Rudimentary implementation of the .it request (input line trap).

[mandoc.git] / TODO

************************************************************************
* Official mandoc TODO.
* $Id: TODO,v 1.152 2013/07/13 12:52:07 schwarze Exp $
************************************************************************

************************************************************************
* crashes
************************************************************************

None known right now.

************************************************************************
* missing features
************************************************************************

--- missing roff features ----------------------------------------------

- roff.c should treat \n(.H>23 and \n(.V>19 in the pod2man(1)
  preamble as true, see for example AUTHORS in MooseX::Getopt.3p
  reported by Andreas Voegele <mail at andreasvoegele dot com>
  Tue, 22 Nov 2011 15:34:47 +0100 on ports@

- .ad (adjust margins)
  .ad l -- adjust left margin only (flush left)
  .ad r -- adjust right margin only (flush right)
  .ad c -- center text on line
  .ad b -- adjust both margins (alias: .ad n)
  .na   -- temporarily disable adjustment without changing the mode
  .ad   -- re-enable adjustment without changing the mode
  Adjustment mode is ignored while in no-fill mode (.nf).

- .ce (center N lines)
  found by naddy@ in xloadimage(1)
  found by Juan Francisco Cantero Hurtado <iam at juanfra dot info>
           in lang/racket(1)  Thu, 20 Jun 2013 03:19:11 +0200

- .fc (field control)
  found by naddy@ in xloadimage(1)
  
- .ns (no-space mode) occurs in xine-config(1)
  reported by brad@  Sat, 15 Jan 2011 15:45:23 -0500

- .ta (tab settings) occurs in ircbug(1) and probably gnats(1)
  reported by brad@  Sat, 15 Jan 2011 15:50:51 -0500

- .ti (temporary indent)
  found by naddy@ in xloadimage(1)
  found by bentley@ in nmh(1)  Mon, 23 Apr 2012 13:38:28 -0600

- \c (interrupted text) should prevent the line break
  even inside .Bd literal; that occurs in chat(8)

- using undefined strings or macros defines them to be empty
  wl@  Mon, 14 Nov 2011 14:37:01 +0000

--- missing mdoc features ----------------------------------------------

- fix bad block nesting involving multiple identical explicit blocks
  see the OpenBSD mdoc_macro.c 1.47 commit message

- .Bl -column .Xo support is missing
  ultimate goal:
  restore .Xr and .Dv to
  lib/libc/compat-43/sigvec.3
  lib/libc/gen/signal.3
  lib/libc/sys/sigaction.2

- edge case: decide how to deal with blk_full bad nesting, e.g.
  .Sh .Nm .Bk .Nm .Ek .Sh found by jmc@ in ssh-keygen(1)
  from jmc@  Wed, 14 Jul 2010 18:10:32 +0100

- \\ is now implemented correctly
  * when defining strings and macros using .ds and .de
  * when parsing roff(7) and man(7) macro arguments
  It does not yet work in mdoc(7) macro arguments
  because libmdoc does not yet use mandoc_getarg().
  Also check what happens in plain text, it must be identical to \e.

- .Bd -filled should not be the same as .Bd -ragged, but align both
  the left and right margin.  In groff, it is implemented in terms
  of .ad b, which we don't have either.  Found in cksum(1).

- implement blank `Bl -column', such as
  .Bl -column
  .It foo Ta bar
  .El

- explicitly disallow nested `Bl -column', which would clobber internal
  flags defined for struct mdoc_macro

- In .Bl -column .It, the end of the line probably has to be regarded
  as an implicit .Ta, if there could be one, see the following mildly
  ugly code from login.conf(5):
    .Bl -column minpasswordlen program xetcxmotd
    .It path Ta path Ta value of Dv _PATH_DEFPATH
    .br
    Default search path.
  reported by Michal Mazurek <akfaew at jasminek dot net>
  via jmc@ Thu, 7 Apr 2011 16:00:53 +0059

- inside `.Bl -column' phrases, punctuation is handled like normal
  text, e.g. `.Bl -column .It Fl x . Ta ...' should give "-x -."

- inside `.Bl -column' phrases, TERMP_IGNDELIM handling by `Pf'
  is not safe, e.g. `.Bl -column .It Pf a b .' gives "ab."
  but should give "ab ."

- set a meaningful default if no `Bl' list type is assigned

- have a blank `It' head for `Bl -tag' not puke

- prohibit `Nm' from having non-text HEAD children
  (e.g., NetBSD mDNSShared/dns-sd.1)
  (mdoc_html.c and mdoc_term.c `Nm' handlers can be slightly simplified)

- When there is free text in the SYNOPSIS and that free text contains
  the .Nm macro, groff somehow understands to treat the .Nm as an in-line
  macro, while mandoc treats it as a block macro and breaks the line.
  No idea how the logic for distinguishing in-line and block instances
  should be, needs investigation.
  uqs@  Thu, 2 Jun 2011 11:03:51 +0200
  uqs@  Thu, 2 Jun 2011 11:33:35 +0200

--- missing man features -----------------------------------------------

- groff an-ext.tmac macros (.UR, .UE) occur in xine(5)
  reported by brad@  Sat, 15 Jan 2011 15:45:23 -0500
  also occur in freeciv-client(6) freeciv-server(6) freeciv-modpack(6)
  reported by bentley@  Tue, 30 Oct 2012 01:05:57 -0600

- -T[x]html doesn't stipulate non-collapsing spaces in literal mode

--- missing tbl features -----------------------------------------------

- implement basic non-parametric .de to support e.g. sox(1)
  reported by naddy@ Sat, 16 Oct 2010 23:51:57 +0200
  *** sox(1) still doesn't work, tbl(1) errors need investigation

- allow standalone `.' to be interpreted as an end-of-layout
  delimiter instead of being thrown away as a no-op roff line
  reported by Yuri Pankov, Wed 18 May 2011 11:34:59 CEST

--- missing misc features ----------------------------------------------

- clean up escape sequence handling, creating three classes:
  (1) fully implemented, or parsed and ignored without loss of content
  (2) unimplemented, potentially causing loss of content
      or serious mangling of formatting (e.g. \n) -> ERROR
      see textproc/mgdiff(1) for nice examples
  (3) undefined, just output the character -> perhaps WARNING

- look at pages generated from reStructeredText, e.g. devel/mercurial hg(1)
  These are a weird mixture of man(7) and custom autogenerated low-level
  roff stuff.  Figure out to what extent we can cope.
  For details, see http://docutils.sourceforge.net/rst.html
  noted by stsp@  Sat, 24 Apr 2010 09:17:55 +0200
  reminded by nicm@  Mon, 3 May 2010 09:52:41 +0100

- check compatibility with Plan9:
  http://swtch.com/usr/local/plan9/tmac/tmac.an
  http://swtch.com/plan9port/man/man7/man.html
  "Anthony J. Bentley" <anthonyjbentley@gmail.com> 28 Dec 2010 21:58:40 -0700

************************************************************************
* formatting issues: ugly output
************************************************************************

- a column list with blank `Ta' cells triggers a spurrious
  start-with-whitespace printing of a newline

- double quotes inside double quotes are escaped by doubling them
  implement this in mdoc(7), too
  so far, we only have it in roff(7) and man(7)
  reminded by millert@  Thu, 09 Dec 2010 17:29:52 -0500

- In .Bl -column,
  .It Em Authentication<tab>Key Length
  ought to render "Key Length" with emphasis, too,
  see OpenBSD iked.conf(5).
  reported again Nicolas Joly via wiz@ Wed, 12 Oct 2011 00:20:00 +0200

- empty phrases in .Bl column produce too few blanks
  try e.g. .Bl -column It Ta Ta
  reported by millert Fri, 02 Apr 2010 16:13:46 -0400

- .%T can have trailing punctuation.  Currently, it puts the trailing
  punctuation into a trailing MDOC_TEXT element inside its own scope.
  That element should rather be outside its scope, such that the
  punctuation does not get underlines.  This is not trivial to
  implement because .%T then needs some features of in_line_eoln() -
  slurp all arguments into one single text element - and one feature
  of in_line() - put trailing punctuation out of scope.
  Found in mount_nfs(8) and exports(5), search for "Appendix".

- Trailing punctuation after .%T triggers EOS spacing, at least
  outside .Rs (eek!).  Simply setting ARGSFL_DELIM for .%T is not
  the right solution, it sends mandoc into an endless loop.
  reported by Nicolas Joly  Sat, 17 Nov 2012 11:49:54 +0100

- in enclosures, mandoc sometimes fancies a bogus end of sentence
  reminded by jmc@  Thu, 23 Sep 2010 18:13:39 +0059

- formatting /usr/local/man/man1/latex2man.1 with groff and mandoc
  reveals lots of bugs both in groff and mandoc...
  reported by bentley@  Wed, 22 May 2013 23:49:30 -0600

************************************************************************
* formatting issues: gratuitous differences
************************************************************************

- .Rv (and probably .Ex) print different text if an `Nm' has been named
  or not (run a manual without `Nm blah' to see this).  I'm not sure
  that this exists in the wild, but it's still an error.

- In .Bl -bullet, the groff bullet is "+\b+\bo\bo", the mandoc bullet
  is just "o\bo".
  see for example OpenBSD ksh(1)

- .Pp between two .It in .Bl -column should produce one,
  not two blank lines, see e.g. login.conf(5).
  reported by jmc@  Sun, 17 Apr 2011 14:04:58 +0059
  reported again by sthen@  Wed, 18 Jan 2012 02:09:39 +0000 (UTC)

- If the *first* line after .It is .Pp, break the line right after
  the tag, do not pad with space characters before breaking.
  See the description of the a, c, and i commands in sed(1).

- If the first line after .It is .D1, do not assert a blank line
  in between, see for example tmux(1).
  reported by nicm@  13 Jan 2011 00:18:57 +0000

- Trailing punctuation after .It should trigger EOS spacing.
  reported by Nicolas Joly  Sat, 17 Nov 2012 11:49:54 +0100
  Probably, this should be fixed somewhere in termp_it_pre(), not sure.

- .Nx 1.0a
  should be "NetBSD 1.0A", not "NetBSD 1.0a",
  see OpenBSD ccdconfig(8).

- In .Bl -tag, if a tag exceeds the right margin and must be continued
  on the next line, it must be indented by -width, not width+1;
  see "rule block|pass" in OpenBSD ifconfig(8).

- When the -width string contains macros, the macros must be rendered
  before measuring the width, for example
    .Bl -tag -width ".Dv message"
  in magic(5), located in src/usr.bin/file, is the same
  as -width 7n, not -width 11n.
  The same applies to .Bl -column column widths;
  reported again by Nicolas Joly Thu, 1 Mar 2012 13:41:26 +0100 via wiz@ 5 Mar

- The \& zero-width character counts as output.
  That is, when it is alone on a line between two .Pp,
  we want three blank lines, not two as in mandoc.

- When .Fn arguments exceed one output line, all but the first
  should be indented, see e.g. rpc(3);
  reported by jmc@ on discuss@  Fri, 29 Oct 2010 13:48:33 +0100
  reported again by Nicolas Joly via wiz@  Sun, 18 Sep 2011 18:24:40 +0200
  Also, we don't want to break the line within the argument of:
  .Fa "chtype tl"

- Header lines of excessive length:
  Port OpenBSD man_term.c rev. 1.25 to mdoc_term.c
  and document it in mdoc(7) and man(7) COMPATIBILITY
  found while talking to Chris Bennett

- trailing whitespace must be ignored even when followed by a font escape,
  see for example 
    makes
    \fBdig \fR
    operate in batch mode
  in dig(1).

************************************************************************
* performance issues
************************************************************************

Several areas can be cleaned up to make mandoc even faster.  These are 

- improve hashing mechanism for macros (quite important: performance)

- improve hashing mechanism for characters (not as important)

- the PDF file is HUGE: this can be reduced by using relative offsets

- instead of re-initialising the roff predefined-strings set before each
  parse, create a read-only version the first time and copy it 

************************************************************************
* structural issues
************************************************************************

- We use the input line number at several places to distinguish
  same-line from different-line input.  That plainly doesn't work
  with user-defined macros, leading to random breakage.

- Find better ways to prevent endless loops
  in roff(7) macro and string expansion.
 
- Finish cleanup of date handling.
  Decide which formats should be recognized where.
  Update both mdoc(7) and man(7) documentation.
  Triggered by  Tim van der Molen  Tue, 22 Feb 2011 20:30:45 +0100