SHORT-TERM TO-DO ITEMS:

Known Problems:

* The code departs from perfect 8-bit cleanness in one respect; you cannot
  specify a character \200 as part of a capability string, because the terminfo
  library interprets \200 as a request to embed NUL (\000) at that point (the
  actual code that does this is in ncurses/lib_options.c:add_keytry().
  This is a legacy terminfo property we can't mess with.

* There is no actual support for terminals with the xmc (magic-cookie) glitch.
  (They'll work fine, unless you try to draw highlights on them and don't
  leave the required A_NORMAL spaces on either side in the screen layout.)
  System V Release 4 never handled these properly either, though it claimed to.
  There is a comment in ncurses_lib_doupdate.c:doupdate() describing how this
  could be done right, if anyone wants to try.

* Recognition of xterm mouse reports is implemented by setting the kmous 
  capability to the mouse-report prefix.  As a result, the kmous sequence is
  eaten even when mouse recognition is disabled.  This could be fixed if 
  necessary with special code that pokes the kmous value back into the
  input fifo when KEY_MOUSE is seen and the mousemask is zero, but that would
  be kind of ugly.

* The window classes defined in the c++ subdirectory need documentation.
  Some C++ programmer could earn a lot of good karma by doing this...

* The wresize() code has been reported to core-dump occasionally.  The
  maintainers haven't been able to reproduce this.

* Mark McKenzie <mmckenzi@undergrad.math.uwaterloo.ca> writes:
 
        I have noticed a difference in the way the ncurses1.9.8a and System V
        curses implementations of getch handles signals.

        1) When in delay mode, getch does not return if a signal is received.
        (With System V curses it does).

        2) When in timeout mode, getch will delay until a character is received
        if it is interrupted by a signal [and then returns ERR].
        (With System V curses it exits when the time limit is up even if
        interrupted by a signal).
 
  These things are basically due to the difference between Systen V and
  BSD-style signals.

* Mark McKenzie also says:

        With SVR4 Curses you can use overwrite to make copy from one pad to
        another.  It seems ncurses' overwrite does nothing when passed the same
        parameters.

        You have to call copywin yourself to copy pads with ncurses.  I didn't
        follow overwrite along to see what parameters were eventually being
        passed to copywin.

  We're working on this.

* On bce terminals (including the Linux console, the Xfree86 3.1.2D xterm,
  and rxvt), ncurses does more work than it needs to when erasing colored
  backgrounds (it blank-fills after issuing clr_eol/clr_eos).

Portability (or lack thereof):

* There have been reports of mouse clicks getting lost in the lib_mouse.c code
  under SunOS and IRIX (no evidence this on Linux, however).

* Users of older System V UNIXes (but not Solaris, and probably not SVr4)
  may trip over a known problem with the signal-handling code which causes
  abrupt termination of ncurses applications following resume from a ^Z
  suspend (this problem was first seen running lynx).  You will not see 
  this problem if you are running Linux or one of the 4.4BSD derivatives
  like FreeBSD, NetBSD, or BSDI.  For details, see the analysis in the
  header comment of ncurses/lib_tstp.c.

* In theory, vwprintw and vwscanf are supposed to use the older varargs.h
  interface for handling variadic argument lists.  Linux doesn't have
  varargs.h, it has the newer X/Open-standard stdargs.h equivalent. So
  these functions use stdargs instead.  This is unlikely to be a problem 
  unless you're building ncurses on a System V old enough to only have
  varargs.h.

* If you're using a BSD earlier than 4.4BSD, or a Linux old enough not to
  have a native vsscanf(3) in its library, vwscanw() will not work.  You lose.
  (It should work on any System V, however).  If you want to fix this, add
  an implementation to ncurses/vsscanf.c.

* The demo build for the c++ library craps out with many link errors under gcc
  2.6.3.  We're told the C++ support in 2.6.3 is broken and that the right
  fix is to upgrade to 2.7.0.

* Under Ultrix, configure craps out (Ultrix sh is lame).  Run it under ksh.

Untested features:

* The code for the HP color model using set_color_pair is untested.

* The code for handling soft labels on a terminal type with built-in support
  for them (num_labels > 0, label_height, label_width, label_format, label_off,
  label_on, plab_norm, lab_f*) has not been tested.  The label_format and
  lab_f* capabilities aren't presently used.

REQUESTED FEATURES:

Larry Schwimmer <rosebud@cyclone.Stanford.EDU> writes:
>       On a somewhat related topic, I have a request if this has not
> already been implemented.  I install software for the campus, but
> would like local sysadmins which access the campus software tree to be
> able to have local control over public software packages.  I would
> also like binaries built with ncurses to not have dependencies to the
> campus software tree where possible.  What this would entail is
> providing a multipart path for TERMINFO.  For example, the path
>
> /usr/local/lib/terminfo:/usr/pubsw/lib/terminfo:usr/share/lib/terminfo
>
> could correspond to "first look in the local terminfo database, then
> look in the campus database, then look in the vendor database".
>
> It would also be nice if tic had a command line flag to specify the output
> directory.  I know TERMINFO is honored, but command line (-d dir?
> -o dir?) would also be nice and intuitive.

LONGER-TERM TO-DO ITEMS:

1. Extended mouse support via gpm integration

We have mouse support under xterm now, implemented and documented.  The
next step is to add code to lib_mouse.c that can detect Alessandro Rubini's
GPM server and use it to implement the same interface.

2. Extended COSE conformance

There is an XPG4 standard recently released which describes a superset
of the SVr4 API.  The library is BASE conformant with this standard.
We would like to make ncurses fully conformant at the EXTENDED level
supporting internationalization.

Here are page references to all material involving wide or multi-byte
characters in Issue 4 of the XSI Curses standard, with notes on their
status in this implementation:

	Page 1 (1.1.2) New Features discussion of internationalization.
	Page 12 (2.4): Definition of cchar_t, wchar_t.
	Page 16 (3.3.2): Introduction of multi-column characts.
	Page 17-18 (3.3.5): Description of non-spacing characters.
	Page 19-21 (3.4.2): Basic character operations.
	Page 34 (addnstr): These should now call underlying wide-
character functions, and do (through waddnstr) if _XOPEN_SOURCE_EXTENDED is on).
	Page 35 (addnwstr): wide-character add-string functions.  All macros 
except waddnwstr() which is not yet defined.
	Page 36 (add_wch): wide-character add-char functions.  All macros 
except wadd_wch() which is not yet defined.
	Page 39 (attr_get): implemented -- we've just made the current-
attributes field of the window an attr_t.
	Page 43 (bkgrnd):  None of these are implemented.
	Page 45 (border_set): Neither of these is implemented.
	Page 47 (box_set): box_set implemented as macro, but the underlying
wborder_set() is not yet defined.
	Page 78 (echo_wchar): echo_wchar() implemented as macro, underlying
wecho_wchar() not yet implemented.
	Page 81 (erasechar): Neither entry point is implemented.
	Page 87 (getbkgrnd): Not implemented.
	Page 88 (getcchar): Not implemented.
	Page 93 (getn_wstr): All implemented (as macros) except the
underlying wgetn_wstr().
	Page 97 (get_wch):  All implemented (as macros) except the
underlying wget_wch().
	Page 99 (get_wstr): Xref to page 93.
	Page 105 (hline_set): All implemented (as macros) except the
underlying whline_set(), wvline_set().
	Page 114 (innstr): Multi-byte character-completeness check is
not implemented.
	Page 115 (innwstr): All implemented (as macros) except the
underlying winnw_str().
	Page 119 (insnstr): Implementation may not be correct for multi-byte
characters.
	Page 120 (ins_nwstr): Not implemented.
	Page 121 (insstr): Xref to page 119.
	Page 122 (instr): Xref to page 119.
	Page 123 (ins_wch): Not implemented.
	Page 124 (ins_wstr): Xref to page 120.
	Page 126 (in_wch): Not implemented.
	Page 127 (in_wchnstr): Not implemented.
	Page 128 (inwstr): Xref to page 115.
	Page 133 (killwchar): killwchar not implemented.
	Page 158 (pechochar): pecho_wchar() not implemented.
	Page 176 (setcchar): Not implemented.
	Page 181 (slk_attroff): slk_wset not implemented.
	Page 200 (ungetch): unget_wch() not implemented.
	Page 203 (vidattr): vid_attr() and vid_puts() not implemented.
	Page 206 (vline_set): Xref to page 105.
	Page 214 (wunctrl): Not implemented.
	Page 216 (curses.h): cchar_t, wint_t, wchar_t references.
	Page 220 (curses.h): KEY_CODE_YES

Basically, the macro superstructure is there but the core is absent.  We
need better multi-locale support guarantees from the OS to finish this.
If you are working on internationalization support, please contact us so
we can cooperate.

3. DOS port

Only 16 of the 55 files in the library depend on the terminfo format.
It should be possible to further kernelize the package, then rewrite 
a small number of core files to produce a functionally-compatible
port that would do updates to a memory-mapped screen area.  The first
result of this would be a DOS port.

4. X port

It would be nice if ncurses could recognize when it was running under X and
maintain its own window.  With this feature, all ncurses programs would
automatically become X programs.  The challenge is to handle resize events
properly.

5. Unused capabilities

The currently unused capabilities fall naturally into several groups:

A. Status-line capabilities:

	Booleans: has_status_line, status_line_esc_ok.
	Numerics: width_status_line.
	Strings: dis_status_line, from_status_line, to_status_line.

System V Release 1 curses made no use of these at all.  SVr4's use, if
any, is unknown.  From the AT&T termcap file it looks like curses, in general,
shouldn't use them; terminal variants with status lines have their line count
decremented by 1, suggesting that curses is supposed to leave the status line
alone.

B. Printer capabilities:

	Boolean: col_addr_glitch, cr_cancels_micro_mode, has_print_wheel,
		row_addr_glitch, semi_auto_right_margin, cpi_changes_res,
		lpi_changes_res.
	Numeric: buffer_capacity, dot_horz_spacing, dot_vert_spacing,
		max_micro_address, max_micro_jump, micro_col_size,
		micro_line_size, number_of_pins, output_res_char,
		output_res_line, output_res_horz_inch, print_rate,
		wide_char_size, bit_image_entwining, bit_image_type.
	String: down_half_line, form_feed, up_half_line, set_left_margin,
		set_right_margin, clear_margins, change_char_pitch
		... set_page_length (all the SVr4 printer caps),

Curses doesn't use these.

C. Printer-control capabilities:

	Boolean: prtr_silent.
	Strings: print_screen, prtr_on, prtr_off, prtr_non.

Curses doesn't use these.

D. Dialer strings:

	Strings: hangup, dial_phone, quick_dial, tone, pulse, flash_hook,
		fixed_pause, wait_tone.

Curses doesn't use these.

E. Window and virtual-terminal capabilities:

	Numerics: maximum_windows, virtual_terminal.
	Strings: req_for_input, create_window, goto_window, set_window.

These seem to be fossils from some AT&T experiments on character-based
window systems that never escaped the lab.  The virtual_terminal cap had
something to do with building terminal emulations into tty line disciplines.

F. Unused VDT capabilities:

	Booleans: erase_overstrike, has_meta_key, insert_null_glitch,
		move_insert, dest_tabs_magic_smso, transparent_underline,
		needs_xon_xoff, hard_cursor.
	Numerics: lines_of_memory, buttons.
	Strings: pkey_key, pkey_local, pkey_xmit, underline_char, 
		enter_xon_mode,	exit_xon_mode, xon_character, xoff_character, 
		display_clock, remove_clock, user[0-5], display_pc_char,
		enter_scancode_mode, exit_scancode_mode, pc_term_options, 
		scancode_escape, alt_scancode_esc.

These are the potentially important ones for ncurses.  Notes:

	i) ncurses doesn't need move_insert; it never uses cup/hpa/vpa while
		insert_mode is on.

	ii) We probably don't care about dest_tabs_magic_smso; only
		Telerays used it and they're all long obsolete.


