of that bug.  More than 90% of this file lists *fixed* bugs.

NOTE: Not all of these are truly bugs.  This file also serves as my "to do"
list, so some items are just unimplemented features that sounded like a good
idea at one time or another.


* After :lpr, the "# pages" message goes away too fast.  This only happens
  when printing to a pipe, not a file.

* In "html" display mode, too many operations change the font back to normal.

* Apparently "elvis -Gx11" doesn't always exit when X does -- it can get
  stuck in a loop where it uses all available CPU.

    Xlib normally kills the application when the connection to the server
    goes down unexpectedly.  Elvis will sometimes disable that behavior
    during initialization -- perhaps it doesn't enable it afterward?

* Windows problems:
  * The "idle" color isn't used for the last elvis window when focus shifts
    to some other application.
  * In Options->Font, the "Font style" is ignored -- can we remove it?
  * WinElvis died while resizing, when it tried to refresh the screen.

* Potential for some minor problems with parsing regular expressions, when
  magicchar is set to something weird.  Specifically, some parts of regbuild()
  and gettoken() assume that '[', '^', and '$' are always in magicchar.  If
  they aren't, then odd (though non-fatal) things will happen.

* RFC1345 defines digraphs?  Maybe lib/elvis.dig should be modified to use
  those digraphs instead of the ones that I just made up.

* When editing the source-code for man pages, there is a bug in the handling
  of { and } paragraph commands, which affects the "ap" and "ip" text objects:
  If the cursor is on the '.' of a .PP or .SH command, then the paragraph seems
  to include the PRECEDING text instead of the CURRENT AND FOLLOWING text.

* Compatibility issue: The `m command should move the cursor back to the
  same column, not the same character.  I.e., it shouldn't care whether any
  text has been inserted or deleted on that line before the mark.  To test,
  go onto a long line and type 40|mm20|10x`m  -- this should leave the cursor
  in column 40, but elvis moves to column 30.

* :10,20s/foo/bar/3 should change lines 20 through 22 -- i.e., 3 lines starting
  at the last line in the range.  Elvis ignores the count when given a range,
  which is wrong.

* :g/^/j should join pairs of lines (it does in the real vi and vim), but
  instead it joins all lines together into a single line.

* The S command looses the last indentation character when autoindent is set.
  Also, c_ (which is what S does) works the first time, but . doesn't, if the
  new text was empty.  If the new text was not empty, then . looses indent.

* When pasting rectangles at the ends of lines, elvis doesn't try to pad
  the lines so the rectangle keeps its shape.  Vim does.

* <img src=...> tags don't work except in HTTP: URLs?

* Should elvis.syn's "documentation" line be extended to support Java?  This
  isn't a critical as Perl's POD text since Java's embedded documentation is
  otherwise displayed as a normal comment anyway.

* Some quirks have been reported in the text-mode Win32 version of elvis,
  when running under Windows2000.  The cursor is sometimes left in the wrong
  place when elvis exits, and while running elvis the ex history is displayed
  incorrectly.

* Change the meanings of some commands when a visual selection is pending:
    * s    Like c (rect repeats the change for each line)
    * C    Change whole lines (except rect changes from left edge)
    * S    Change whole lines
    * D    Delete whole lines (except rect deletes from left edge)
    * X    Delete whole lines [:map select X :d^M -- but cut buffers?]
    * Y    Yank whole lines [:map select Y :y^M -- but cut buffers?]
    * I    Insert at left edge (for rect, do it to each line)
    * A    Append at right edge (for rect, do it to each line)

* Some useful 'g' commands that elvis doesn't support yet:
    * gv    reselect previously highlighted text
    * ge    go backward to end of previous word.
    * gE    go backward to end of previous WORD

* Some useful 'z' commands that elvis doesn't support yet:
    * zh    sidescroll leftward.  If cursor forced off right edge, move it.
    * zl    sidescroll rightward.  If cursor forced off left edge, move it.
    * zs    sidescroll to move cursor's column to left edge
    * ze    sidescroll to move cursor's column to right edge

* Maybe add vim's "scrolloff" option.  It causes scrolling to occur when the
  cursor is a few lines away from the top or bottom of the window, so you
  always have a few lines visible around the cursor.

* Refine the way that proportional fonts are used.  If the proportional text
  is wider than fixed-pitch text (which should be rare) then still squeeze it
  for PS and Windows.  But if narrower then don't stretch it.  Text in other
  fonts should be drawn immediately after the end of the proportional text,
  with one exception: The starting point for graphic text should be shifted
  rightward if necessary, to the point it would be at if all preceding text
  had been fixed-pitch.

  This will tie in nicely with the idea that indentation and tabs should be
  drawn in a distinctive font.  If it is drawn as *graphic* spaces, then the
  columns will line up correctly.

* When you mark a rectangle via ^V, and then run an ex command on the
  highlighted text, elvis should IDEALLY limit the changes to the highlighted
  columns.  This would be very difficult though, and it isn't at all clear
  what some ex commands should do -- the ex command set is very line oriented.

* In "normal" display mode (and "syntax" too?), tabs (and any spaces that
  follow a tab?) should be drawn in a distinctive font.  This is necessary
  so that printing with ":color normal proportional" will still allow left
  margins to be aligned.

  Similarly, in markup display modes, leading whitespace shouldn't be italic.
  If italics are converted to underlined, then that'll cause the whitespace
  to look like underscore characters.

* Maybe add an option that makes the <Tab> key use shiftwidth instead of
  tabstop.  Or maybe allow inputtab=shift or something like that.  Vim
  uses softtabstop as an alternate width, for both the <Tab> key and <BS>.

* It would be nice if the "showtags" option was clever enough to check the
  tags file in some other directory, when editing a file in that directory.

* The "untar.c" program needs to convert ':' into something harmless on
  Windows systems.  Windows doesn't allow ':' in a filename, except in a
  drive letter.  Maybe use '_' instead?

* When reading an HTTP response that has a header, elvis should skip the
  empty line that marks the end of the header.  It would also be nice if
  the Content-Type line was stored for later use.

* The :fold command seems to confuse the :g command, perhaps because :fold
  changes the meaning of "line".  The :unfold command is okay, though.  In
  particular, this means you can't do this...

    :g/^{/,/^}/fold (current("tag"))

  ... but this works...

    :g/^{/,/^}/unfold (current("tag"))
    :%fold!

* Need to update http://www.cs.pdx.edu/~kirkenda

* When lpcolumns!=columns, printing will often use the wrong font attributes
  when printing text in a markup display mode.  Apparently the array of
  line attributes is tripping it up.  NEWSFLASH: It can cause core dumps too.
  Try setting lpcolumns=300, for example.

    This should go away with the new markup code (when written).
    Until then, just be sure to ":let lpcolumns=columns" before printing.

* hlsearch doesn't mark all matches within a markup display mode.  Its
  behavior is very consistent, in that it fails to highlight the same
  instances each time.

    This would be hard to fix -- probably not worth the effort at this
    time.  The ideal solution would involve modifying regexec() to search
    across line boundaries.  The problem is caused by the fact that
    markup display modes have a different definition of "end of line"
    than the regexp matcher, so any text between a newline character
    and the end of the "markup" line is unchecked.

* Giving too many "Hit <Enter> to continue" prompts.

* "spellsuffix" doesn't always work.  It seems to skip looking for base words
  defined via :words -- it only checks the "spelldict" dictionary.  Example:
  "GUI" is added via :words, and spellsuffix contains "'s", but "GUI's"
  is still flagged as an error.

    I believe there are two things happening here.  First, it doesn't
    look for the root word in the tags dictionary; this allows lowercase
    "digraphs" to be flagged as an error even though "digraph" is accepted
    because it is a tag.  This is not a bug.

    The second thing is that it looks up the root word in a case-
    insensitive way (converting it to lowercase) so it doesn't find
    the uppercase "GUI" when trying to check "GUI's".  This is a bug...
    how to fix?

? From Stephane Billiart:
  There is a bug in the "configure" script -- it doesn't know that the
  internet files are required even if the network protocols are not
  used, with the "x11" interface.  Here's a patch that fixes it for Solaris...

    529,531c529
    <       why "   to ensure that 'const' is ignored.  Also, if X11 is to be supported then"
    <       why "   we also need -lsocket -lnsl.  A -R flag may also be added so the X11"
    <       why "   shared libraries can be found at run time."
    ---
    >       why "   to ensure that 'const' is ignored."
    535a534,535
    >               why "For Solaris2, X11 requires -lsocket -lnsl.  A -R flag may also"
    >               why "   be added so the X11 shared libraries can be found at run time."
    538a539,542
    >       elif [ X"$inet" != X ]
    >       then
    >               why "For Solaris2, any network access needs -lsocket -lnsl"
    >               TLIBS="$TLIBS -lsocket -lnsl"

  Okay, here's my question: How could the network files be present but not
  used?  The "configure" script always uses HTTP and FTP if the network files
  are present, so that should always cause those files to be included under
  Solaris, right?  Or does Solaris sometimes have the libraries but not the
  header files?

? The :au command doesn't accept {...} notation for multi-line commands.

    This is probably more trouble than it's worth.  Because the syntax
    for :au is so different from every other ex command, ex_autocmd()
    must parse the args itself.  To get the normal ex command parser to
    hand off unparsed arguments to ex_autocmd(), the cmdnames[] array
    just says its entire argument string is a single command line.
    So unless the whole command line starts with a '{' at the end of a
    line, elvis will treat it as a one-line command.  Period.  Fixing
    this would require major changes to the parser in ex.c, purely for
    the benefit of this single, rarely-used command, for an elvis-only
    extension.

? Non-HTML URLs are loosing their FORM parameters.  This is a problem,
  because user-defined protocols such as mailto: need parameters.

    Hmm... I remember seeing this problem, but it seems to working now.
    Either I fixed it and forgot to cross of this item, or I had added
    it twice and only crossed off one copy.  Or I guess I could have
    fixed accidentally while making some other change, I suppose.

? When writing via FTP, can't overwrite an existing file, even with ":w!".

    I was unable to reproduce this; it worked fine for me.  Perhaps the
    person who submitted this bug was leaving off the "~user" part of
    the URL?

? When running a program via :!cmd, the last line isn't printed unless it
  ends with a linefeed.

    I tried this, and it works okay for me.  Running "perl -e 'print 5'"
    does indeed show "5" -- with "Hit <Enter> to continue" appended,
    as one would expect.  I suspect this is a buffering problem in the
    cmd, not in elvis.

? Can't interrupt a looped :s command.  Not sure why.  Apparently guipoll()
  doesn't work correctly, at least with the "x11" interface.

    Oh, now it seems to be working again.  Damn.

? Ex scripts sometimes dump core.  Here's an example from Kris Wehner:
      ex - /usr/local/apache/conf/httpd.conf << EOF
      /^# LoadModule foo_module/a
      LoadModule ct_auth_module libexec/libct_apache_plugin.so
      AddModule ct_apache_mod.c
      .
      wq!
      EOF

    I tried this on my system with 2.2d-alpha and it worked perfectly.
    I can't think of anything I changed that could explain why it works
    now, though, so I'm kind of worried.  Perhaps I should try it on
    a Solaris machine...

? The :foldc alias has been reported to cause core dumps on some files --
  specifically, the gdb/infrun.c file from gdb-5.0's source.

    I downloaded the source for GDB 5.0, but was unable to reproduce this.
    I suspect the core dump was caused by a bug in the :normal command,
    which the :foldc alias uses.  That bug is fixed in 2.2d.

-------------------------------------------------------------------------------
Fixed in 2.2g-beta:

/ The spell-checker didn't like words that start with '_', even though
  tag names often do.

/ Maybe allow lpheader font to be controlled via ":color header" ?  Still
  make it default to "fixed underlined", but some users might prefer boxed.

    Actually, it was already doing that for non-markup modes.  I tweaked
    the markup modes to do the same, and documented it.

/ Gnome patches from Klaus Elsbernd, needed for compiling the gnome interface
  with a non-GCC compiler, and run in SunOS

/ ctags *needs* support for "enum".  And probably "class" wouldn't be too hard.

/ ctags needs to reset the class hint when switching from one file to another.

/ The show=link option is slow if there are no <a ...> tags shortly before
  the cursor position.  It should probably stop looking if it sees a tag
  which is a paragraph/section marker.

    I thought I had this one fixed, but apparently the fix wasn't good
    enough.

/ ":map" doesn't list cursor key maps, and even some user maps such as "gw".
  Actually, I kind of like the fact that it omits cursor keys since the user
  probably already knows what they're mapped to, and it's nice to avoid clutter.
  But there should be *SOME* way to see them.

      Fixed.  Also made ":map" list non-cursor keys, while ":map all" lists
      all maps including cursor keys.

/ Use mapmode to make <Enter> follow a link in HTML.  Maybe also use <space>
  to page forward.

/ "map mode=html ^V  ^F" isn't saved correctly.

/ Don't set hlspell for the help files.  Why embarrass myself?

/ Elvis dumps core in a C++ file (or C header file) that starts with a comment
  if the "showlink" option is set.

/ "show=link" shouldn't affect any display modes other than "html".

/ The :r command displays the number of lines read, even if report=0

/ In "hex" mode, the line before the column headings is in a funny font.
  This only shows up when you do a ":sp file.bin" on a binary file; as soon
  as any event causes the screen to be redrawn, it looks okay.

/ The "modified, not saved" message uses filename, should show bufname.

/ There's a bug which causes a core dump when "theme:parchment" is loaded.

    Apparently this is in the "1;/^BEGIN /,$d" command, which deletes
    everything but the initial comment.  I suspect that it is failing
    because the last block is deleted, in a buffer which requires
    multiple blklist blocks.  (The "parchment" theme is the biggest
    one.)

    If I run that command manually it doesn't dump core...  until the
    third time!  This suggests that buffer cloning may be involved.

    SOLVED!  It took six hours, but I finally found bug in lowbuf.c's
    delblock() function, when a buffer is big enough to require a chain
    of BLKLIST blocks.  To delete a CHARS block in the n'th BLKLIST block,
    it was copying the first item from the n+1'th BLKLIST block and then
    recursively calling delblock() to shift the n+1'th block (and any later
    ones, the same way).  But delblock() frees the deleted CHARS block,
    so I have INCREMENT THE ALLOCATION COUNT FOR THAT BLOCK BEFORE THE
    RECURSIVE CALL!

/ For the termcap interface, the "italic" attribute can interfere with
  colors.  It is implemented as "half brightness", which Linux's console,
  for one, can't do in color.

    Added a "ttyitalic" option, similar to "ttyunderline".

/ Changing the 'background' option should cause colors to be reevaluated,
  if the actual background color is unknown.  This would mostly benefit
  the "termcap" interface, since the GUI interfaces always know exactly
  which background color they're using.

    Basically, this action could be done by calling recolor() for any
    font which has no "like" clause.  (The "like" fonts will automatically
    be reevaluated when the font that they're like is reevaluated.)
    Triggering this action will require writing a custom option-setting
    function.

/ The MS-DOS version has a hard time using bright colors for "normal" plus
  other things: color normal yellow|color comment bright green.

/ Modelines don't work?

    This was because the first "1,5s/ex:\(.*\):/\1/x" was failing, which
    prevented the elvis.arf script from continuing with the second
    "$-4,$s/ex:\(.*\):/\1/x" command.

/ "safer" mode is too strict.  Disallowing :alias and :autocmd causes problems
  in the initialization scripts.

  / Make elvis.ini check "safer" before running elvis.ali
  / Make :augroup be unsafe, so the script generated by :mkexrc will work
    whether safer is on or not.

/ Weird problem with PERL doc lines.  See Merijn's example.

    This was caused by a scanalloc() call without a matching scanfree(),
    in dmsyntax.c.  Each screen update left one more scan context lying
    around, and that caused elvis' buffer functions to get confused --
    there should never be any scan contexts active when the buffer is
    changed.  There are assert() statements to verify this, but those
    are normally disabled, so that's why this looked subtle.

/ Shouldn't draw guide lines on ex scrolled rows at bottom of window.

    Fixed, in an interesting way.  I added a new text face, "bottom",
    and made it immune to guide lines.  All ex commands and their output
    are drawn in this font, so they don't get guide lines.

/ Make "smartargs" be a buffer option?  Sometimes there will never be tags
  for functions; e.g., in HTML source where "functions" are really tag
  parameters.  Searching for a nonexistent tag can take a while!

    Done.  Also modified "elvis.arf" to turn off smartargs for html.

/ If you delete the last character from "boxed" text, the right edge of the
  box is not redrawn in its new location.  The box appears to be open-ended.

/ Add a BgChanged event for :autocmd.

/ add a "showface" option, which shows the name of the current font.  Maybe
  also add a current("face") function which returns the name of the face at
  the current cursor position.  If the face is a combination, then return "".

    Added...
        current("face")
        current("background")
        color("name")
        color("name","fg")
        color("name","bg")
        color("name","like")

    Replaced many showXXX options with a single versatile "show" option.
    file, tag, cmd, face, link, spell, region

/ The 'x' flag of the :s command should move the cursor to the current line
  before evaluating the command.  This affects the readTHEME alias.

    This is very quirky.  The bug mostly resides in parsewindowid(), but
    really it is just a manifestation of the fact that my ex parser doesn't
    allow ex commands to specify a cursor position -- it tries to use the
    window's cursor, or the top of bufdefault.  I think in readTHEME, it
    is using the latter.

    Fixed!  I just made sure that x's quirks matched parsewindowid()'s.

/ Some easy ideas for background images in X11:
  / Allow background color names to contain either a simple color name, or
    the name of an image file, or both.  When both are given, use the simple
    color to "tint" the image.  Return a pseudo-colorcode.  Also return the
    average RGB value of the image, so contrasting foregrounds can be chosen.
  / When an unscrollable image is being used, GUI->scroll() should always fail.

/ problems with partially specifying the "normal" color, and probably
  any of the special colors -- When you run :mkexrc it only remembers the
  most recent partial color specification.

/ The "or color" clause can be chosen before the background is set. E.g.,
  ":color normal black or white on black" chooses the new foreground based
  on the old background.

    Note that this is related to the "partial specify" problem -- the
    only reason for specifying "or" colors with the background is so
    that you'll be prepared when the next foregroundless :color command.

    Possible combined solution: Write a parser which returns the foreground
    and background colors AS STRINGS.  When doing this, treat the "or"
    word as plain text, so it is included in the foreground string for
    later processing.  Other attributes are returned as flags.  When
    doing a partial specification, parse both the old & new attribute
    strings; if the new one omitted a foreground or background, then
    use the foreground or background from the old string.  Process the
    background first, so the proper "or" color can be chosen.  Regenerate
    the combined text by using the foreground & background strings,
    with flags generated from the attribute bits.

/ In "x11", a bold 'm' leaves bits to the left it its character cell --
  maybe clip the cell?  It would make scrolling look better.  However,
  italic characters often slop to the right, which apparently produces far
  fewer problems, so the clipping should only be on the left.

/ Make WinElvis read its color names from a file, preferably in the same
  format as /usr/X11R6/lib/X11/rgb.txt

/ ":help ex" fails.  It tries to load "elvisex.html" in the current directory.

/ Describe the meaning of ";" in ex addresses.

/ The postscript print driver uses the video colors, not the carefully chosen
  lp colors.

/ It is so tempting to give only one foreground color, I really should add an
  "lpcontrast" option.

/ For the sake of uniformity, x11's "normalfont" option should be renamed to
  just plain "font", like in Windows.

/ ":help set ttyunderline" doesn't go to the ttyunderline message, unless
  you're using the "termcap" interface.  I wouldn't expect name completion
  to work, but you should at least be able to give the full name and find it.

/ "nottyunderline" shouldn't completely disable the use of underline --
  still try to use it if no background color has been set.  This will allow
  the window separator to be red.

    Actually, this is more of a personal preference.  But users who
    prefer to have no redness in their window borders could just say
    "color normal on black" so no harm is done.

/ Can't use "bold" attribute to distinguish between normal/idle.  Elvis
  appears to only be sensitive to the colors, not other attributes.

/ Windows problems:
  / The Options->Gui dialog can't change colors, and doesn't display the
    colors of that have been set elsewhere.
  / Text type names aren't sorted, and the label should probably be "face"
    instead of "text type" anyway.
  / Combo boxes should show more items.
  / The Options->Syntax menu item crashes elvis.  It is obsolete anyway,
    and should be deleted.
/ The :let command never triggers an OptChanged or OptSet autocmd event.

/ Add "kvt" to the list of terminals recognized by -Gx11.  Also, the termcap
  interface should assume "kvt" has a light background.

/ "make install" should create the /usr/local/lib/elvis/tags file.

/ The smartargs option doesn't allow a space before '('.

/ Make the toolbar's colors be configurable via resources (instead of the
  scrollbar), and then make the default settings of the scrollbar and
  statusbar be "like toolbar".

/ In the Windows, OS/2, and MS-DOS binary distributions, make the contents
  of the "lib" directory be read-only.  This is important for making
  ":set prefersyntax=writable" work right.

/ Writing an anonymous file should never change the name of an INTERNAL buffer.
  Currently, :mkexrc seems to do that.

/ The "elvis untitled #n" buffer names should correspond to bufid numbers.
  It's less confusing that way.

/ In "termcap" interface, sometimes the hlsearch highlighting doesn't show
  everything that it should.  This seems to depend on color settings.  See
  samples/hlsearch.cpp for an example.

    Not a bug.  This is due to the interaction of "bold white" text and
    "bright red" highlighting.  In the termcap interface, bold+bright
    always results in bright white, regardless of whatever color you're
    trying to use.

/ For HP, the default level of optimization doesn't work in elvis.  Need to
  make "configure" drop it to +O2.

    I tried changing the "configure" script to handle this, but I can't
    test it myself.

    Merijn Brand reports that it does indeed work.

/ Sometimes the cursor gets stuck at the bottom of the screen (not the status
  line where ex commands are entered, but the last one normally used in visual
  mode).  Hitting zz fixes this condition.

    This may be related to the "calling delimage() endlessly" bug, which
    is now fixed.  The "stuck at bottom" bug hasn't occurred since that
    fix.

-------------------------------------------------------------------------------
Fixed in 2.2f-beta:

/ Changing the "normal" background should cause all colors to be reevaluated,
  so that "or color" can be chosen properly.  Merely switching a window from
  "idle" to "normal" shouldn't do this -- only a ":color normal" command should.

/ Hey!  ":color normal gray" changes the background to black, in -Gx11?
  It should leave the color unchanged.

/ Need a way to make maps that only apply in certain display modes, or
  possibly some other factor.  Maybe add a "map=XXXX" flag, and a "map"
  buffer option?  Or a generic "option=value" flag?  Maybe add a :mapgroup
  command, like autocmd, and allow groups to be selectively enabled/disabled.

    I implemented this as a "mode=XXXX" flag, and a "mapmode" option.

/ After running an external command, the termcap interface restores colors
  BEFORE switching back to the edit screen, so the original screen colors
  are clobbered.

/ Make :alias list the user aliases, and :alias! list the system aliases.

/ Make :alias indent aliases to reflect the control flow.

/ Need a way for aliases to access the tail of a command.  The csh way to
  do this would be to put a * after the digit, as in !:2* to access everything
  except the first word.

/ Can the Edit autocmd event be partially dependent on bufdo()?  Should it be?
  This is desirable because it would reduce the number of events, but what if
  somebody "moves" the cursor while in input mode?

    Actually, moving while in input mode should probably call bufdo()
    anyway, so that individual changes can be undone.

    A more complex problem: global changes, via either :g or :%s.  There
    should probably be flag added to each buffer, to indicate that Edit
    events should occur more often than "undo" saves.

/ Make the window title show the filename if there is one; else the buffer name.

/ Add a way to mark regions.

/ With "optimize" turned off, the : command causes hlobject text to be
  highlighted weirdly.  I think this happens only when there are multiple
  windows.  It tries to recolor the current window, using the temporary
  fonts from the other window.

  Exposure events can be similarly weird, regardless of optimization.

    This might be solvable by comparing the color number against the
    npermanent.  If >= then allocate a new temporary color with the
    required attributes... BUT WHY SHOULD THAT BE NECESSARY?

/ Make :mkexrc generate a script that uses :try instead of :if.  This is
  important if FEATURE_CALC is turned off.

/ Allow more features to be disabled via FEATURE_XXXX -- tags, regexp, etc.

/ Modify the [[ and ]] commands to recognize functions that start with '{'
  at the end of the function header line, instead of in column 1 of the
  following line.  Assume function header lines are either unindented.
  This notation is standard in GNU source code, Perl, and (oddly enough)
  in elvis' own alias notation.  This should also affect the gd command.

    Done.  Also added "tweaksection" option to control it.

/ The "locked" and "readonly" option should never be saved in ~/.elvisrc
  Maybe other options too.

/ Adding "tag=" to the status line had the unintended side-effect of adding
  it to the result of current("tag") too.  And it looks uglier than I expected
  anyway.  Since the real source of confusion is when both "showtag" and
  "showspell" are set, it would probably be better to add numbers before
  the showspell words.  This makes them distinctive, and also makes the gs
  command easier to use.

/ Add link anchors to example aliases in lib/elvistip.html.  Refer to them
  in other parts of the manual.

/ Bug in suffixes -- it searches random memory past the end of the suffix
  list.

/ Something is wrong with subscripts.  Try :calc "foo"[,2...];"ps".

    There were two bugs: The precedence of [] was too low, and non-numeric
    arguments were being interpreted as an infinite list of 0's.

/ The "x11" user interface dumps core.  Try "e2 exmake.c", ":sp ex.c", resize
  the ex.c window to be 4 lines high, run ":window" or ":set".  This does not
  dump core in the "termcap" interface.  It may be related to spell checking.

    I suspect this has something to do with static variables being set
    with the expectation that one window will be used next, and then the
    other one is winds up being used instead.  I say this because in at
    least one occasion, the status line (below the text window) showed
    the same tag name for both windows even though they were for different
    buffers.

    Unfortunately, one side-effect of this is typically that the stack
    is clobbered, so it is difficult to tell what's going wrong.

    FIXED!  It turns out that some important variables were being left
    unset by drawopenedit().  The variables are normally set by the
    drawupdate() function which calls updateimage().  But drawopenedit()
    also calls updateimage(), so it had to set those variables too.

/ It would be nice if ":make" didn't switch buffers if there's already a
  buffer showing the erroneous file.

/ There should be a way to disable spelling for each buffer.  This is partially
  dependent upon whether the window is readonly or not, but it should be more
  versatile than that.  Maybe replace the spelltags option.

    I simply made "hlspell" be a buffer option instead of a global one.
    I also deleted the "spelltags" option since I no longer consider it
    to be useful... or not useful enough to justify making users read
    about it and understand it.

/ Sometimes elvis gets stuck calling delimage() endlessly.  In the "termcap"
  interface, if you start on an HTML file and then use :sp to load some other
  file, then elvis gets stuck in an endless loop trying to update the HTML
  window.

    Fixed!  This was happening because eventfocus() was calling draw()
    for the original window, which in turn tried to draw it using the
    "html" functions -- even though the cursor was still in the ex history
    buffer at the time.

/ Make the absolute() function be smart about ".." in path names.

/ Elvis should recognize different paths to the same file.  This is important
  because some compilers mess up names (javac, for one).

    The best solution here is probably related to the idea of giving each
    window its own virtual current working directory.  But that's more
    than I was planning to do for 2.2.  Instead, I can just make the bufname
    option store the full pathname; then buffind() will find the existing
    version of a buffer no matter how you express its filename.

/ Add g<Tab> -- move backward to previous link.  Make GUIs map <Shift-Tab>
  to g<Tab> if possible.

/ Include /usr/dict/words in the windows binary distribution.

/ Perl scripts aren't being recognized if they have no extension?

    Merijn Brand tells me this is due to the fact that he sets the
    magicchar and magicperl options to make elvis use Perl-like regexps,
    but the default elvis.arf script file tries to recognize Perl scripts
    by using traditional vi regexps.

    There are two ways to solve this.  One is by resetting the magicchar
    and magicperl options to their defaults at the start of each script
    or alias -- either hardcode the equivalent of ":lo mac=^$.[* nomap"
    or add a new OptLocal autocmd event to make it configurable.

    The other way is to add a special metacharacter which forces the
    remainder of the regexp to be parsed using traditional syntax, and
    then add that metacharacter to the start of the regexps in elvis.arf.
    I'm thinking \V would be nice.  Maybe also support \Q...\E quoting?

/ When pasting chars into an empty buffer, add a newline first.  This is
  particularly important when pasting via the GUI, but it may also be
  important for pasting between buffers.

/ When searching for tags: if there are both a macro and a function with the
  same tag name, use the function first.  This should probably be both an
  optional flag in ctags, and the sorting behavior or :tag and the ref program.

    What if I just initialize the heuristic tables with a higher
    preference for functions initially?  I'd need to do this in both
    "ref" and "elvis".

/ Add "color emphasized like bold" somewhere, for the benefit of syntax
  coloring.

    Fixed a different way: Since syntax coloring fonts should be given
    names that reflect their intended purpose, I modified elvis.syn to
    use other font names, and then added "color" lines to define the
    default appearance of those fonts.

/ Termcap "transparent" background doesn't work with Eterm.  If some text
  uses a different background, then the display code doesn't switch back
  to normal afterward.

-------------------------------------------------------------------------------
Fixed in 2.2d-alpha:

/ The "howto.html" file doesn't look good in Netscape.  This is due to the
  ugliness of the non-elvis <dt><dd> list.

/ The "showspell" option shouldn't suggest natural words where only tags are
  allowed.  For example, when editing source code, it shouldn't suggest natural
  words when the cursor is on a variable.

/ The newbuffer() function and :file command can both create different buffers
  with the same bufname.  So can writing an anonymous buffer out to a file.

/ Change the "showname" and "showtag" options' display formats to prefix
  their value with "file=" or "tag=" -- there are so many different things
  that can be shown down there that it can get confusing.  Oh, and maybe
  rename "showname" to "showfile".

/ Make :mkexrc save aliases.

? Occasionally, elvis will mess up the session file.  When this happens, the
  screen looks normal but the cursor doesn't move correctly.  I think this is
  a recent bug -- added after 2.2d.

    Maybe this has something to do with the message queue?  Elvis doesn't
    allow buffers to be modified while a scan operation is pending.  The
    (Elvis message queue) buffer is used for queuing messages, so it is
    modified each time a message is generated.

  FIXED? There was a bug in the indentation code, used by the shift commands
    and autoindent, among other things.  It was mixing scan operations
    with changes.  Also, wordatcursor(), used by the spell checker among
    other things, was sometimes neglecting to free a scan context when
    it was done, which eventually caused large numbers of scan contexts
    to accumulate.

/ Need a way to fire off commands when certain events happen.  Maybe vim's
  ":au" command, or something related to the :gui command which already runs
  events for GUI events.  Combine them somehow?  The immediate goal is to
  provide a way for a [buffer] menu to change when buffers are created or
  deleted.

/ In XFree86 4.0, the "x11" interface's toolbar hints have an extra character
  on the end -- maybe not trimming newline?

/ Maybe add "transparent" as a termcap background color -- it will just
  reset the colors via \E[m and then not set a background.  This will allow
  you to specify some background colors while still leaving others set to
  the default.  It won't always look good, but it is another option.

/ Under XFree86 4.0.3, the image under the cursor doesn't appear to be saved
  before the cursor is drawn.  On the next blink, it get screwed up sometimes.
  This also affects selections.

/ If multiple tags are found with equal sorting values, then sort them by
  the timestamp of the file where the tag is defined.  Do this in ref, in
  addition to the built-in tags searcher.  Use dirtime() to find the
  timestamp... but will that cause problems in ref?

/ osunix/osblock.c shouldn't trust the access() function, because broken
  links look like nonexistent files to it.  It should instead try calling
  open(..., O_RDWR|O_EXCL).  Also, think about what happens when blkopen()
  fails, because currently it leaves the tty in "raw" mode.

/ Windows patch from Mauro Persano: Use GetModuleFileName() to get the full
  pathname of the winelvis.exe file, if possible.  This should be more
  reliable than explicitly searching the path (though it will still search
  the path if GetModuleFileName() fails).

/ Spelling bug: If a word is known to be bad or incomplete, then any words
  that can be derived from it via suffixes will be considered good.  I.e.,
  the suffix test only checks whether the base word is known, not whether it
  is good.

/ The showspell option doesn't update the status line correctly when running
  in the termcap user interface.  It isn't updated as often as it should be.
  Specifically, it is only changed from empty to suggestions.  (Scrolling
  causes it to be empty, briefly.)

    This is exactly the way other status messages work, but other status
    messages aren't sensitive to the cursor position.  The genlastrow()
    function will need to be extended to clear the last row before the
    next update, if it is displaying alternative spellings.

/ The "gs" command complains about words that start with digits.

/ Spell checker bombs on words longer than 100 characters.  Any word that
  long should be assumed to be misspelled, and not have any suggested
  alternatives.

/ With incsearch set, elvis could dump code on some partially-entered
  regular expressions, such as "/[^[".

/ In expressions, allow $_ to be used as a synonym for line().  Allow $ before
  options, too.  This would make Perl users happy.

/ Can't yank from a locked buffer?

/ Change the description of makeprg and ccprg, to include a description of
  how to add double-quotes around $2.

/ Odd things happen when you edit a filename that looks like a file number.
  E.g., ":e 1".  It can clobber an existing buffer.

/ d/regexp/+0 retains the line containing regexp, but should delete it.

/ Spelling gives false errors when sidescrolling is enabled -- partial words
  along the left edge of the screen are flagged as being incorrect.

/ Speller should skip over words that begin with a digit.

/ Could incremental searches be smart enough to skip searching if more
  keys are in the type-ahead buffer?  That might avoid some useless search
  time.

? How to dump core: 1) start editing an existing file, 2) run ":make",
     3) run ":("Eel)sp", 4) run ":!ref ...".

? The :configure alias doesn't work when there are spaces in the
  directory names.

/ Create a script that map's vim's commands to the corresponding elvis commands
  including the following:
    rX    changes each non-newline to X [:map select r g=]
    x    Delete (like d) [:map select x d]
    U    make uppercase [:map select U gU]
    u    make lowercase [:map select u gu]
    ~    toggle case [:map select ~ g~]
    J    Join whole lines (like :join) [:map select :j^M]
    ^]    Tag search, for highlighted text [:map select ^] y:ta ^P^M]
    R    Change whole lines (same as S)
    o    Go to other end
    O    go to other edge of rectangle

/ Allow v/V/^V to change the type of a pending selection.

/ When a rectangular selection is pending, $ should cause it to extend all
  the way to the end of each line.

/ New motions for v/V/^V selections:
    g%    go to other end of selection  (same as vim's o)
    g^V    go to other vertical edge of rectangular selection (vim's O)

/ :>> doesn't shift twice, but should.

/ The :join command doesn't work correctly when multiple lines are visually
  selected -- it still joins just the current line & following line.

/ Wasn't there a window() function?  There should be!

/ Describe why "-Gx11" is needed when adding elvis to an X start menu in Linux.

/ ":help ^Wd" doesn't work.  Apparently multi-key vi commands aren't
  recognized by the viname() function.

/ Speller dumps core if a spell_t node has a max=255.  That's a valid Latin-1
  letter, but it messes up some loops that use a CHAR as a counter, since
  the counter can never be incremented past 255.

/ Support keyboard timeouts (the usertime option) in the "x11" user interface.

/ If the "find next misspelled word" command is used with a count, and the
  cursor is on a misspelled word, then use the count'th corrected spelling.
  For example, ^K moves to a misspelled word, and 1^K uses the first
  suggested correction.

/ Need to support 'g' commands, so we can standardize keystrokes for spelling.

    / g0    go to leftmost char in current row (for wrapped lines)
    / g^    go to first non-whitespace char in current row (for wrapped)
    / g$    go to rightmost char in current row (for wrapped lines)
    / gj    move down to same column of next row (next line, or next row
        of a long wrapped line)
    / gk    move up to same column of previous row.
    / gd    go to local definition of word that the cursor's on (first
        mention within the current function).
    / gD    go to global definition of word that the cursor's on (first
        mention in the current file).
    / gI    like "0i"
    / gJ    join lines without space, like ":join!"
    / gU    operator, makes text uppercase
    / gu    operator, makes text lowercase
    / g~    operator, makes text lowercase
    / ga    show ascii code of current char in ascii, decimal, hex, octal
  and...
    / gh    move left, skipping hidden characters (for markup modes)
    / gl    move right, skipping hidden characters (for markup modes)
      / g=X    operator, replaces text characters with X
      / gs    go to next misspelled word.  If given a count, then before
          moving, replace the current word with the count'th alternative
          spelling.
    / gS    go to end of current spelling word. (Useful for cgS, if none
          of the alternative spellings are correct.)

  The first task is to create a means by which these commands can be parsed.
  Probably this should be a second table, similar to the vikeys[] array in
  vi.c.  An extra BOOLEAN flag in the vi input state will indicate whether
  the main table should be used, or the g table.

  Then I can start implementing the actual G commands.  Some will be easier
  than others, but none should be terribly hard... except gv.

  Tests:
  / Make sure invalid g commands are rejected safely.
  / Make sure valid ones are recognized.
  / Make sure ":help gs" works.
  / For operators, make sure "g~g~" and "g~~" both change the whole line.
  / Verify that g=x works for motions, text objects, and selections.
  / Verify that other operators work for motions, text objects, and selections.
  / Make sure changes can be undone.
  / Make sure . can repeat all of the commands that change text.

/ Elvis hangs on :%s/\( ,\)\+//g

    This turned out to be due to a recently introduced bug in non-greedy
    closure operators: They weren't checking the minimum number of
    repetitions.

/ When incsearch is set, ":map! >^[?<^Mye" dumps core.

    For backward incremental searches, the regexp's leavep field wasn't
    being saved.

/ Elvis hangs when hlsearch is set, for any regular expression which could
  match a zero-length string.

/ WinElvis doesn't update the screen correctly if it scrolls while part of
  the window is off-screen.

/ Allow :map to support more flags, in addition to "visual".  Some specific
  words that would be handy are:

    command    Map works in command mode
    input    Map works in input mode on a normal buffer
    history    Map works in ex and regexp history buffers
    motion    Map works when a motion is expected after an operator
    select    Map works when a visible selection is pending
    noremap    Don't expand maps in in rhs

  :map doesn't imply any of the "Map works..." modes, but if no modes are
  specified then it'll default to "command history motion".  :map! implies
  "input history".  Note that these are compatible with traditional vi,
  and aliases can be used to make :map act like vim's mode-specific mapping
  commands.

  To keep the map table clean, there are two ways a new entry can affect
  an existing entry:
   1.If the LHS matches an existing entry and some of the contexts are the
     same, then the existing entry should have those contexts removed.  If
     this means the existing map doesn't have any contexts left, then it
     should be deleted.
   2.If the LHS matches an existing entry, and the RHS also matches, then
     the existing entry's contexts should be merged into the new entry's
     contexts, and the existing entry should be deleted.
  The new entry is always added, regardless.

  Also, make sure that if the rhs starts with the lhs, then that portion
  is never subjected to remapping.

  The keywords should be allowed before either the lhs or the rhs.  Elvis
  has always supported "visual" before the rhs, but putting these before
  the lhs is more convenient in an alias.  To support this, the :map command
  can't depend on the ex parser to separate the lhs from the rhs, which seems
  odd.

  Abbreviations are similar to maps.  Although the flags used with maps
  don't really apply to abbreviations, there is a chance that the map changes
  could break the abbreviation table, so abbreviations should be tested too.

-------------------------------------------------------------------------------
Fixed in 2.2c-alpha:

/ The command ":map! X visual J" doesn't work.  ":map! X visual :normal J^M"
  sort of works, but it looses the whitespace.

    The first map should work now, but it'll work exactly like the
    second one -- without any whitespace between them.  I figure
    that's probably close enough.

/ One user reported that ^V{motion}d deletes text okay, but then pastes it
  all on one line.

    This happens when you're near the end of the buffer.  The pasted rows
    from a rectangle can only be pasted into existing lines.

/ Either the "fold" color or "spell" color isn't being allocated correctly.
  Elvis mixes them up.

    The "fold" color was being allocated the first time a fold was
    displayed -- which could be after some spelling errors have been
    highlighted via a temporary color, hence the confusion.

/ Should autowrite write files before :make, like vim does?  Yeah, I guess so.

/ Allow text object to work with the V/v/^V commands.  This is important!
  For example, "Va{d" deletes a block using whole-lines.  There is no other
  way to do that; "da{" and "%d%" both do character deletions, not line
  deletions.

/ Modify the behavior of ":set! option=value" so it changes the default value.
  I.e., store the new value as the default, and ONLY if the options wasn't
  explicitly set before it should change the actual value.  This should never
  set the option's OPT_SET flag, but may reset it if the new default is the
  same as the current actual value.

    This is about half-done now.  In its present state, there is a
    bug in the way it decides which options to specify in .elvisrc.

    Keep the desc->dflt pointer.  Leave it NULL for most options,
    indicating that they've never been set.  When an option is set
    via ":set" (no !) then store the old version as desc->dflt.  When
    set via ":set!", then store the new version as desc->dflt.  There
    is never any time when we take a snapshot of all options' values
    and call those the default.

/ In html mode, $ moves to the start of the following line, but looks like
  the start of the current line.  BOTH ARE WRONG!  It should move to the
  last character on the current line.

/ In visual mode, typing :<enter> should do nothing.  Currently it prints
  the line and advances.  (Patch supplied by Derek Peschel.)

/ ":vi +line" ignores the line when there's no filename.

/ Regexp changes:
  / Add a new "magicchar" option which lists the chars that don't need to
    be preceded by a backslash in order to be considered metacharacters.
  / Modify the !\x notation (backslash in an alias parameter) to support
    the new magicchar option.
  / Make the regexp error messages use \ only where needed.
  / Add support for alternation, the \| metacharacter.
  / Add support for using closure on complex subexpressions.
  / Add metacharacters for edge of a word: \B for not-an-edge, and either
    \b or \h for edge.
  / Add an option to choose the meaning of \b.  And eventually \uXXXX.
  / Add an option that disables Perl-style $name substitution.
  / Support $name substitution in replacement text of :s/re/text/ command.

/ There should be a way to handle Perl POD text as formatted text.  Basically
  this would require everything from =head1 through =cut to be treated as a
  big comment in the "documentation" font.  It probably isn't worth the effort
  to do anything special with markups within POD text, since elvis won't be
  used as the formatter, just the editor.  There should be a line in the
  elvis.syn file to indicate that the language uses POD, and what starts &
  ends it -- "documentation =cut", where "=cut" ends it, and anything that
  starts with "=" starts it.

/ Modify PostScript to automatically detect size of paper.

    Use "clippath pathbbox" to obtain the drawing boundaries.

/ Link error when config.h contains "#undef FEATURE_TEXTOBJ"

    This is because it also used "#define FEATURE_HLOBJECT", which
    requires text objects.  I fixed it by making "elvis.h" define
    FEATURE_TEXTOBJ whenever FEATURE_HLOBJECT is defined.

/ Allow <Shift-Arrow> to select text, like in most Windows programs.
  / For WinElvis
  / For XElvis

/ Need a way to distinguish between variable names and other unclassified
  text, in a syntax-colored buffer.  Perhaps this is another use of the
  "initialtoken" option.  Or a "mostly" option, which allows individual
  languages to be configured to use something other than "variable" for
  unclassified words.

  This is more important now, because spell-checking of variables is normally
  disabled, but in "syntax html" mode all words look like variable names.

/ The "syntax" display mode allows single characters to be declared as
  keywords (for example, to change their font), but if you do that and also
  use the character in a "useregexp" list, then the *character* is marked as
  preceding regexps, but the *keyword* is not.

/ The "showlink" option has the weird side-effect of making the current word
  be displayed on the status line, while in "man" display mode.  Harmless,
  but very odd.

/ The :browse and :sbrowse commands are broken -- the links don't work.
  This also affects the :kwic alias.  Apparently, somehow elvis lost the
  ability to handle file:...?command URLs.

/ Can't use incsearch as the target of an operator.

    Maybe disable "incsearch" when searching is used as the target of
    an operator.  That should be easy to do, and would probably be
    satisfactory to most users.

/ Incsearch doesn't set "wantcol" -- after a successful search, hitting j or k
  will cause the cursor to jump to its previous column, instead of staying in
  the same column.

/ In incsearch mode, the new $name variable interpolation doesn't work.

    The fix for this will probably involve calling regbuild() function,
    since that's where $name is implemented.  And if we're going to be
    calling that, then we might as well support the search flags too.

/ The visual command "/$" moves the cursor to the newline character, instead
  of the last character in the line.

/ Need a clean way to allow HTML files to be displayed initially in either
  "syntax" or "html" mode (whichever the user prefers).  Currently, all the
  user can do is modify elvis.arf to set bufdisplay differently.  This isn't
  good enough because setting it to "syntax" inhibits the ^Wd command, and
  may also affect some other commands.

    Added an "initialsyntax" buffer option, and a global "prefersyntax"
    option which is used to set "initialsyntax" automatically via the
    elvis.arf script.

/ Maybe add a "mkexrcfile" option.  String, defaults to "".  If set to some
  other value, then that's where :mkexrc writes by default, and :mkexrc is
  allowed to overwrite that file if it exists.  Whenever :mkexrc writes a
  file anywhere, it first sets mkexrcfile to the file name, so that the same
  file can be overwritten easily.

  The :cd command makes this dangerous.

/ In incsearch, ^V can't be used to escape control characters.

/ "hlsearch" and "selection" fonts don't mix well.  The foreground color
  is lost, at least with the "x11" interface.

/ Change the way "Elvis custom initialization" (for :mkexrc) is updated.
  Only consider updating it if the event counter has changed, so that complex
  initialization scripts only update it once.  The :mkexrc command regenerates
  the buffer before writing, just to be sure it is up to date.

  Ideally, it should be updated from event.c, near the end of each event
  handler function.  The ex.c file should simply set a flag indicating that
  the update is needed.

/ Need an easy way to store regexp text in variables.  Suggest Perl's $name
  or ${name} notation.

/ In expressions, make ":" operator concatenate path elements, using the
  OS-specific path delimiter (either ":" or ";").  This shouldn't interfere
  with its use in the "?:" ternary operator.

/ The :normal command has a bug.  The second invocation causes a core dump,
  at least when used inside an alias.  This prevents the :spell alias from
  working.

/ Modify the "x11" gui to support font names of the form "font*size", like
  the "windows" gui already does.  Rename "normalfont" to "font".

/ The cfmt alias is broken.  It seems to think all comments are C++ comments.
  Perhaps the error status of :s/// has changed?  I know I fixed the :g//s///
  behavior recently.

/ Folding interferes with <PgUp>/<PgDn>.

    This is probably because the dmnormal.move() function doesn't
    handle folds when the absolute value of the linedelta is > 1.

/ The :help command should allow = at the end of n option name.

/ It would be nice if the "aL" text object ignored preprocessor lines, the
  same way that it ignores blank lines.

/ In arithmetic expressions, using a regular expression as a function argument
  sometimes causes core dumps.

/ In incsearch, /<Enter> and ?<Enter> don't repeat the previous search.

/ The "autoselect" option was recently broken.

/ hlobject updates are too optimized.  If multiple windows, it won't always
  update the highlight range even after the cursor has moved outside of the
  highlight region.  Setting nooptimize hides this bug.  It seems to go
  away if there is only one window.  Maybe it is confused about the "current
  window"?

  Serious quirks in screen updates while editing, too.

/ Add an error message when a script contains an unterminated {...} block.
  Also, warn users to enclose the hlobject value in quotes so "a{" doesn't
  cause the parser to become confused.

/ Does hlobject sometimes cause foreground colors to be lost?

    Merijn Brand:
    If hlo/hll is on and the block (on the first level) is rather
    large, the block is backgrounded with the appropriate color, but
    the syntax highlighting of the bottom half of the screen
    disappears. I realize this a weak  which is rather
    hard to verify, but I've not come up yet with a
    example.

    Here's one way to reproduce it: ":man pcre".

    This is almost certainly an example of using up all of the colors,
    so any attempt to merge two fonts returns just one -- the hlobject<N>
    font.  Why are so many fonts being used, in such simple contexts?

/ <Enter> doesn't advance the cursor.

/ In incsearch, /\ will hang elvis.

/ After an incsearch is cancelled, hide the status info.

/ To support highlighting of whole lines, allow the hlobject value to use
  a 'V' prefix, as in `:set hlobject="Va{"'

/ Some objects cause core dumps, if the buffer is empty.  This is most critical
  when you have "set hlobject=a{" in your .exrc file.

/ Shouldn't draw guide lines on ~ rows after last line of file.

-------------------------------------------------------------------------------
Fixed in 2.2c-alpha:

/ Still need to implement "boxed" text in "windows" printer.  Also, the screen
  needs to support COLOR_LEFTBOX and COLOR_RIGHTBOX, plus the "guide" color.

/ :g/re1/ s/re2/text/ stops at the first line which contains re1 but not re2.

/ Syntax descriptions aren't being cached correctly.

/ "/\@" only looks for the first letter (anywhere in the found word), even
  though ^A still searches for the whole word.

/ Add vim's :normal command.  When invoked without arguments, it can still
  switch to "normal" mode for backward compatibility.

/ The "syntax" display mode should allow tabs between the function name
  and the ( character.

/ Here's a weird bug: When running "elvis +make", sometimes the screen
  is cleared to a funny color during compilation, and then when the cursor
  has moved to the first error and is responding to keystrokes, it is
  stuck on the second-to-last line (just above the status line).  This is
  *not* open mode -- Hitting <z><z> fixes the scrolling.  One big clue: screen
  updates are very slow, as though it was redrawing the whole buffer from the
  top each time.

/ ":color selection" can only set attributes, not colors.

/ In the manual, the <title> sections all say "Elvis 2.1".

/ Backspacing while in indentation with autoindent is very quirky.  Apparently
  it moves the cursor back too far, when the shifting operation doesn't delete
  as much as it might in some other context.  It only occurs when smarttab is
  set.

/ The showlink option is slow if there are no <a ...> tags shortly before
  the cursor position.  It should probably stop looking if it sees a tag
  which is a paragraph/section marker.

/ Many web pages uses non-Latin1 characters for quote marks and such.  Perhaps
  ":set nonascii=hack" should translate the illegal characters into legal
  ones.  It might also be nice if nonascii characters could be flagged with
  a distinctive font, too.
    0x85    :
    0x91    `
    0x92    '
    0x93    "
    0x94    "
    0x96    -

/ The command "/[[:space]]" (without the second ":") causes a core dump.

/ The "termcap" interface's graphic chars don't work?  Since when?

    Actually, it works but rxvt doesn't support graphic characters when
    using the Sony "12x24" font, because that font doesn't have graphic
    characters in the usual places.  xterm works around that, but rxvt
    apparently does not.

/ Regular expressions should support \x notation inside character lists
  (the [] notation).

/ Revise the way the "draw.c" module compares the old screen image to the
  new one.  Instead of storing the old font code, it should store the old
  font's attributes: foreground RGB, background RGB, and bits.  The new
  screen image should continue to store the font code though; we need it
  to find the color code for the RGB values.  This solves the following
  problems:

  o Temporary font codes are recycled after the screen image has been
    updated, so we don't know what attributes it had.

  o Some attributes (such as guides and box-ends) are synthesized inside
    "draw.c", and are not stored in any font code (temporary or permanent).
    Synthesizing them this way requires code to be scattered throughout
    "draw.c" instead of collected in reasonable places, and it also limits
    the efficiency of updates.

  This requires replacing the di->curfont array or 1-byte codes with a
  di->curattr array of 8-byte attribute bundles.  It also requires some
  changes inside draw.c, because the way fonts are stored and compared
  will be changed.

  HOWEVER, THIS WILL NOT FIX ANOTHER TEMPORARY FONT PROBLEM: The markup
  display modes store the font code of the start of each line.  That code
  could be temporary.  A separate solution should be found for that.

/ Another good idea: Add guide lines -- vertical lines superimposed at
  certain column numbers, configurable for each window (or for each buffer?).
  Multiple guide lines should be supported.

  This should be implemented as a new "guide" option, whose value is a
  comma-delimited list of widths.  Ideally, the "tabstop" and "shiftwidth"
  options should be modified to use the same type of values.

/ Modify the way "boxed" text is drawn.  In addition to the basic "boxed"
  attribute, it should also pass bits indicating whether the top, bottom,
  left, and right edges should be drawn.

    This would be more complicated than it sounds.  Characters could
    be forced to be redrawn even if they haven't changed.  Strings of
    characters that are all in the same font could be forced to be
    drawn in different gui->text() calls, because some use the top/bottom
    edges and some don't.

    A simpler alternative might be to pass in separate bits for the
    left & right edges (of the entire chunk of text, not each character),
    but always draw the top & bottom whenever boxed text is drawn.  This
    would also give an easy, portable way to draw the guide lines.

/ Extend the :color command to support an "or <colorname>" clause.  If the
  <colorname> has a higher contrast against the background than the current
  color, then then use it; else ignore it.  This would allow commands such
  as ":color prep light green or dark green" -- if the background is light,
  preprocessor directives would be dark green; if the background is dark,
  they'll be light green.

  This *might* not need to store the color that isn't used; the command
  string would be enough.  The only time this would cause problems would
  be if the foreground was inherited but the background was specified, and
  the specified background happened to be close to the foreground.  Elvis
  couldn't choose the alternate color then, because it would have forgotten
  what it is.  Perhaps that's a good thing though -- easier to understand.

  Storing only the chosen foreground and background would also allow more
  than one alternate foreground to be specified.  For example, on a green
  background, "light green or dark green or pink" would select pink.

  This would work best if the background color was always parsed first, so
  "white or black on white" would always resolve to black on white, even if
  the inherited background would have been black.  THIS PROBABLY ISN'T
  IMPORTANT since the "or <colorname>" clause is unnecessary if you're
  forcing it to use a known background anyway.

  The :lpr command would always choose colors with the assumption that the
  background is white.  This is actually trickier than it sounds -- colors
  are only selected when a :color command is given, not when some other
  factor causes backgrounds to change.  So :color would need to compute
  both the video foreground and the lpr background, and store them both.

/ "ref write" hangs.

/ The guimoveto() function calls (*gui->moveto)() even if it is NULL.

/ The readeol option is recalculated by lib/elvis.brf even if the buffer
  already exists.  That's wrong; it should leave the option unchanged.
  Also, maybe make -b force binary I/O.  Suggested by Alexandros M Manoussakis
  <[e-mail]>

/ Some changes to :color

    / Allow "+=" to merge new attributes into old attributes.

    / Maybe allow "-=" to support removal of attributes.  This is a little
      tricky for colors and fixed/proportional.

    / After "+=" or "-=", synthesize a new string which configures all of
      the set attributes in a single statement.  This is required for :mkexrc
      to work reliably, and for ":color" to list complete attributes.  Doing
      this will require foreground & background color names to be stored.

/ "3ofoo^[" will insert 3 foo's, but they'll all be on the same line.  They
  should each be on a separate line.  (This also affects the . command.)

    Actually, the standard "o" command doesn't accept a count.  But if
    elvis' version does, its behavior should be more intuitive that it
    currently is.

/ When :a or :i has text on the command line (instead of following), it
  leaves the cursor on the line below the inserted text, instead of on the
  last line of the inserted text.

/ Create a "showlink" option which displays (on the status line) the value
  that tagatcursor() returns.  This would be nice in "html" mode but a pain
  in normal mode, so perhaps it should be a display option.  Either that, or
  make it sensitive to the same thing that changes the <Enter> key's behavior.

    This option would be tested in the draw.c module, so making it a
    display option would be awkward.  Make it global and check for the
    <Enter> key thing (i.e., check whether display->tagnext is defined).

-----------------------------------------------------------------------------
Fixed in 2.2b-alpha:

/ In "man" display mode, any blank lines before a .P or similar command will
  confuse elvis, preventing you from moving the cursor upward over a paragraph
  boundary.

/ The :browse and :sbrowse commands need to be tweaked.  They contain a
  work-around for a minor bug in the "html" display mode, but that bug has
  now been fixed, so the workaround (an extra space after <pre graphic>)
  should be removed.

/ Modify lowline() and lowoffset() so that if looking for a line near the end
  of a block, the newlines are counted from the end of the block instead of
  the beginning.  That should be faster.

/ Add an "%%Orientation: Landscape" comment to ps2 output after "%%CreationDate"

/ UNC ("\\host\dir\file") names still don't work correctly under Windows.

    This turned out to be due to a bug in the way elvis was checking
    Cygwin's mount table.  The bug only affected Cygwin users who had
    something mounted as "/".  For example, if "C:\" was mounted as "/"
    then "\\host\dir\file" was being interpreted as "C:\host\dir\file."

/ On terminals with "brain damaged newlines" (the :xn: termcap flag),
  modifying the last column on a line would have unpredictable results.

/ Add color support to the "html" lptype. <font color="#rrggbb">...</font>

/ Can't print HTML documents via "-c" anymore?  It dumps core.

/ lptype=html produces bad output.  ">" is output as ">" without a ";".
  Also, since 2.2a-alpha, it has been changing the font for each character,
  instead of only between fonts.

/ Bug: Can't interrupt elvis with ^C or anything else, with any interface.

/ New features for searches:
  / 'i' flag for ignorecase, 'c' for noignorecase, 's' for smartcase.
  / 'w' flag to imply \< \> around regexp.
  / 'x' flag to imply ^ $ around regexp.
  / Vim's smartcase.  Regexps that contain any uppercase letters force off
    the "ignorecase" option temporarily.
  / Support for \i and \I character classes. \I is any char that can start
    an identifier, and \i is any character that can appear later in an
    identifier.
  / Vim's hlsearch option -- highlight *ALL* matches, without selecting them.

/ From Merijn Brand:
  Something is broke in syntax highlighting.  Try this on an excerpt from
  the ColourChooser module [using ":display syntax perl"]:

    $colour[1] =~ s/\s+//g ;        # Colourname
    $colour[2] = lc $colour ;       # colour name
    $colour[3] = $colour[2] ;
    $colour[3] =~ s/\b(\w)/\u$1/g ; # Colour Name
    $colour[4] = $colour[3] ;
    $colour[4] =~ s/\s+//g ;        # ColourName
    for my $i ( 5..9 ) {
        $colour[$i] =~ s/\d+$// ;   # Remove trailing digits
        }

  and see what happens :-( to the first line.
  now change the s/\s+//g to s/\d+//g. Ooooops!

/ ":color role1 like roll2 etc" includes all following text in the "roll2"
  name.  The "roll2" name *SHOULD* be exactly one word long every time.

/ On last line of a file, do "$v^=" to evaluate an expression.  The expression
  is stored on the last line, but the final newline is missing.

/ Add a :safely command.  It temporarily sets the safer flag (like :safer)
  while executing the rest of the line as an ex command (like :try or :then).
  This would be handy if the "ex" protocol was implemented via a :readex
  alias -- the alias could be smart enough to make "ex" links safe.

/ Add the :push command already!  Like :e, except that it saves previous
  position on the tag stack.

/ In "windows" GUI, after ":set number" you can't highlight whole lines via
  the mouse.  (The strip along the left edge of the window goes away?)

/ In "x11", the scrollwheel only works right in the text area.  Other windows
  treat it like any other mouse click.  All wheel events should be treated
  as though they occurred in the text area.

    Fixed... but the statusbar never requests mouse events, so that area
    is dead to scrollwheel events too.  That's okay.  I mostly wanted
    to make sure scrollwheel events wouldn't do anything disastrous in
    the toolbar.

/ Add a new() function, which creates a buffer but not a window, and returns
  the name of the buffer.  Optional argument is a desired buffer name, else
  it uses "Elvis untitled %d".

  This is the first time that functions will have side-effects.  If you want
  to evaluate an expression without displaying the result, just to get the
  side-effect, then you could use the :if command without a :then.

/ "make clean" should remove the executables.

-----------------------------------------------------------------------------
Fixed in 2.2a-alpha:

/ Font/color changes:
  / Support the new :color syntax
  / :color can add other attributes: bold, italic, underlined, boxed, fixed.
  / This only specifies *CHANGES* to colors or attributes.  Unspecified
    colors are retained from previous font.  Attributes are ORed.
  / Limits: 128 font names (so we can have standout versions).
  / Each font is stored as a struct containing a pointer to its name,
    fgcolor index, bgcolor index, attribute bitmap.  The name-to-index
    conversion can be slow -- it is only used when a display mode initializes
    itself.
  / When constructing a new screen image, for each character we store a font
    index, optionally with a bit set to indicate a selection.  This is compact
    and should be good enough when comparing images.
  / After an image is generated, it is combined with the "normal" font to
    inherit from it any colors/attributes which are not explicitly set.
    elsewhere.  The can also be an "idle" font.
  / Change all display modes to use new font convention.
  / Change all GUIs to use new font convention.
    (Except "vio", for now.)
  / Change all printer drivers to use new font convention.
    (Except "windows" is untested.)
  / Update the manual.

/ In input mode, make ^Z move rightward across any backspaced text, keeping
  it instead of deleting it.

/ Change the version number to 2.2a-alpha

/ Add smartargs option for displaying a function's arguments.

/ Modify "ref" to look through elvispath.  Put a file of ANSI C stubs into
  elvis' lib directory and generate a "tags" file from it.  This will allow
  "ref" to act as a reference for the standard library.

-----------------------------------------------------------------------------
Fixed (or not) in version 2.1_4:

/ The built-in calculator's shell() function gives an ugly error message
  if the safer option is set.  It fails correctly; it simply looks ugly.

/ POSIX says that ^F/^B should leave two lines overlapping, not just one

/ POSIX says that $ should accept a count, and move forward (count-1) lines.

/ Executing ":do source somefile", where somefile contains an inner :while
  loop, messes up the condition for the outer loop.

? The text-mode version of elvis sometimes has problems drawing characters
  near the left edge of the window.  Apparently it gets confused about where
  the cursor is located.

    I've seen this on some occasions, but I not recently.  I've never
    been able to reproduce it reliably.  Perhaps it was a side-effect
    of some other bug?

/ Under MS-DOS with TERM=nansi (and NANSI.SYS installed, of course) the screen
  is shifted up one line.  TERM=ansi works, but doesn't use nansi efficiently.

/ For gui="x11", if a window scrolls while some other window has pointer
  focus, then the cursor isn't drawn.  This is important when, e.g., you
  use a dialog to search for text.

/ When the same buffer is shown in two windows, inserting text into one
  window can cause the other to scroll.  This seems to occur only when
  the location of the change is above the top line of the other window.

    Apparently what's happening is this: In a series of blank lines,
    (i.e., lines which contain only a newline character), after any
    character is inserted before line n, line n-1 has the same offset
    than line n used to have, so the window update functions assume
    that line n-1 is really still line n, so it shows line n-1 at the
    top of the window instead of line n.

    Probably the win->di->topline variable, which is currently a long
    which stores an offset, should be replaced with a MARK.  Since
    MARKs are updated when a buffer changes, this should fix it.

* Given a URL such as "../untar.c", referenced from "/pub/elvis/unreleased",
  elvis 2.1_3 tries to fetch "/pub/elvis/unreleased/../untar.c", but it *MUST*
  be normalized as "/pub/elvis/untar.c" -- i.e., delete instances of "dir/../".
  Some web servers depend on this.

? In "html" mode, if a single displayed line contains more than one link,
  elvis always seems to follow the *last* link regardless of which one you
  click on.  This is bad for some menu bars.

    Actually, this only seems to occur when multiple images appear
    on a line, and you want to download one of the images.  For
    real links, elvis works correctly.

    I took a stab at fixing this one.  It seems better now.

? There appears to be a bug in scanning -- a scanned block is not necessarily
  being locked for the whole time that it should be, which means that it may
  be removed from the block cache, in which case NULL may be returned by some
  of the blkXXXX() functions.  Also, on at least one occasion elvis appeared
  to be trying to free a block which wasn't allocated.

    This may have something to do with the FEATURE_LITRE feature, or my
    recent attempts to make searches run faster by keeping at least one
    scan context for the current location at all times.

    The FEATURE_LITRE bug is fixed; hopefully that fixes this too.

/ There are still problems with running external programs under Win32.

/ When viewing an HTML document at a URL such as "http://foo/bar.html?boo/far",
  containing a link to "zot.html", elvis will resolve that link to be to
  "http://foo/bar.html?boo/zot.html", but it should be "http://foo/zot.html".

  Similarly, I suspect that fileext() fails on URLs like that.

/ Hitting ^] on an operator doesn't search for an "operatorXX" tag?  I
  thought it did.

    It appears to work now.  I don't know why it didn't work before.

/ The alias.c program (which is compiled as vi.exe, ex.exe, and view.exe)
  has been reported to cause problems under WindowsNT, and more recently
  under Windows98 as well.

    It appears that a Microsoft has introduced a bug into their versions
    of the "exec" system calls.  I switched to "spawn", which waits a
    little RAM but seems to work correctly.

/ Under Windows95/98/NT, the "Terminal" font has funny spacing and sometimes
  characters overlap each other.

    This is a bug in the font, not in elvis.  Elvis only supports
    fixed-pitch fonts, and although "Terminal" claims to be fixed-pitch,
    it is really a variable-pitch font.

/ Control chars in an alias aren't displayed in a printable form by the
  :alias command.  ":alias clean s/.^H//g" looks like "alias clean s///g".

/ Add VIM's "smarttab" option -- <Tab> keys at the beginning of a line act
  like ^T, so they indent by the shiftwidth amount using a combination of
  tabs and spaces.  <Tab> keys elsewhere in a line are real tabs.

/ One user reported that elvis was sensitive to keystrokes during quitting.
  I suspect that gui->poll was being called while the file was being written
  (possibly just when written via FTP).  That write should not be quittable.

------------------------------------------------------------------------------
Fixed in 2.1_3

/ One user reported crashes in WinElvis when the "Options->syntax" menu item
  is activated while in "syntax" display mode.

/ Can't identify tags whose whose contains a new-style comment.  Apparently
  the tag-selector can't parse \/\/ in an regular expression.

/ Make elvis assume "-Gopen" if termcap can't be used.

? The :make command is still slightly weird.  "elvis +make" waits until after
  a keystroke to display the first error message.

/ Document the ftp://host/~user/resource URL notation.

/ The :mark command should save the column number as well as the line number.
  (I.e., if no explicit address was given, then use the cursor position instead
  of the start of the cursor's line.)

/ Add a :cbshow alias which displays cut buffers.  To do this right, elvis
  will need a text() function for fetching text lines...
    text()        return the current line of the current buffer
    text(n)        return line #n of the current buffer
    text(b,n)    return line #n of buffer (b)

/ ":map ^H visual X" doesn't work when the cursor is at the end of the line.
  It deletes the last character, but then moves the cursor back two characters
  instead of just one.

/ Assertion fails at buffer.c:1206 when attempting to write an empty buffer.

? dmmarkup.c:1066: htmlinit: Assertion `first == True && col == 0' failed.

/ "Dead keys" don't work in "x11" gui.

/ If the .elvisrc file contains "set wrapmargin=0" then that option is
  displayed by a later ":set" command.  It shouldn't be.

    This is a complication arising from the fact that wrapmargin is
    actually implemented via the textwidth option.  Wrapmargin doesn't
    exist until a buffer is shown in a window, but defaults are normally
    recorded before that.

/ If there is no tags file in the current directory and inputtab=tagname
  then there is a long delay each time <Tab> is hit.

/ Is this a good time to fix the bug where :make outputs the first error
  message *before* prompting the user to hit <Enter>, when it should be after?

/ Add a "howto" describing Perl's error messages.

/ The "noanyerror" option shouldn't prevent the cursor from moving into locked
  files in the current directory -- only unwritables in some other directory.

/ It's back: Sometimes in "man" mode the cursor can't move up.  This appears
  to happen only for .IP or .TP commands with long labels.  I should probably
  change that code anyway, so the labels are considered to be separate lines
  in that situation.

/ It would be nice if FTP URLs supported user names as in ftp:host/~user/...
  in addition to just ftp:host/~/...  When a user name is specified, the
  ~/.netrc file is searched for an account at that host which has that login
  name; the password is then taken from that entry.

? More importantly, it has been reported that the Windows versions don't
  read "elvis.ftp" correctly.  (It works under Unix; I've tested it.)

    This appears to work under Win32, as far as I can tell.  Perhaps
    the user who reported it didn't have "elvis.ftp" in a directory
    where elvis could find it?  Or maybe the other changes I made to
    that file (to support ~user) had the side-effect of fixing this?

/ Add support for backticks in filenames -- the text between backticks is
  interpreted as a shell command, and the command's first line of output
  is substituted into the line.

    The main reason for supporting backticks is to do things like
    ":n `grep -l sometext *.c`".  The backtick expression would return
    multiple file names.  Because of this, it must be done at a fairly
    high level in the ex command-line parser.

    It would also be handy to have a "shell" function which runs a
    command string and returns its output.  That function would be
    handy when implementing backtick support, and could also be used
    for things like...

        :let d=shell("date")

/ Under Win32, can't always redirect an external program's stdin.  For example,
  ":%!\windows\command\sort" fails, but ":%!\usr\bin\sort" works.

    This is ridiculously difficult.  I/O redirection depends on whether
    the program being run is 16-bit or 32-bit, whether you're running
    it from WinElvis or elvis, whether you're running Win95/98 or
    WinNT, and... something else that I haven't figured out yet.
    I give up.  Instead of trying to do it the efficient way (with
    pipes when possible, and always using Win32 stdio redirection
    facilities), I decided to rewrite it from scratch to do it the
    brain-dead way: By constructing a shell script, and invoking
    COMMAND.COM to run that script.  This sucks, but it appears to
    work reliably.

? On some systems, elvis sometimes reports "# is illegal during initialization"
  and then it skips some initialization steps for the first file.  Later files
  will be initialized correctly; it only affects the first file.  This has been
  reported under SCO and (I think) Irix.  Unfortunately, it seems to interfere
  with the "-VVV" verbose feedback.

    I've improved error messages -- they now include the filename and
    line number where the problem was detected.  Hopefully that'll allow
    me to get a little more feedback.

    I just tried the newest code under SCO, and this didn't happen.
    It worked fine.  This is definitely a change in its behavior, and
    that's good... but I wish I knew why it started, and why it may
    have stopped.

/ Add a "digraph" topic to the howto file, and improve the documentation of
  digraphs in the main manual.

/ Lose the "tagprgonce" option; the :local command offers a cleaner solution.
  You can just say "local tagprg=..." in the alias, and be done with it.

/ The [[ and ]] commands shouldn't be line-mode moves.  I.e., from the middle
  of a line, d]] should not delete the entire current line.

/ When a character-mode deletion ends at a newline, that newline should not
  be deleted.  The same goes for yanking.

? The tagprg option (or something similar) should be used when reading tags
  for tag name completion.  Apparently it isn't.

    Actually I think it is, but maybe not quite correctly.  The code
    assumes the typed part of the tag name is always going to match the
    initial part of each match.  If the tagprg program doesn't do that,
    then name completion gets confused.  Changing this behavior will be
    hard.

/ In a shell command, file name completion should match binary files.

/ After a repeated insert such as 25i*<Esc>, typing <.> will only repeat 24
  insertions, not 25.
------------------------------------------------------------------------------
ELvis 2.1, official released version

/ Some HTTP sites require a "Host:" line in the GET request.  The argument
  to "Host:" is the site name and port of the URL elvis is trying to download.

/ Check for licensing wordage in manual or README file.

/ In the "configure" script, xinc is set to /usr/include/X11, when it should
  be set to /usr/include.

? Under WinNT 4.0sp3, the "vi.exe" and "ex.exe" commands have been reported
  to cause problems.  Input doesn't work right, and the whole DOS window
  goes away when elvis exits.

/ In macros, !(.)% should work but doesn't.

? Elvis has been reported to run out of memory when reading a large LaTeX
  file -- "http://www.loria.fr/services/tex/general/latex2e.html", although
  it isn't clear whether elvis is reading it via the net, or from a local
  file.

    It worked for me.  The user didn't report what OS he's using.
    I suspect he's running the DOS version, in which case it isn't
    surprising that he's running out of memory (even with Win32 on
    a 64meg Pentium).

/ Change the default configuration to load lib/exalias.eg if there is no
  custom version in the user's home directory.  Maybe rename it to
  "elvis.ali" or something like that.

/ Add an "lpheader" option, and modify dmnormal.c:header() to use it instead
  of the "lplines" option.

/ :s/re/cmd/x sets the cursor to the wrong line -- cmd should run on the line
  where re was found, but it is incorrectly running on the following line.

/ With textwidth set, if the first line of a file contains no whitespace
  and extends past the textwidth column, elvis dumps core.

/ Aliases tend to step on each other's use of options.  Many macros temporarily
  set report=0 to avoid chatter while the alias runs.  To restore it, they save
  the old value in the "r" option, and copy it back afterward.  If if one alias
  calls another and they both do "let r=report", then the original value of
  "report" is lost.  This is one specific example of a larger problem: aliases
  need local options.

    I've added a ":local" command.  It is like ":set", except that it
    saves the old values of any options on a stack, so they can be
    automatically restored when the alias exits.  Also, ":local" never
    outputs any options; this allows you to mark an option as being local
    without immediately changing its value.

/ Both "x11" and "windows" have an option named "toolbar".  It behaves the
  same on both.  Soon they'll both have "statusbar".  They already had
  "scrollbar" but I changed the name of the X version since they behaved
  differently.  I should try to unify this somehow -- both in the program
  and in the manual.

/ The ":gui ~Label" notation for deleting a button doesn't work, but the
  postfix ":gui Label~" version does.

/ It would be nice if "gui <label>" would output all info about the given
  button, under X11.

/ The "tex" mode needs some work, especially for \xxx{text} notation.
  Several common fonts are not supported, including \code{text}.

/ Aliases can't run during initialization?  Remove that restriction.

/ For gui="x11", cutting & pasting (between applications) large amounts
  of text doesn't seem to work.  The text gets jumbled up.

/ Write a "howto" file, and an alias to access it.

/ In an empty buffer, search commands seem to get stuck in an infinite loop.

/ The "inputtab=identifier" code should avoid looking up matches when the
  word is a keyword.  When I type "int<Tab>i;", I don't want to get
  "intlmsgi;"

/ Add support for cygwin (GNU utilities and API under Windows).  The only
  unusual things are 1) it needs the local inet_aton() function even though
  that function is declared in a header, and 2) It doesn't seem to open files
  in binary mode.  It should, both for session files and for (non-binary)
  text I/O.

/ On some Linux systems, the variable name "restrict" in tagelvis.c seems to
  clash with a header symbol.  Changing it to "rest" avoids that.

/ Core dump from io.c:564 when bufdefault is NULL.

/ When abbreviations are expanded via the <Tab> key, elvis inserts a newline
  after the expanded form, instead of the tab.

/ Apparently ":try s/foo/bar/" doesn't turn off the the "then" flag when the
  substitution fails.

/ Some QNX problems:
  / <sys/select.h> isn't included by every file that needs it.
  / ftpperms is defined in two places.  (Lacks "extern" in header file.)
  / inst*.sh files don't return a good exit code, which interferes with
    "make".  Add "exit 0" to the end of each script.

--------------------------------------------------------------------------
Fixed in 2.1j-beta....

/ File name completion doesn't always work on directory names, when the
  completebinary option is off.

/ It would be nice if, when listing multiple matches, filename completion
  would leave off the directory path.  That way more matches would fit on
  a single line.

/ It would be nice if <Tab> completed tag names when editing normal source
  code.  That would save you a lot of typing if you like long identifiers.

/ On HP, "configure --with-gcc" isn't honored -- it still uses "cc -Ae".
  Also, HP (and probably some other platforms) expect main() to return
  an int, not void.

/ If "elvis.arf" executes a :s///x command, then the window doesn't realize
  that the buffer has changed so it doesn't set the display mode accordingly.
  This shows up when you're editing non-HTML text, and do "e url" where
  "url" is a URL or file which doesn't end with ".html", but which does
  contain HTML text -- the buffer's bufdisplay option gets set correctly,
  but the window isn't switched to html mode.

/ The "alias.c" file can be compiled for Win32 and OS/2, which may be
  handy for some people.  To eliminate warnings, it needs <stdlib.h>
  and <process.h>.  Also, there's a "pid" variable which isn't used.
  Some shells don't pass the program name in argv[0] so we need to
  handle that -- probably compile it with a -Dxxx to make it default
  to "ex", "vi", or "view".

/ File name completion should be case-insensitive on non-Unix systems.

/ The command "elvis -i" (with no file name) creates a buffer which
  doesn't contain any text.

/ From some URLs, a reference to a full pathname generates an error message
  about an "unknown site" or something like that.  For example, from
  http://www.fh-wedel.de, try following the "English pages" link.

/ After ":se report=1", hitting x says "1 line yanked".  Need to make
  reporting sensitive to the partiallastline option.

/ Make elvis.arf strip off the perl version number from #!/usr/bin/perl-XXXX
  lines.  This makes syntax easier to recognize.
-------------
* If xcurses exists, it should probably be used in preference to curses.
  (This would be a change to the "configure" script.)

? Elvis gets confused if you load a URL which has no filename component,
  such as "http://www.yahoo.com".  Adding a trailing slash avoids that.

/ Document the meaning of "!" for each ex command.

/ If a :tag fails because the current file has been modified, then that tag
  is still deleted from the list so ":w|tag" doesn't work.

/ Add ~/.exalias to the distribution.  Maybe extend it with a few other
  useful aliases such as "courier", "copying", "rm".

/ When an href in a remote web page contains just a file name elvis tends
  to mangle it.  From ftp://ftp.cs.pdx.edu/pub/elvis/unreleased/README.html,
  <A HREF="/pub/elvis/untar.c"> is treated (under Unix anyway) as a reference
  to ftp://ftp.cs.pdx.edu//ub/elvis/unreleased/pub/elvis/untar.c

/ Got stuck in an infinite loop when doing =} at the end of the buffer.
  There was no blank line at the end of the buffer.  The bug seems to
  occur after the buffer has been changed, when elvis is trying to update
  the screen.

    This appears to be a reliable way to duplicate the same big
    session file bug I've been tracking for months!  At last!

    It goes away when LINECACHE is #defined to be 0.  Apparently
    the caching in lowoffset() is broken.

/ Can't search for 0x00 bytes.  /^X00^M doesn't work.

    Added \0 for matching NUL bytes.

/ Trouble with the :& and :~ commands:
   / They are supposed to discard any flags from the original :s command,
     and accept new flags.  (This implies that the visual ~ command also
     discards the old flags.)
   / :& should always use the regexp from the previous :s command, but
     :~ should use the most recent regexp used in any context.
   / :s with no args should act like :&
   / Also, the count isn't supposed to be an instance specifier, it is
     supposed to be the quantity of lines to scan.  (The current instance
     specifier is handy, though, so invent a new notation for it -- perhaps
     a number after a '.' could be an instance.)
   / Add 'c' flag.  This will require the creation of a new edit mode which
     searches forward for next match within a range of lines, moves the
     cursor there, highlights it, and waits for a keystroke.  <Esc> exits
     that mode; <y> replaces and moves on, anything else moves on without
     replacing.
   / The print flags don't work.
   / global subst for a zero-length regexp doensn't work.
   / If a match is found at the start of the first line for a :s///c command,
     the cursor is displayed in the wrong location.
   / "confirm" doesn't work right in ex mode -- only in vi mode.
    Modify it to use the same state-insertion technique as visual-mode,
    and also make it exit ex mode.  The subst structure should be
    extended to have a flag which indicates whether it should revert
    ex mode when the last substitution is complete.  To revert, it
    should replace the "confirm" state with either the old "ex" state
    or a new one.
   / check for memory leaks

/ Add an alias() function, which tests for the existence of an alias with a
  given name.  Modify the default elvis.[ab][rw]f files to check for an alias
  and use that if it exists.  This will make customization easier.

/ When using <a name=...>...</a> to define a name anchor, elvis shouldn't
  display the anchor's text like an href link.

/ In "man" mode, blank lines should not affect indentation.  I.e., they
  should be interpreted  as ".sp", not ".LP".  Exception: If the blank
  lines are followed by any "." command which causes a line break, then
  ignore the blank lines.

  Also, for readonly files it should allow any number of args to a .BI
  command (or similar commands).

/ In "tex" mode, } should restore the current font to the stored font
  (which it does in 2.1i) and then set the stored font to "normal" (which
  is new).  This will help it handle embedded {...} pairs better.

/ The autowrite option is too eager.  After :e thisfile, :sp, :e otherfile,
  elvis writes thisfile even though it is still visible in the first window.

/ Add the PERL "Artistic License" to the distribution.  Earlier I'd
  announced plans to distribute elvis under both the old agreement and
  the GPL, but the Artistic License does pretty much what I wanted.

/ In macros and aliases, the current buffer and the current line within
  the buffer are not changing as one would expect.  This makes macros
  much harder to write than they should be.

/ ":let x=TERM" causes a core dump.

/ With Sun + fvwm + -Gx11 + ":set warpback", the pointer is warped back to
  the xterm's scrollbar when you exit.  This is unfortunate because Sun's
  xterm's scrollbar eats keystrokes.  The pointer should be warped back
  to the middle of the top edge of the window.

/ Add an example of command-line editing to the manual.

/ Under DOS and Win32, the ctags program doesn't work.  It doesn't seem to
  accept lowercase letters at the start of a word.

/ There's a problem with file name completion.  When nocompletebinary,
  if the only match is a binary file then it backspaces over the file name.

/ Problem with "ref" program -- it doesn't seem to handle "file:" correctly.

/ In "tex" mode, a % on a line by itself causes the following line to be
  ignored.  It shouldn't!  Also, \' should be displayed as \', not just '.

/ Option value completion: If while entering an ex command, the user
  types something like ":se kp=<TAB>" then elvis should insert the
  current value of the "kp" option.  The user can then edit it.

/ For Option value completion, need to insert a backslash before each
  '|' character, because '|' normally marks the end of the command.
  NOTE: Check whether this is done for :mkexrc -- it should be!

? The "windows" GUI needs to insert backslashes before '(', ')', and '$'.
  Look for addquotes() in guiwin32/gw{cmd,msg}.c

/ Bug in :calc (filename " " (333 * 100 / 400)) >> 58

    This appears to be due to a quirk of the concatenation notation.
    Since there is no explicit concatenation operator, it's difficult
    to assign it a precedence level.

/ Maybe support name completion for tags?

/ If "showtag" is set, and the tags file is out of date, then you get a
  bunch of "no match" messages for the tags that it can't find.

/ On an empty buffer, it still says "partial last line".  That flag shouldn't
  be set for new, empty buffers.

/ Need to be able to create new files on FTP sites.  (And if you can do
  that, then update the manual too.)  It would also be nice to have a
  better way to detect directories.

    I have been unable to find any FTP protocol commands for testing
    the attributes/existence of a file.  So I'll need to abuse other
    commands and make some guesses.

    To test whether a name is a directory, I can try to "cd" into it.
    Elvis always uses full pathnames for FTP access, so the current
    directory doesn't matter.  But if I can "cd" into it then it must
    be a directory; else it either doesn't exist, or is unreadable, or
    is something other than a directory -- hopefully a normal file.

    To test for the existence of a file, I should probably try to read
    it; if the read process starts okay, then I can abort it and know
    that the file exists; otherwise it is unreadable, not a file, or
    doesn't exist... and in all of those situations I should assume it
    is a new file.

    If I know it exists and is a file, then I can test for writability
    by attempting to append to it.  If the APND command can be started
    correctly, then shut it down without actually appending anything and
    make note that the file is writable; else it is read-only.

/ In replacement text, \^M should be a literal carriage return character,
  not a newline.

/ In the "man" display mode, 2 or more consecutive newlines (i.e., one or
  more blank lines) should act like a ".PP" command.

/ In the "man" display mode, add .TS and .TE -- like .nf and .fi, except
  that .TS has the added quirk of hiding the following line.  This will
  allow elvis to display tables, sort of.

/ The buffers created when reading a directory, browsing, etc., should be
  marked as "readonly".  Without this, it is too easy to get a "Hit ^Wd
  to see source" message.

/ Should :@@ execute the anonymous buffer?  Elvis has trouble parsing it,
  and after that it executes the same buffer as was most recently used.

/ It should be possible to edit cut buffers, particularly so you can edit
  @ macros.  To do this, I'll need to make sure a cut buffer doesn't get
  yanked into itself.  It would also be nice if the line/char/rect status
  wasn't embedded in the buffer's text (make a new "putstyle" option).

/ Problems in untar.c?  Reported to fail on some archives.

    Walter Briscoe figured this one out for me.  I was altering
    a block in the sliding window, when that block was interpretted
    as a tar header.

/ A little bit of name caching would be a big help for network name lookup.

/ Filename completion really needs work.
  / At the very least, it should exclude '<' and '>' from the list of
    characters that can appear in a file name.
  / It shouldn't be *too* hard to make a parser which guesses whether
    filename completion makes sense in the current context.
  / It might be a good idea to make filename completion be disablable via
    #undef FEATURE_FILENAME.
  / Another nice addition would be to make it ignore binary files.

? :make can't parse AIX error messages, which look like this...
    "cfunc.c", line 511.38: 1506-045 (S) Undeclared identifier incase.
  I suspect the ".38" column number is tripping it up.  It would also be
  nice to ignore the "1506-045 (S)" garbage.

    That's almost certainly the right diagnosis.  When elvis divides an
    error message into words, it permits '.' characters in those words.
    But it didn't allow '.' in a number, so "511.38" was perceived as
    garbage.  I changed the allowable number format to permit '.'.

/ The autoselect option is broken.  If set in .exrc, then it is reset the
  first time you give a search command.

    This is an unexpected consequence of changing the definition of
    "explicitly set options".  To fix it, I'll need to add a static
    variable in search.c, and set the variable to True if 'v', to False
    if 'n', and to the value of autoselect if neither 'v' nor 'n' is
    given.  Then use that variable in place of the autoselect option,
    when a search successfully finds a match.

? Problem with .exrc scripts (and other scripts too) -- :map lines frequently
  end with a ^M character.  Since this ^M appears immediately before the LF
  that marks the end of the line, it looks like a CR-LF pair and so the file
  may be seen as a DOS text file.  When interpretted on a DOS/Windows platform,
  the ^M will be deleted.  It also messes up editing of the script on any
  platform.

    I changed the way the fileeol() function recognizes file types.
    If the first 100 bytes of the file contains at least one instance
    of a LF which isn't preceded by a CR, then the file is assumed to
    be a Unix text file.  This should allow Unix .exrc scripts to be
    edited on a DOS/Windows system.

    To actually execute them, they should be converted to DOS/Windows
    format.  That way, when a map command ends with ^M, the line that
    it's on will end with CR CR LF, which will be handled properly by
    elvis.

/ Options aren't locked by "safer".  Some should be, including "safer"
  itself.  How long has this security hole been there?

/ Really need a way to queue up messages.  Add :message, :warning, :error.

? When GCC is used under HP, don't use the "-Ae" flag.  The normal "cc"
  compiler doesn't work under HP.

    I suspect this was due to some new-style function declarations
    that were in 2.1i.  These have been fixed, so HP's "cc" should work
    now.  "cc" is preferable to GCC on HPs because "cc" supports shared
    libraries.

* Two ^Cs in sequence can kill elvis.  (OS-dependent)

* When in "html" mode, maybe also make the <Space> key page forward, and the
  <Backspace> key pop.

* Can't run a shell command in the background under X11, because elvis waits
  for an EOF condition on its stdout/stderr.  Should something be done about
  that?  (You can do it if you redirect the program's stdout and stderr.)

    Maybe make the x11 prgopen() wrapper check for a '&' command, and
    force "willread" to False in that situation.  But then prgread()
    would need to be smarter, or something.

* After <Shift-K>, the window's wantcol value is messed up -- if you type
  <Shift-K><j>, the cursor will move to the first character on the following
  line, instead of staying in the same column.  (This may affect all external
  commands -- test it!)

    This only affects <Shift-K>, and even then only in some situations.
    The column that it moves to varies.  I wasted a whole day trying
    to figure this one out, with little success.  It seems to have
    something to do with the copies of wantcol that are stored in
    the state stack.

* ":set global?" scrolls off the top of the screen.  Something should be done
  about that.

    The fix will probably have to wait until "shell windows" are
    supported.  They'll allow you to scroll back through the output,
    like an xterm window.

* Under MS-DOS, "-o file" can cause a NULL dereference.  File permissions?

    Or memory limitations?  I looked through the code and it appears to
    be correct.  It is simple enough that it I don't think there could
    be any subtle bugs in it.

* Big clues on why sometimes session files can't be recovered:

    > This may be interesting. I will pursue it if given instructions.
    >
    > I got a GPF on a telnet session in which I was running elvis.
    > `elvis -r` got "blkread failed". The read failed with errno set to
    > 22 which is a singularly unhelpful EINVAL.

    Probably the EINVAL was left over from some earlier system call, so we
    can ignore it.

    blkread() failed because it was asked to read a non-existent block
    number.  Block 29474 would be at offset 60362752, but the session
    file you're recovering is only 344064 bytes long.  So the read() in
    blkread() returned 0 bytes, which would not affect errno.

    > I infer there is a window of time during which a session file is
    > not clean.  I don't understand the details of this and don't feel
    > like deducing them.  What do you want to be done?

    Apparently elvis needs to inspect the session's low-level buffers more
    carefully before restoring them.  In your session file, there is a
    low-level buffer (at block 52) which contains a reference to a bogus
    block number, and that's preventing you from restarting elvis to
    recover the other buffers in that session file.

    As I recall, the most common source of bogus block numbers was when
    a blklist block was reused as some other type of block.  I suspect
    that in your session file, the buffer at block 52 was freed, along
    with its blklist blocks, but the superblock was never flushed after
    that.  And the blklist block was recycled as a chars block for
    some other buffer, and that *was* flushed.  (Block numbers are
    two bytes long, and 29474's bytes are '"' and 's').

    So probably in addition to making recovery stricter (so it can ignore
    bad buffers), elvis should also probably flush the superblock
    immediately after freeing a buffer so that this particular type of
    bad buffer can't occur.

    (Yes, I am going to save a copy of this response for myself.)

/ :mkexrc doesn't insert enough ^Vs in front of Esc characters.

/ Under X11, trouble with "-fn -sony...." argument.  I think the "-f logfile"
  option is causing problems.

    Changed to "-o logfile".  Sadly, "-o" *MUST* be handled before
    the GUI initializes itself, so that we can log problems that
    occur then.

/ With no address := should report the total number of lines.

/ In replacement text, \U and \L should affect literal text as well as
  & and \1 ... \9.

? :N, :previous, and :rewind don't work at the end of the args list.
  Apparently they check the wrong end of the args list.

      Unable to reproduce this one.

/ In "untar.c", the typedefs are apparently clashing with similar typedefs in
  <stdio.h>, on some systems.  Change the names of those types in untar.c.

/ Check the manual for typos!

/ Need to add NEED_MEMMOVE, and use it for older SunOS.

/ The "man" display mode doesn't handle long section names.  At the very least
  it should truncate them!

/ To compile elvis under Solaris2 with "cc" (not "gcc"), you need to add
  "-Dconst=" to the value of the CC macro in Makefile.

/ In WinElvis, the "windows" lptype is counter-intuitive.  It changes the
  lplines and lpcolumn options to correspond to the font size and paper size.
  This is the opposite from the "ps" and "ps2" lptypes, which change the
  font size to squeeze lplines by lpcolumns characters on a page.

/ The sesclose() function shouldn't attempt I/O on the session file unless
  the session file has been successfully opened.  Currently, elvis can get
  stuck in an endless loop giving fatal error messages if you set SESSIONPATH
  to a bad value.

/ Sometimes crashes after running multiple external commands.  Particularly
  noticible with :make, but I think it affects all external programs.
  Reported under Linux.

    I used to be able to reproduce this easily, but now I can't.  Either
    some other change fixed this or changed its symptoms, or I just forgot
    how to reproduce it.

/ A core dump has been reported while entering a ":x" command.  The dump
  apparently occurred inside the mapdo() function after most of the local
  variables got scrambled.

    This may be a variation of the "crash after running an external
    program" bug.  The :x command can save the file, and if the backup
    option is set, then elvis runs "cp" or "copy", depending on the OS.

    In addition, a potential bug in the termcap interface was fixed.

? Some crashes/hangs have been reported under Linux.  These may have something
  to do with <u> and <Shift-U> commands.  Usually there are no clues, but at
  least twice the buffer filled with ^@ characters.  Reported by Steve Woodard,
  [e-mail]

    I enabled assert() checking and allocation debugging, and played
    around with those commands hundreds of times without being able to
    reproduce it.  It may be compiler- or OS-dependent.

    1/20: It happened to me while I was just editing a plain old text file.
    One of the numbered cut buffers' bufinfo blocks was recycled as a text
    block, and that cause blkread to fail as it tried to find a bogus
    blklist block.

    This happened to me again, shortly after a <Shift-U>.  Definitely a
    problem there.  I've inspected the code though (bufundo() in buffer.c)
    and it seems correct.

* The "ref" program doesn't understand compound tag addresses, such as
    bar    foo.h    /^struct foo$/;/^ int bar;$/;"    struct:foo

* The "ref" program should be modified to support "./tags".  In fact, does
  it even support "tags" (file name instead of directory name)?

    How can it support "./tags" when ref doesn't know which file you're
    editing?  To do this right, I'd probably need to add a "-f fromfile"
    option.

/ The value of the blkcache option can have a *HUGE* effect on the speed of
  elvis.  Why should a large cache be necessary?  A macro may crawl when
  blkcache=150, and fly when blkcache=200.

* An ex search command (:/foo) doesn't highlight text when autoselect is true.
  This is unfortunate, because the X11 [Search] button uses that command.

* Problems with the "ctags" program:
  1) `extern "C"' seems to confuse it a lot.
  2) It doesn't know that #defines can't be in a class.
  3) A predeclaration of a class name (e.g., "class MyClass;") generates a
     tag but shouldn't.
  4) Bodies of class declarations aren't scanned but should be (with all
     generated tags being declared as being in that class).
     And could enum values maybe be detected as part of the same fix?
  5) C++ destructors (e.g., ~MyClass(void)) aren't found.

/ Modify "elvis.ps" to sense the paper size automatically.  Replace "lppaper"
  with a list of driver-specific options; for PS, this should include frame,
  bar, punch, clipboard, font, and tray.  (The frame, bar, and punch options
  are already implemented, but you need to edit elvis.ps to change them.)

* Under X11, if you replace the controlfont with another of the same height,
  elvis doesn't redraw the toolbar.  Labels don't fit in the buttons anymore!

* Under X11, if a dialog has a locked field which is wider than a normal string
  field, it is shifted.  This is bad, because locked fields have no cursor so
  there is no way to shift it back again.  Also, the overall dialog width is
  chosen with the assumption that the whole locked field is visible.

* Problem: The "squid" HTTP proxy, when handling an FTP request to read a
  directory, returns an HTML document similar to Elvis' own format, but
  it uses relative file names in a weird way.  I suspect squid's method is
  incompatible with that used in "index.html" files on some FTP sites
  such as ftp://sunsite.unc.edu/pub/Linux/!INDEX.html, so this is probably
  really a squid bug, but I need to check.

/ Some FTP servers make non-existent files look like directories.
  Specifically, the SIZE command returns "550-Argument is not a plain file".
  Need a better way to detect non-existent files.

/ If the cursor is past the end of the buffer, then it looks weird.  This
  could be fixed by adding a "virtual" newline to the end of a file which
  doesn't end with a real one.

* If you use the scrollbar to scroll to the bottom of a large file in html,
  man, or hex mode, then when you reach the very end of the file (where you
  would expect see no part of the file, just ~'s) it jumps back so the window
  is completely filled with lines from the file.

* The "hex" display mode does strange things to 0x0a characters.  Those quirks
  are hard to eliminate, because that's the newline character and several
  commands are sensitive to newlines.  But eventually something should be done.

* In the "syntax" display mode, if a multi-line comment ends on the last
  character of the line just above the top of the window, and the comment
  ender string is one character long (which happens in Pascal, but not C),
  then elvis assumes that the top line of the window is a continuation of
  that comment.

* Ex command-line history is very weird.  If you recall an earlier command
  and edit it, you're actually editing the historic version.  I would prefer
  to have a new copy of that line created at the end of the history buffer,
  and have the editing take place on that new copy.

  The same thing happens with regular-expression history, and filter-program
  history.

* elvis's prompt cannot be answered from a keyboard map, an
  example (simplified and therefore useless):

          :map #1 :!more %^M^M

  the second ^M should answer elvis's prompt, that occurs after
  the shell escape, automatically (only for convenience, because
  `more' has its own prompt), but this doesn't work: I have to
  press <Return> manually - but then elvis recognizes the ^M - and
  moves the cursor one line down (to my surprise: if I map a
  single character to any key I can use this key to answer the
  prompt!)

* In the installation routine, create links named "vi", "ex", and "view".
  Similarly, supply .BAT files or something for Win32.

* Ctags doesn't produce "Mfilename" tags for main() functions in filename.c

* Sometimes text changes as you move the cursor over it in HTML mode.  This
  has also been reported in Hex mode, but I wasn't able to reproduce that, and
  it could easily have been a configuration problem anyway. (:set nonascii=...)

-------------------------------------------------------------------------------
? Under X11, elvis was reported to get stuck in a loop while trying to step
  through the toolbar buttons in the x_tb_predict() function.

    It looks like there was only one button in the toolbar, and the
    buttons are organized in a circular list.  They shouldn't be a
    circular list!  How could that happen?

    This may have some indirect relationship to the way that the default
    geometry is parsed.  The system where this was reported had a
    geometry string that looked like "80x25+130" -- with no "+Y" value.

? Trouble decoding URL-encoded addresses?  Compare elvis' code to that of
  VFY.  Also compare the osnet.c file to the VFY's code.

? Sometimes the X11 interface refuses to give up keyboard focus.  I've seen
  this *VERY RARELY* with FVWM under Linux.  It has also been reported under
  OLVWM, and is supposed to be very common under TWM.  I suspect there is a
  race condition, perhaps because elvis (or something else?) isn't including
  a timestamp on a request?

    Hopefully this was a side-effect of the bug which caused elvis to
    request an invalid window size (width != basewidth + integer * inc).
    That bug caused elvis and the window manager to fight about window
    sizes (and thus burn a lot of CPU cycles!) which may have lead the
    window manager to treat elvis windows differently.

? Does elvis handle ^M in macros differently than the standard vi?

    MS-DOS handles them differently than the standard Unix.  I suspect
    that's the main problem.  Elvis should try to avoid dependencies
    like that, though; perhaps ":source" should always read ex scripts
    in "unix" mode?

? In the Win32/console version, the handling of the "dead" keys (^, ' and `)
  is wrong, they produce three characters, for instance ^<space> inserts #)^

/ the read command omits the message 'read x lines' in some
  circumstances, one case that I can reproduce:

  call elvis for a new (empty) file, then do

      :r !echo "abc\ndef"
  twice

      I couldnt reproduce this here.  Hopefully that means it is fixed,
      but I doubt it.

/ In lib/elvis.bwf, the Win32 copy command doesn't like the ">NUL".

    Unable to verify this bug; it works for me.  I have made a few small
    changes to osprg.c, but as far as I can tell, it should also have
    worked before those changes.  From the description of this bug, it
    sounds like ">NUL" was being passed to COPY.EXE as an argument.

/ A URL with empty parameters ("http://localhost/sample.html?") has been
  reported to cause core dumps.  I was unable to reproduce this myself.
  It is possible that some other change that I made to the code has fixed
  this.

    Could not reproduce this.

/ Elvis got hung on a slow machine running X11.  The cursor continued to blink,
  but elvis didn't respond to any events, including Expose or KeyPress events.

    Unable to reproduce this.

/ These commands, taken from elvis.brf, cause a trap under NT:
  > I suppose you could make "elvis.brf" check the file's write permissions,
  > and check it out if it is read-only.  The following UNTESTED lines should
  > do the job:
  >
  >         if dirperm(filename) == "readonly"
  >         then !co -l %
  >         if dirperm(filename) == "readwrite"
  >         then set noreadonly

    Worked for me.  (Actually, I used "then !attr -r %" since I don't
    have RCS for Win32.)

/ The NT console port throws away any characters that the user types into
  the console window before elvis creates its own console buffer.  Ideally
  it shouldn't do that.

    I don't think it does.  It can be hard to tell, since (on my NT system
    at least) elvis creates its console buffer after only a second or so.
    But it looks like it works, and if the SetConsoleMode() functions do
    clobber the typeahead buffer, I'm not sure how to tell it not to do
    that.

/ A long (>255 chars) error message from a compiler caused elvis to dump core.
  This was while running :make, of course.

    There's no obvious reason for this to happen.  All of the buffers for
    holding parsed parts of the error message are dynamically allocated
    and grow when necessary.  Perhaps the msg() command caused calc.c's
    buffer to overflow?

    This was actually reported for 2.0, not 2.1.  I was able to reproduce
    it under 2.0 but not 2.1, so I'll assume it was fixed somewhere along
    the way.

-------------------------------------------------------------------------------
/ Change version to "2.1i-beta".

/ The "-b blksize" option doesn't verify that the requested size is a power
  of 2.  It must!

/ The command "elvis -Gquit -cq -b16384" causes a core dump.

    Actually, anything that generated a non-fatal, non-status message
    was causing core dumps when elvis was invoked with the "quit" or
    "script" user interfaces.

/ It would be nice if regexp character classes allowed extra characters after
  the :alnum: or whatever.  E.g., /[:alpha:_][:alnum:_]*/ find C identifiers.
  MORE IMPORTANTLY, did I really implement the right syntax?  The POSIX specs
  could be interpretted as meaning that you need an extra set of brackets
  around the named classes.

    Apparently POSIX does require an extra set of brackets, as in
    /[[:alpha:]_][[:alnum:]_]*/ for matching C identifiers.  I've changed
    elvis' regexp code, and the lib/elvisre.html documentation file.

/ The visual / command doesn't allow / in a character class.  In fact, it
  appears that *all* regular expression parsing code has this problem:
  ex addresses, and :g/:v/:s commands.  Tags may also have this problem,
  but since tag addresses never use character classes it is a moot point.

/ Winelvis' "About" box should show the version number of the VERSION macro,
  or the o_version option.  (They're the same.)  In 2.1h and earlier, it is
  just hardcoded to show "2.1".

/ All aliases seem to be enclosed in curlies, even the ones which were really
  defined on a single line.  Since the presence of curlies affects the way
  arguments are handled, this is a problem.  Apparently arguments are appended
  on the *following* line, so they look like a separate command.

/ Added OS/2 port to general distribution.

/ Describe blksize in the manual.  Mention that it can only be set via -b#
  and explain why.

/ The manual entry for MS-DOS (9.1) states incorrectly that you can change
  the blksize option in a configuration file.

/ The manual needs to say more about filenames: wildcards, %, #, \, $env,
  (expr), ~, {a,b}, `command`.

/ Typo in manual: "altername" should be "alternate".

/ In elvisinp.html, ^R/^L get confused about the terms "window" & "screen".

/ In elvis.syn, MS-DOS file names should accept uppercase extensions as well
  as lowercase.  E.g., .BAT and .bat.

/ Make the X11 version check for WINDOWID *or* one of the known TERM values.

/ If $name isn't found, should elvis try for $NAME?  This would be handy under    Win95 because most programs are insensitive to the case of environment
  variables but getenv() is case-sensitive.  It could also be construed to be
  more CSH-like, which is probably a good thing.

/ How about a command-line option for redirecting stderr (and maybe stdout?)
  to a log file?  This would provide an easy-to-document means by which
  Windows95 users can store error messages.

/ The manual entry for "elvis.ses" still tries to use the old "binary" option.

/ If the "backup" option is set, :wq can cause a core dump at guitcap.c:1547

/ Tag searches on operators don't work?  E.g., if you double-click on <<
  in a C++ program, it should search for the tag "operator<<" but instead it
  just fails.

-------------------------------------------------------------------------------
/ Change version number to "2.1h-beta".

/ For MS-DOS, the ^C key doesn't interupt macros, or long commands.  This is
  apparently due to the fact that MS-DOS only checks for ^C during a few
  specific system calls, and elvis doesn't use those calls when it is busy
  executing commands.

    I fixed this by adding a bogus kbhit() call to the ttypoll() function.

/ The :a, :c, and :i commands are dropping the newline from the last line of
  multi-line text.

/ For X11, the XGetInputFocus() function now returns void instead of a success
  indicator.  Need to modify guix11/guix11.c

/ The sigaction() calls should not use sa_restorer, because it is deprecated.

/ The [charlist] notation in regular expressions doesn't allow ] to be included
  in a regexp.  It should be allowed as the first character, or after an initial
  ^ character for negating the list, or after a - as the end of a range.  Also
  need to document this.

/ In osunix/osnet.c, the function netread() tries to return -1 to indicate
  an error.  It should return False.

/ Problems remain with terminals that support dim text but not bright text.
  md/me/mh termcap attributes.

/ The backslash changes for aliases seems to have broken backslash handling
  for commands like...

      g/foo/a\
      one\
      two

  ... which now adds an extra blank line.  Probably the :append command should
  be modified to expect a newline after the last line, and the normal '.'
  marker should ensure that the last added line ends with a newline.

/ In ex mode if you hit <Enter> at the end of a file, elvis should give a
  better error message than "bad delta".

/ In guiwin32/gwmsg.c, the gwclient_WM_DROPFILES() function needs to quote
  the dangerous characters (such as space) in the filename.

/ The X11 interface can get stuck in a loop in which it continually resizes
  the window.  Apparently if there are two or more resize requests in the
  queue, and elvis adjusts itself for the first one, then the others will
  can change it to still another size.  This has been reported under TWM
  after a simple resize action.

    A somewhat sleazy way to reproduce this: Start elvis.  From within
    elvis, start xkoules.  While elvis is waiting for xkoules to exit,
    resize the elvis window *twice*.  Exit xkoules.

    Possible fix: When elvis gets one configure request, it can try to
    read as many more as possible, and only use the last one.

? Problems with German keyboards under X11.  The modifiers don't appear to
  work correctly, which prevents { } [ ] ~ @ \ | from being enterable.
  Reported by Juergen Preuss ([e-mail]).

    I borrowed some code from RXVT, which had borrowed it from XTerm.
    Hopefully this will solve the problem.  It doesn't seem to have
    broken anything for US keyboards, but that's all I can test myself.

/ ctags can't handle derived classes: "class foo: public bar {...}" should
  generate an entry for "foo" and ignore "bar".

/ The "tags" option needs to support a notation for "look in the directory of
  the current file".  Vim uses "./tags" for that.

/ After a <L command, I got the message "15 lines NULLed".

/ Some problems with the manual: The use of multiple -V flags should be
  explained somewhere.  The missing features of the DOS port should be
  writ large.

/ The "version" option isn't being initialized correctly.

/ Add a dummy "redraw" option.  Or maybe a real one.

/ Change the html entry in "elvis.syn" to include more tags.

/ When the "number" option is set, clicking the mouse sets the preferred
  column to the wrong value; If you click and then hit j/k, the cursor
  jumps to a different column.

/ A crash was reported in the TeX mode.  Overflowed a token, perhaps?

    Perhaps; the token-getting function wasn't smart enough to handle
    long comments or expressions.  It should be that smart now.

/ The online manual title says "elvis 2.0" -- change to "2.1".  Also, I saw
  a reference to "14.x" which should actually read "16.x"

/ The "Tips" chapter of the manual should have a section on how to make elvis
  run faster.

-------------------------------------------------------------------------------
/ Change version number to "2.1g-beta".

/ Consistent crashes have been reported under Linux in the lp() function
  when elvis is compiled with gcc -O2, with ELF binaries.  Unfortunately, it
  works correctly when compiled with gcc -g, so we can't narrow it down any
  more than that.  Reported by William Cheung, [e-mail]

/ In text-mode Win32, there are freaky problems reading from a program.
  Just running a program works; filtering works; reading from a text-mode
  win32 executable such as ls.exe works.  Reading from anything else
  generally causes the output to show on your screen but not be inserted
  into the edit buffer.

    Got pipes working under Win95 in text mode, but not graphic mode.
    Would like to have pipes under graphic mode (WinElvis).

/ Add the ability to run an external tag search program.  A new "tagprg"
  options will be invoked, and its stdout will be parsed as a series of
  tag lines, all of which are assumed to match the requested tag.
  If tagprg is undefined, then use the standard built-in tag searches.

/ Both "windows" and "x11" have a scrollbar option, but they work differently.
  Should probably rename the x11 option to "xscrollbar".

/ Temp files tend to accumulate under Win32.  Problem in oswin32/osprg.c?

/ Rapid left-arrow while in input mode causes characters to be lost.  This
  only seems to show up in Win32 text-mode.

    This was fixed once before, I thought, in input.c:cleanup().
    I suspect new ctrl_o code in inputchange() is to blame for its
    reappearance.

/ Using backslashes for multi-line commands doesn't seem to work right.
  It looks like there's no terminating '\0'.

/ In WinElvis, the toolbar and scrollbar options don't quite work -- you can
  apparently turn them off, but you can't turn them back on again.

/ The names of the "windows" font options conflict with the "x11" options.
  This is a problem because both sets of names appear in the elvisopt.html
  file, and :help needs to be able to find the right one.

    Changed the names of the "windows" options from normalfont, boldfont,
    etc., to normalstyle, boldstyle, etc.  This makes more sense anyway,
    since their values aren't really font names.

/ ":w>>file" doesn't set the alternate file name correctly.  Also, the
  >>file notation doesn't seem to *create* a file if it doesn't exist;
  it should.

/ The :set command doesn't permit trailing spaces.  It should.

/ Need to find a way to nest the { ... } notation.

/ The following should be undoable in a single step: @x macros, :map macros,
  :@x macros, :ab abbreviations, aliases, and :g/:v commands (which are already
  undoable as a single step).

    One possible way to do this: make bufwilldo() check the eventcounter,
    and do nothing if it hasn't changed since the previous bufwilldo().

/ Under X11, toolbar labels should accept characters which aren't defined
  as meaning something else -- namely :"=?~ -- and trailing spaces should be
  ignored.

/ Aliases should not affect the behavior of the exthenflag variable.

/ Add :while and :do commands.  The while expression should be saved
  wherever the exthenflag flag is saved.

/ The htmlescape() function in dmmarkup.c is reported to cause core dumps
  on an RS6000 running AIX, when compiled with optimization enabled.  If
  compiled for debugging, it doesn't dump core so specifics are hard to
  come by on this one.

/ Under MS-DOS, elvis seems to write each character twice.  This was reported
  by a blind user who has a text-to-speech device hooked into his console.

    It doesn't do this with the termcap interface under Linux.  I suspect
    that it is having a hard time because the attribute and character
    bytes are written out seperately.

/ Large :copy commands can get screwed up.  Try editing a large file, such
  as ex.c, and insert line numbers at the start of each line ("%!cat -n").
  Then do a ":100,500co1000" command.

/ Sometimes keywords from elvis.syn are lost?  Saw some PERL code where
  "if" and "while" weren't in boldface.  The large number of PERL keywords
  may be a factor.

    Yes, that was it.  There was a silent limit of 256 keywords.  I have
    now increased that limit to 300 keywords.  I really need to recode
    the keyword hash table though -- it is unnecessarily kludgy.

/ X11 dialogs should support a "locked" data type, because some options can't
  be edited.  If a row doesn't specify an option name, then that row's type
  should be "locked"; this will allow dialogs to display immutable information
  which doesn't come from an option.  If a row has neither an option nor a
  value, then its label should be centered in the dialog.

/ Under MSDOS with TERM=dosansi or TERM=nansi, mouse highlighting is
  flakey, and linewrap causes some problems.  Also, running an external
  command causes the mouse cursor to vanish.

/ Should allow certain words to be customized by language:  Submit, Cancel,
  Help, True, False.  Create options named "true" and "false" for storing
  the local versions of those words; this will also have the pleasant side
  effect of allowing the english words "true" and "false" to be used as
  though they were Boolean literals in expressions.  The other words are
  only used by the "x11" user interface.

/ The saveregexp option doesn't do enough.  It protects the regexp, but
  it needs to protect the substitution text and the substitution flags too.

/ Need to replace <xmp></xmp> with <pre></pre> in the lib/*.html files.
  <xmp> is obsolete, and elvis doesn't implement it correctly anyway.
  Also add <code> to dmmarkup.c, as a synonym for <tt>, and use <code>
  in the lib/*.html files.

/ The :w command affects the value of the "exthenflag" flag.  The elvis.bwf
  and elvis.awf scripts (and others) should probably save the value of
  exthenflag, and restore it afterward.

/ In a :ab! abbreviation, expansion doesn't take place if the short form
  is the last thing on a command line.

/ Under Win32, it should be able to parse UNC names (\\machine\directory\file)

    Apparently the only missing piece is that elvis should not change
    the leading \\ to \.

/ The termcap interface can't use "dim" unless there is also a "bold" escape
  sequence.  There's no need for that!

/ The :ta command doesn't offer any way to supply values which contain
  colons or whitespace.

/ Long moves don't necessarily center the cursor line in the window.  See
  the "!!!" comment in dmnormal.c.  To fix this, I'll need to pass the
  window pointer into the setup function.  (Some other setup functions could
  benefit from this, too.)

/ The :ta command should be smarter about setting the previoustag option.
  Whenever you jump to a tag, previoustag should be set to the name of that
  tag.

/ Add a Windows version of the "quit" GUI to WinElvis.  The normal "quit"
  interface isn't supported there, and the text-mode version of elvis doesn't
  support "lptype=windows".

/ When ^L is defined as forcing a redraw in input mode, it can break some
  common macros.  Perhaps it should be left as a normal "insert me" character.
  Users could still redraw the screen while in input mode by typing ^O^L.

/ Bottom button of X11 scrollbar isn't drawn initially.  It appears after
  an expose event, or after any colon command including colon-backspace.
  I suspect the scrollbar is first drawn with a height of 0, and the top
  button is drawn over the top of the bottom button.

/ On AIX systems, the toolbar is drawn incorrectly.

/ Control-V should display a carat on the screen, while waiting for the
  quoted character.

--------------------------------------------------------------------------------
/ Change version number in version.h to "2.1f-alpha".  (n.b. I haven't been
  doing a consistent job of updating this list as bugs are fixed, or as new
  versions are uploaded.  There *were* some fixes between 2.1b and 2.1f.)

/ An expose event causes the scrollbar to be redrawn, even if it should really
  be blanked out (because the cursor is on the bottom row).

/ Under Win32 or DOS, text-mode "elvis -?" outputs a line of garbage after
  the help message.  The garbage consists of uninterpretted escape sequences,
  emitted as the termcap interface shuts itself down.

/ When invoked in the current directory, with the "exrc" option set, elvis
  will source the ./.exrc file twice.

/ Abbreviations can get stuck in a loop.  Self-referential abbreviations
  should be disallowed.  (Actually, should an abbreviation *ever* expand
  another embedded abbreviation?)

/ The :& command doesn't seem to be working correctly in all circumstances.
  After doing a ":%s/.$" on one buffer, and then switching to another buffer,
  the command ":%&" did not work.

    This is because the "elvis.arf" file uses a :s command to locate and
    execute modelines.  There is no easy way to fix this, unless you think
    ":set nomodelines" is an acceptable solution.

    Here's an idea: Create a new "saveregexp" option.  Leave it on
    normally, but turn it off when doing the modeline search.  Also force
    it off while searching for tags for the "showtag" option.

/ The command ":g/old/s//new/g" doesn't work in 2.1e.  It worked in 2.0, so
  some recent change must have broken it.

/ When returning after a :bb command, the cursor is moved to the top of the
  buffer instead of its previous position.  The best way to fix this would
  be to add a "winpos" field to the BUFFER struct, and set it to the cursor
  position before each ^W command or ex command.

/ The X11 interface doesn't handle Expose events during time-consuming
  commands such as :make.

/ If bufdisplay is set to "man", "html", or "tex", then ^Wd should probably
  toggle between "syntax" and bufdisplay, rather than "normal" and bufdisplay,
  so the source text will be displayed with syntax coloring.

/ In ioopen(), the usestdio flag is never turned off.

/ The :make command is insensitive to the gui->poll function.

    It does set the pollfrequency option to 1 while it is running, but
    I think maybe it also needs to call the guipoll() function to reset
    its internal count-down.

    Update: It does detect it, but not until the next line comes in from
    the compiler.  This can take a while.

/ The :g and :v commands don't work correctly on last line.  If the last line
  is supposed to be affected, it appears to actually affect the *FIRST* line.

    The behavior of this bug is more complex than that, actually.  More
    testing is needed.

/ The :s command doesn't handle counts correctly.  It should treat a count as
  a quantity of lines, but instead it treats it as a selector for the instance
  of matching text to change in each line.

/ In the "x11" interface, the tool buttons should wait until the mouse button
  is released.  Otherwise the "quit" button could leave the mouse pointing at
  some totally unrelated window WITH A BUTTON PRESSED!  Surprise!

/ If the "number" option is set, then the printout should show line numbers.
  It should also cause page headers to be printed.

/ Often it would be handy to allow ":e !command" or ":sp !command" to fill an
  untitled buffer from a filter command.  For example, ":sp !man \@ | col -b"
  would cause a window to pop-up, showing the man-page for the word under the
  cursor.

/ In the syntax display mode, the directory in which the current file resides
  should assumed to be in the search path.  For example, if you're editing
  osunix/tcaphelp.c, and you double-click on "tcaposix.h", then elvis should
  be smart enough to load osunix/tcaposix.h.  Currently it isn't.

/ The assert() statement in ex_qall() needs "|| xinf->command==EX_ONLY".

/ If the backspace key sends ^? (and the tty settings indicate this) then
  elvis should treat ^? as ^H.

/ It would be nice if elvis supported HTML frames, in the least-complex
  manner possible: Treat <FRAME SRC=... ALT=...> like <A HREF=...>...</A>

  Similarly, it would be nice if an <IMG> which isn't an hypertext reference
  would fetch the image data (perhaps reading it into a temporary edit buffer)
  and then send it to a configurable program; the default should be "xv -"
  so the image is displayed by xv.

/ If ":make" is used in a window which is displaying an internal buffer
  (especially the "Elvis error list" buffer), and there is another window
  which is displaying a non-internal buffer, then the command should be run
  in that other window.  This is simply for convenience -- after going to the
  trouble of displaying an internal buffer in a window, the user probably
  doesn't want to use that same window to show a buggy source file.

/ Some older X-windows servers don't have XrmCombineFileDatabase().

/ If you are displaying a cut buffer in a window, then you can try to yank
  text from a cut buffer into itself... but cutyank() isn't smart enough to
  handle that correctly; elvis hangs.  Probably cut buffers should be locked.

/ The :pop command should wipe out the list of matching tags.  This will
  allow elvis's enhanced tags to act more like traditional tags for users
  who are in the habit of using ^]^T to glance at the definition of a tag.

/ The X11 resource "*foreground: somecolor" makes toolbar buttons unreadable.

/ The install procedure should copy lib/elvis.xpm and lib/elvis.xbm into the
  /usr/include/X11/{pix,bit}map directories.  Also, small icons would be nice
  (pixmaps of size next/normal=56x46, small=21x18, and mini=17x14)

/ The "Delete Window" button of the last elvis window should not simply fail;
  it should do a ":close" command instead.

/ In the "html" display mode, if a token begins with   then elvis
  mistakes the token for a whitespace token; the remaining characters aren't
  displayed.

/ The x, X, ~, and r commands don't work correctly when the count exceeds
  the number of characters in the line.  They should implicitly reduce the
  count; but instead they're just ignoring the command.

    Actually, r and X are supposed to fail, and they do.
    x should fail, but elvis just deleted to left.  Now it fails correctly.
    ~ should reduce the count.  It didn't before, but it does now.

/ The character search commands stop at a newline, even if the current display
  mode uses some other definition of a "line".

/ The X11 interface should handle GraphicsExpose events when scrolling a
  window which is partially obscured.

/ Add support for hardtabs/ht option.  And then ignore it, since elvis never
  outputs hard tabs.

/ The "flash" option doesn't work.

/ The current("word") function sometimes causes a core dump.  This occurs when
  the cursor isn't located on a word (but sometimes that works too).

/ Elvis uses an option named "buffers", but that clashes with a variable
  declared in some systems' <stdio.h> file.  Change to "elvis_buffers".

/ Elvis checks for :mh=: even if :me=: is undefined.  It shouldn't.

/ Under NT, in the GUI version, the Options/Syntax menu item brings up
  Dr. Watson.

/ Cut/Paste don't work in Win32/gui?

    Verified.  A true bug.  It attempts to cut/paste whole lines (?) via
    the ">/"< buffer, but the underlying GUI clipboard functions don't
    seem to work.

/ Win32: There is a limit to how small a window can be: 2 rows of 30 characters.
  Resizing the screen to be smaller than that (by dragging an edge/corner of
  the window frame) causes elvis to crash.

/ Win32: If you use a mouse click to position the cursor, and you jiggle
  the mouse slightly, then WinElvis starts highlighting text.  Xelvis is less
  sensitive to jiggles; it requires you to drag the mouse into a different
  character cell before it'll start highlighting, which seems to work well.

--------------------------------------------------------------------------------
/ Change version number in version.h to "2.1b-alpha"

/ The X icons still say "2.0", even though we're working on 2.1 now.

/ Some commands move the cursor to a different buffer when they shouldn't.
  For example, ":(otherbuf)1,20co." copies the lines correctly, but then it
  moves the cursor to line 20 of otherbuf.

/ The "x11" interface should ignore MotionNotify except after a ButtonPress.
  Although elvis only requests MotionNotify events while a button is held down,
  some servers send them at other times as well.

/ Ex parser doesn't support backslashes, for things like...

    :g/XXX/a\
    line1\
    line2

/ Under X, pasting from the clipboard doesn't work during, e.g., a "5s"
  command.  (Probably true of Win32 gui, too.)

/ Environment variables with underscores are not recognized in elvis scripts.

/ A more sophisticated method for handling EOF would be appreciated.
  Suggestion:
      * Add a "readeol" option to each buffer, one of {unix, dos, mac, text,
        binary}, defaults to "text".
      * Add a "writeeol" option to each buffer, one of {unix, dos, mac, text,
        binary, same}, defaults to "same".
      * Add a "fileeol(filename)" function, returns the file's probable
        format, one of {unix, dos, mac, text, binary} by reading the first
        part of a file.  Returns "text" if no hard evidence of any other
        format.
    * Meanings of values: unix=LF, dos=CRLF, mac=CR, text=local convention,
      binary=no modifications, same=current value of readeol.
    * The "elvis.brf" file could do a "let readeol=fileeol(filename)"
      to detect non-text files automatically.

/ Need a way to prevent :make from loading an erroneous BINARY file.

/ The built-in calculator can easily overflow when fetching the value of an
  environment variable.  Possibly other situations, too.

/ Under Unixware, when the termcap interface writes a '\n' character, the
  cursor is forced to column 0.  Should use :do=: or :DO=: instead.

/ ":q!" doesn't turn off a buffer's "modified" flag or delete the buffer.

/ The guitcap file never checks the LINES and COLUMNS environment variables.

/ When entering a command line, ^Ocw places a $ as though it will replace
  the current word, but actually any text is inserted.

/ The "program" option isn't documented.

/ Under X11, some sites are missing the default fonts, or have goofy fonts
  aliased to "fixed" and "variable".  Probably appdefaults fix this for other
  applications.  Elvis should support some method of reading X resources.

/ It would be nice if untar.c kept the date and permissions from the archive.
  This can't be done in ANSI C, but if POSIX calls are supported then it can
  be.  Maybe check for _POSIX_SOURCE?  Also, it would be nice if the untar
  executable could be prepended to the archive, to produce a self-extracting
  archive.

/ Add a way to abbreviate (Elvis error list) and (Elvis map log).

    Done: You can now give use a quote character followed by the initials
    of any buffer in place of the buffer name.  So ("Eel) means the same
    as (Elvis error list).

/ The "warningbells" and "warpback" options are both abbreviated "wb".  Change
  "warpback" to "xwb".

/ The termcap gui should not attempt to catch SIGHUP.  It isn't necessary,
  and catching it can, in fact, cause the elvis process to go into a tight
  loop as it attempts to read from a dead tty.

/ Add ^Wo and :only commands -- like :qall except they only close (don't
  delete buffers) and don't affect the current window.

/ Maybe add a "showname" option, to display the buffer name on the status
  line?

/ The console's "Edit/Paste" menu item doesn't paste all of the characters
  that it should.  (Also, it doesn't work correctly for Chinese text... but
  that may be too big a problem for now.)

/ The manual should mention that "ignorecase" doesn't affect character classes.

/ The status line isn't redrawn when colors change.

/ The "instman.sh" script doesn't invoke elvis correctly to format the
  man-pages.  It uses "-gquit" instead of "-Gquit", and the EXINIT variable
  as no effect because ELVISPATH=dummy.

/ there seems to be a `!' substitution in command lines, this may
  be a "generalisation" of vi's command repetition `!!' (which
  allows command repetition and appending to a command, i.e.
  `!' substitution occurs only at the beginning of a command);
  IMHO the `original' behaviour (which was also implemented in
  elvis18) is better, because a substitution inside of a command
  is rarely needed, but now you have to quote all exclamation marks

/ in vi the `print' and `list' commands query the `number' flag
  (and behave according to it, i.e. print line numbers or not)

/ the join command could display a message like 'x lines joined'

/ In the termcap interface, if you hit <Esc> in command mode, the following
  character is eaten.

    This is normal.  Hitting <Esc> causes the terminal to be "alerted,"
    which has the side effect of throwing away any characters already in
    the type-ahead queue.

/ The defaultreadonly option is ignored.  The bufload() function resets each
  buffer's readonly flag before loading the file.

/ if I call the (ex) join command with an adress range of 1 line
  (not a single adress, sounds strange, but see below ...) it
  joins 2 lines, but -per definition- it should do nothing

  e.g.:           :3,3j             joins lines 3 and 4!

  why bother? ... try the following command:

                :v/./,/./-j

  in vi it compresses consecutive blank lines to one, but,
  due to the behavior of `join', elvis deletes all blank lines

/ the following I've already reported, but now I can describe the
  error more precisely, also I can give you a hint, how to fix
  it:

      elvis deletes one `level' of backslashes before passing a
      command to the shell (concerns ex's `!', read and write
      commands, NOT vi's filter command !)

  I think it's easy to fix, but you should prove it; in `ex.c'
  change line 1228:

        if (*refp && !CHARchr(toCHAR("%#!@\\"), **refp))
                          ^^
                          delete this

/ In the relative part of a line address, the "+" is optional.  I.e., the
  command ":.,.3y" should be identical to ":.,.+3y".

/ The command ":g/^/m0" should reverse the order of all lines, but elvis
  just gives the message "destination can't be inside source".

/ WinElvis doesn't print graphic characters correctly.

/ Add support for multi-line strings which don't require a backslash.

/ Typing some text and then (without hitting Esc) allowing the left arrow to
  autorepeat causes some characters to be lost.  This only happens under the
  X11 user interface.

/ The description of the "keywordprg" option says that $1 is replaced by the
  current word, but currently the word is appended to the option's value.

--------------------------------------------------------------------------------
/ Change version number in version.h to "2.1a-alpha"

/ Changing one option via the Options menu causes all options in that dialog
  to be set, so ":set" with no arguments causes them to be displayed.

  FIXED: This is actually due to a quirk in elvis' options.c file.  I've fixed
  it.  Previously elvis didn't store the default value for each option; now
  it'll save the values of all options immediately after executing "elvis.ini"
  as the defaults.  Each option's OPT_SET flag will be set or cleared after
  a value is assigned to the option, depending on whether the new value
  matches the default value.

/ Bug in filename completion under all Microsoft OSes (not just in Win32 gui):
  If you enter a directory name and partial file name using forward slashes,
  the substituted name will drop the directory name.  If I use a backslash,
  it works correctly.  For example, "guiwin32/gu<Tab>" becomes "guiwin.c" but
  "guiwin32\gu<Tab>" becomes "guiwin32\guiwin.c".

  FIXED, at least for Win32.  It is due to minor bugs in the osdir.c functions.

/ In the X11 user interface, I've received some requests for more configuration
  options for the cursor.  Some folks don't like the hollow rectangle.  Add a
  new "cursorstyle" option which can be set to hollow, opaque, or xor.

/ In addition to "xterm", perhaps guix11.test() should check for "iris-ansi"
  (for SGI) and "dtterm" (for CDE).

/ In text mode, the cursor jumps to lower-right corner after each keystroke.
  Annoying on slow terminals.  This appears to be an attempt to clear the line,
  by drawing a bunch of spaces before the statusline text (showmode, ruler).

/ Errors that occur interactively should not set exitcode to 1.

/ eventreplace() doesn't move the cursor to the correct location.

/ Add WinElvis and WinTags to the "makwin32.bat" file.

/ In WinElvis, if the text doesn't completely fill the window, then the
  scrollbar will look funny.  This is unavoidable.  We would like to have
  the scrollbar's thumb completely fill the scrollbar (since the text
  completely fills the window) but Win32 tries to do us a favor by removing
  the scrollbar in that situation.

/ :only can cause a core dump.  Try starting elvis without a file, running :sp
  and then :on

/ :close can close the last window.  It shouldn't.

    Actually, that's a pretty convenient behavior to support.  The window's
    "close" button does a :close, which should be able to close the last
    window... if there aren't any modified buffers.

/ Under Win32, if you move the pointer outside of the text window, it isn't
  restored to its intended shape when you move it back in again.

/ Under X11, the resource for button colors must not be named "foreground".
  That leads to a foreground-on-foreground problem, so button labels aren't
  legible.

/ Under X11, -client isn't sensitive to which machine the command is running
  on.  Perhaps the "ELVIS_SERVER" attribute name should be renamed to
  "ELVIS_ON_{hostname}".  Another problem is that user ids may differ --
  running in root and as a real user, for example.

/ Backslash needs to be treated literally in more situations, especially
  under Win32.  A name like "C:\tmp\_tempfile" should be treated literally.

  The worst case is something like "C:\temp\$wc\foo", where the "$wc" is
  supposed to be literal.  I don't think anything can make this be rational!
  The current code interprets the backslash as a quote for the $ character,
  so elvis loads "C:\temp$wc\foo".  If backslash was literal, it would try
  to perform environment variable substitution for "$wc".  The current
  behavior is about as good as it gets.

/ Does the ":lpr" command support ">>" to append to files?  It should.

/ If the buffer doesn't end with a newline, then the Y command will go into
  an endless loop.

/ Newly created windows should acquire input focus... and when entered
  interactively they do.  But for macros and the "elvis -client -c ..."
  command, this doesn't work correctly.

/ The :s command (not :s/old/new/) is supposed to repeat the previous
  :s/old/new/ command, but doesn't.  The :& command works correctly.

/ Under Win32, the italic font is shifted slightly to the left -- which is
  fine normally, but not after a ":set italicfont=n".  Then it clips off
  the uprights of many letters.

/ Skip the "Hit <Enter> to continue" prompt if we're in the middle of a map.

/ Tweak the tags stuff to use the proposed standard.
    - Allow :" after the address in field 3
    - Interpret file: as file:(tagfile) -- i.e., file: marks static tags
    - In the extra fields, the values are subjected to the following
      translations: = becomes =3D, tab becomes =09, newline becomes =0A

/ Writing via ftp doesn't turn off the "writing" flag when complete.  This
  causes the next read to fail in an assert() statement.