path: root/TODO

************************************************************************
* Official mandoc TODO.
* $Id$
************************************************************************

Many issues are annotated for difficulty as follows:

 - loc = locality of the issue
    *    single file issue, affects file only, or very few
    **   single module issue, affects several files of one module
    ***  cross-module issue, significantly impacts multiple modules
         and may require substantial changes to internal interfaces
 - exist = difficulty of the existing code in this area
    *    affected code is straightforward and easy to read and change
    **   affected code is somewhat complex, but once you understand
         the design, not particularly difficult to understand
    ***  affected code uses a special, exceptionally tricky design
 - algo = difficulty of the new algorithm to be written
    *    the required logic and code is straightforward
    **   the required logic is somewhat complex and needs a careful design
    ***  the required logic is exceptionally tricky,
         maybe an approach to solve that is not even known yet
 - size = the amount of code to be written or changed
    *    a small number of lines (at most 100, usually much less)
    **   a considerable amount of code (several dozen to a few hundred)
    ***  a large amount of code (many hundreds, maybe thousands)
 - imp = importance of the issue
    *    mostly for completeness
    **   would be nice to have
    ***  issue causes considerable inconvenience

Obviously, as the issues have not been solved yet, these annotations
are mere guesses, and some may be wrong.

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

- The abort() in bufcat(), html.c, can be triggered via buffmt_includes()
  by running -Thtml -Oincludes on a file containing a long .In argument.
  Fixing this will probably require reworking the whole bufcat() concept.
  loc **  exist *  algo *  size **  imp **

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

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

- .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).
  loc ***  exist ***  algo **  size **  imp **  (parser reorg would help)

- .fc (field control)
  found by naddy@ in xloadimage(1)
  loc **  exist ***  algo *  size *  imp *
  
- .nr third argument (auto-increment step size, requires \n+)
  found by bentley@ in sbcl(1)  Mon, 9 Dec 2013 18:36:57 -0700
  loc *  exist *  algo *  size *  imp **

- .ns (no-space mode) occurs in xine-config(1)
  reported by brad@  Sat, 15 Jan 2011 15:45:23 -0500
  loc ***  exist ***  algo ***  size **  imp *

- .ta (tab settings) occurs in ircbug(1) and probably gnats(1)
  reported by brad@  Sat, 15 Jan 2011 15:50:51 -0500
  also Tcl_NewStringObj(3) via wiz@  Wed, 5 Mar 2014 22:27:43 +0100
  also posix2time(3) Carsten Kunze  Mon, 1 Dec 2014 13:03:10 +0100
  loc **  exist ***  algo **  size **  imp ***

- .ti (temporary indent)
  found by naddy@ in xloadimage(1)
  found by bentley@ in nmh(1)  Mon, 23 Apr 2012 13:38:28 -0600
  loc **  exist **  algo **  size *  imp ** (parser reorg helps a lot)

- .while and .shift 
  found by jca@ in ratpoison(1)  Sun, 30 Jun 2013 12:01:09 +0200
  loc *  exist **  algo **  size **  imp **

- \h horizontal move
  found in cclive(1) DocBook output
  Anthony J. Bentley on discuss@  Sat, 21 Sep 2013 22:29:34 -0600
  loc **  exist **  algo **  size *  imp ** (parser reorg helps a lot)

- \n+ and \n- numerical register increment and decrement
  found by bentley@ in sbcl(1)  Mon, 9 Dec 2013 18:36:57 -0700
  loc *  exist *  algo *  size *  imp **

- \w'' improve width measurements
  would not be very useful without an expression parser, see below
  needed for Tcl_NewStringObj(3) via wiz@  Wed, 5 Mar 2014 22:27:43 +0100
  loc **  exist ***  algo ***  size *  imp ***

- using undefined strings or macros defines them to be empty
  wl@  Mon, 14 Nov 2011 14:37:01 +0000
  loc *  exist *  algo *  size *  imp *

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

- fix bad block nesting involving multiple identical explicit blocks
  see the OpenBSD mdoc_macro.c 1.47 commit message
  loc *  exist ***  algo ***  size *  imp **

- .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
  loc *  exist ***  algo ***  size *  imp **

- 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
  loc *  exist ***  algo ***  size **  imp **

- \\ 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 -centered implies -filled, not -unfilled, which is not
  easy to implement; it requires code similar to .ce, which
  we don't have either.
  Besides, groff has bug causing text right *before* .Bd -centered
  to be centered as well.
  loc ***  exist ***  algo **  size **  imp **  (parser reorg would help)

- .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).
  loc ***  exist ***  algo **  size **  imp **  (parser reorg would help)

- implement blank `Bl -column', such as
  .Bl -column
  .It foo Ta bar
  .El
  loc *  exist ***  algo ***  size *  imp *

- explicitly disallow nested `Bl -column', which would clobber internal
  flags defined for struct mdoc_macro
  loc *  exist *  algo *  size *  imp **

- 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
  loc *  exist ***  algo **  size *  imp **

- 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
  loc *  exist *  algo *  size *  imp **  (already done?)

- have a blank `It' head for `Bl -tag' not puke
  loc *  exist *  algo *  size *  imp **  (already done?)

- check whether it is correct that `D1' uses INDENT+1;
  does it need its own constant?
  loc *  exist **  algo **  size *  imp **

- 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)

- support translated section names
  e.g. x11/scrotwm scrotwm_es.1:21:2: error: NAME section must be first
  that one uses NOMBRE because it is spanish...
  deraadt tends to think that section-dependent macro behaviour
  is a bad idea in the first place, so this may be irrelevant
  loc **  exist **  algo **  size *  imp **

- 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
  loc *  exist **  algo ***  size *  imp **

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

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

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

- look at the POSIX manuals in the books/man-pages-posix port,
  they use some unsupported tbl(7) features.
  loc *  exist **  algo **  size **  imp ***

- use Unicode U+2500 to U+256C for table borders
  in tbl(7) -Tutf-8 output
  suggested by bentley@  Tue, 14 Oct 2014 04:10:55 -0600
  loc *  exist **  algo *  size *  imp **

- 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
  loc **  exist **  algo **  size *  imp **

--- missing eqn features -----------------------------------------------

- The "size" keyword is parsed, but ignored by the formatter.
  loc *  exist *  algo *  size *  imp *

- The spacing characters `~', `^', and tab are currently ignored,
  see User's Guide (Second Edition) page 2 section 4.
  loc *  exist *  algo **  size *  imp **

- Mark and lineup are parsed and ignored,
  see User's Guide (Second Edition) page 5 section 15.
  loc **  exist **  algo **  size **  imp **

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

- italic correction (\/) in PostScript mode
  Werner LEMBERG on groff at gnu dot org  Sun, 10 Nov 2013 12:47:46
  loc **  exist **  algo *  size *  imp *

- When makewhatis(8) encounters a FATAL parse error,
  it silently treats the file as formatted, which makes no sense
  at all for paths like man1/foo.1 - and which also contradicts
  what the manual says at the end of the description.
  The end result will be ENOENT for file names returned
  by mansearch() in manpage.file.
  loc *  exist *  algo *  size *  imp **

- makewhatis(8) for preformatted pages:
  parse the section number from the header line
  and compare to the section number from the directory name
  loc *  exist *  algo *  size *  imp **

- Does makewhatis(8) detect missing NAME sections, missing names,
  and missing descriptions in all the file formats?
  loc *  exist *  algo *  size *  imp ***

- 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
  loc ***  exist **  algo **  size **  imp *** (parser reorg helps)

- kettenis wants base roff, ms, and me  Fri, 1 Jan 2010 22:13:15 +0100 (CET)
  loc **  exist **  algo **  size ***  imp *

--- compatibility checks -----------------------------------------------

- is .Bk implemented correctly in modern groff?
  sobrado@  Tue, 19 Apr 2011 22:12:55 +0200

- compare output to Heirloom roff, Solaris roff, and
  http://repo.or.cz/w/neatroff.git  http://litcave.rudi.ir/

- look at AT&T DWB http://www2.research.att.com/sw/download
  Carsten Kunze <carsten dot kunze at arcor dot de> has patches
  Mon, 4 Aug 2014 17:01:28 +0200

- 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

- look at pages generated from ronn(1) github.com/rtomayko/ronn
  (based on markdown)

- look at pages generated from Texinfo source by yat2m, e.g. security/gnupg
  First impression is not that bad.

- look at pages generated by pandoc; see
  https://github.com/jgm/pandoc/blob/master/src/Text/Pandoc/Writers/Man.hs
  porting planned by kili@  Thu, 19 Jun 2014 19:46:28 +0200

- 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

- check compatibility with the man(7) formatter
  https://raw.githubusercontent.com/rofl0r/hardcore-utils/master/man.c

- check compatibility with
  http://ikiwiki.info/plugins/contrib/mandoc/
  https://github.com/schmonz/ikiwiki/compare/mandoc
  Amitai Schlair  Mon, 19 May 2014 14:05:53 -0400

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

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

- 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
  loc *  exist ***  algo ***  size **  imp ***

- 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
  loc *  exist ***  algo ***  size *  imp **

- .%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".
  loc **  exist **  algo ***  size *  imp **

- 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
  loc *  exist **  algo **  size *  imp **

- global variables in the SYNOPSIS of section 3 pages
  .Vt vs .Vt/.Va vs .Ft/.Va vs .Ft/.Fa ...
  from kristaps@  Tue, 08 Jun 2010 11:13:32 +0200

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

- 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

--- PDF issues ---------------------------------------------------------

- PDF output doesn't use a monospaced font for .Bd -literal
  Example: "mandoc -Tpdf afterboot.8 > output.pdf && pdfviewer output.pdf".
  Search the text "Routing tables".
  Also check what PostScript mode does when fixing this.
  reported by juanfra@ Wed, 04 Jun 2014 21:44:58 +0200
  instructions from juanfra@  Wed, 11 Jun 2014 02:21:01 +0200
    add a new <</Type /Font>> block to the PDF files with /BaseFont /Courier
    and change the /Name from /F0 to the new font (/F5 (?)).
  loc *  exist **  algo **  size *  imp **

--- HTML issues --------------------------------------------------------

- <dl><dt><dd> formatting is ugly
  hints are easy to find on the web, e.g.
  http://stackoverflow.com/questions/1713048/
  see also matthew@  Fri, 18 Jul 2014 19:25:12 -0700
  loc *  exist *  algo **  size *  imp ***

- jsg on icb, Nov 3, 2014:
  try to guess Xr in man(7) for hyperlinking

- The tables used to render the three-part page headers actually force
  the width of the <body> to the max-width given for <html>.
  Not yet sure how to fix that...
  Observed by an Anonymous Coward on undeadly.org:
  http://undeadly.org/cgi?action=article&sid=20140925064244&pid=1
  loc *  exist *  algo **  size *  imp ***

- consider whether <var> can be used for Ar Dv Er Ev Fa Va.
  from bentley@  Wed, 13 Aug 2014 09:17:55 -0600

- check https://github.com/trentm/mdocml

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

- .Fn reopens a new scope after punctuation in mandoc,
  but closes its scope for good in groff.
  Do we want to change mandoc or groff?
  Steffen Nurpmeso  Sat, 08 Nov 2014 13:34:59 +0100
  loc *  exist **  algo **  size *  imp **

- .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.
  loc *  exist *  algo *  size *  imp * (already done?)

- In .Bl -bullet, the groff bullet is "+\b+\bo\bo", the mandoc bullet
  is just "o\bo".  The problem is to not break ps/pdf when fixing.
  see for example OpenBSD ksh(1)
  loc **  exist **  algo **  size *  imp **

- In .Bl -enum -width 0n, groff continues one the same line after
  the number, mandoc breaks the line.
  mail to kristaps@  Mon, 20 Jul 2009 02:21:39 +0200
  loc *  exist **  algo **  size *  imp **

- .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)
  loc *  exist ***  algo **  size *  imp **

- 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).
  loc *  exist **  algo **  size *  imp **

- 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
  loc *  exist **  algo **  size *  imp **

- 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.
  loc *  exist **  algo **  size *  imp **

- .Nx 1.0a
  should be "NetBSD 1.0A", not "NetBSD 1.0a",
  see OpenBSD ccdconfig(8).
  loc *  exist *  algo *  size *  imp **

- 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).
  loc *  exist ***  algo **  size *  imp **

- 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
  reported again by Franco Fichtner Fri, 27 Sep 2013 21:02:28 +0200
  loc ***  exist ***  algo ***  size **  imp ***
  An easy partial fix would be to just skip the first word if it starts
  with a dot, including any following white space, when measuring.
  loc *  exist *  algo *  size *  imp ***

- 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.
  loc **  exist **  algo **  size *  imp **

- 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
  loc *  exist *  algo *  size *  imp *

- 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).
  loc **  exist **  algo **  size *  imp **

************************************************************************
* warning issues
************************************************************************

- check that MANDOCERR_BADTAB is thrown in the right cases,
  i.e. when finding a literal tab character in fill mode,
  and possibly change the wording of the warning message
  to refer to fill mode, not literal mode
  See the mail from Werner LEMBERG on the groff list,
  Fri, 14 Feb 2014 18:54:42 +0100 (CET)
  loc *  exist **  algo **  size *  imp **

- warn about attempts to call non-callable macros
  Steffen Nurpmeso  Tue, 11 Nov 2014 22:55:16 +0100
  Note that formatting is inconsistent in groff.
  .Fn Po prints "Po()", .Ar Sh prints "file ..." and no "Sh".
  Relatively hard because the relevant code is scattered
  all over mdoc_macro.c and all subtly different.
  loc **  exist **  algo **  size **  imp **

- warn about "new sentence, new line"
  loc **  exist **  algo ***  size *  imp **

- mandoc_special does not really check the escape sequence,
  but just the overall format
  loc **  exist **  algo ***  size **  imp **

- integrate mdoclint into mandoc ("end-of-line whitespace" thread)
  from jmc@  Mon, 13 Jul 2009 17:12:09 +0100
  from kristaps@  Mon, 13 Jul 2009 18:34:53 +0200
  from jmc@  Mon, 13 Jul 2009 17:45:37 +0059
  from kristaps@  Mon, 13 Jul 2009 19:02:03 +0200
  (mostly done, check what remains)

- -Tlint parser errors and warnings to stdout
  to tech@mdocml, naddy@  Wed, 28 Sep 2011 11:21:46 +0200
  wait!  kristaps@  Sun, 02 Oct 2011 17:12:52 +0200

- for system errors, use errno/strerror/warn/err

************************************************************************
* documentation issues
************************************************************************

- mention hyphenation rules:
  breaking at letter-letter in text mode (not macro args)
  proper hyphenation is unimplemented

- talk about spacing around delimiters
  to jmc@, kristaps@  Sat, 23 Apr 2011 17:41:27 +0200

- mark macros as: page structure domain, manual domain, general text domain
  is this useful?

- mention /usr/share/misc/mdoc.template in mdoc(7)?

- Is all the content from http://www.std.com/obi/BSD/doc/usd/28.tbl/tbl
  covered in tbl(7)?

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

- Why are we using MAP_SHARED, not MAP_PRIVATE for mmap(2)?
  How does SQLITE_CONFIG_PAGECACHE actually work?  Document it!
  from kristaps@  Sat, 09 Aug 2014 13:51:36 +0200

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 
  loc *  exist **  algo **  size *  imp **

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

- Use libz directly instead of forking gunzip(1).
  Suggested by bapt at FreeBSD among others.

- 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

- struct mparse refactoring
  Steffen Nurpmeso  Thu, 04 Sep 2014 12:50:00 +0200

- Consider creating some views that will make the database more
  readable from the sqlite3 shell.  Consider using them to
  abstract from the database structure, too.
  suggested by espie@  Sat, 19 Apr 2014 14:52:57 +0200

************************************************************************
* CGI issues
************************************************************************

 - Enable HTTP compression by detecting gzip encoding and filtering
   output through libz.
 - Sandbox (see OpenSSH).
 - Enable caching support via HTTP 304 and If-Modified-Since.
 - Allow for cgi.h to be overridden by CGI environment variables.
   Otherwise, binary distributions will inherit the compile-time
   behaviour, which is not optimal.
 - Have Mac OSX systems automatically disable -static compilation of the
   CGI: -static isn't supported.