xmltv-util 0.5.69-1 / usr / bin / tv_extractinfo_en

#!/usr/bin/perl -w

=pod

=head1 NAME

tv_extractinfo_en - read English-language listings and extract info
from programme descriptions.

=head1 SYNOPSIS

tv_extractinfo_en [--help] [--output FILE] [FILE...]

=head1 DESCRIPTION

Read XMLTV data and attempt to extract information from
English-language programme descriptions, putting it into
machine-readable form.  For example the human-readable text '(repeat)'
in a programme description might be replaced by the XML element
<previously-shown>.

B<--output FILE> write to FILE rather than standard output

This tool also attempts to split multipart programmes into their
constituents, by looking for a description that seems to contain lots
of times and titles.  But this depends on the description following
one particular style and is useful only for some listings sources
(Ananova).

If some text is marked with the 'lang' attribute as being some
language other than English ('en'), it is ignored.

=head1 SEE ALSO

L<xmltv(5)>.

=head1 AUTHOR

Ed Avis, ed@membled.com

=head1 BUGS

Trying to parse human-readable text is always error-prone, more so
with the simple regexp-based approach used here.  But because TV
listing descriptions usually conform to one of a few set styles,
tv_extractinfo_en does reasonably well.  It is fairly conservative,
trying to avoid false positives (extracting 'information' which
isnE<39>t really there) even though this means some false negatives
(failing to extract information and leaving it in the human-readable
text).

However, the leftover bits of text after extracting information may
not form a meaningful English sentence, or the punctuation may be
wrong.

On the two listings sources currently supported by the XMLTV package,
this program does a reasonably good job.  But it has not been tested
with every source of anglophone TV listings.

=cut

use strict;
use XMLTV::Version '$Id: tv_extractinfo_en,v 1.70 2015/07/12 00:46:37 knowledgejunkie Exp $ ';
use XMLTV::Date;
use Date::Manip;
use Carp;
use Getopt::Long;

BEGIN {
    if (int(Date::Manip::DateManipVersion) >= 6) {
	Date::Manip::Date_Init("SetDate=now,UTC");
    } else {
	Date::Manip::Date_Init("TZ=UTC");
    }
}

# Use Log::TraceMessages if installed.
BEGIN {
    eval { require Log::TraceMessages };
    if ($@) {
	*t = sub {};
	*d = sub { '' };
    }
    else {
	*t = \&Log::TraceMessages::t;
	*d = \&Log::TraceMessages::d;
	Log::TraceMessages::check_argv();
    }
}

# Use Term::ProgressBar if installed.
use constant Have_bar => eval { require Term::ProgressBar; 1 };

use XMLTV;
use XMLTV::TZ qw(gettz offset_to_gmt);
use XMLTV::Clumps qw(clump_relation relatives fix_clumps nuke_from_rel);
use XMLTV::Usage <<END
$0: read English-language listings and extract info from programme descriptions
usage: $0 [--help] [--output FILE] [FILE...]
END
;


# There are some seeming bugs in Perl which corrupt the stop time of
# programmes.  They are less in 5.6.1 than 5.6.0 but still there.  The
# debugging assertions cst() and no_shared_scalars() have the effect
# of stopping the problem (it's a Heisenbug).  So one way of making
# stop times correct is to call this routines regularly.
#
# Alternatively, we can limit the script's functionality to work
# around the bug.  It seems to affect stop times, so if we just don't
# add stop times things should be okay.
#
# This flag decides which of the two to pick: slow but with maximum
# information about stop times, or fast without them.  (Stop times can
# easily be added back in by tv_sort, and they weren't that good
# anyway, so you should probably leave this off.)
#
my $SLOW = 0;
warn "this version has debugging calls, will run slowly\n" if $SLOW;

# It might turn out that a particular version of perl is needed.
# BEGIN {
#     eval { require 5.6.1 };
#     if ($@) {
# 	for ($@) {
# 	    chomp;
# 	    s/, stopped at .+$//;
# 	    warn "$_, continuing but output may be wrong\n";
# 	}
#     }
# }

sub list_names( $ );
sub check_same_channel( $ );
sub special_category( $ );
sub special_multipart( $ );
sub special_credits( $ );
sub special_radio4( $ );
sub special_split_title( $ );
sub special_film( $ );
sub special_new_series( $ );
sub special_year( $ );
sub special_tv_movie( $ );
sub special_teletext_subtitles( $ );
sub special_useless( $ );
sub check_time_fits( $$ );
sub cst( $ );
sub no_shared_scalars( $ );
sub has( $$@ );
sub hasp( $$$ );
sub pd( $ );
sub ud( $ );
sub nd( $ );
sub bn( $ );
sub munge( $ );
sub multipart_split_desc( $$ );
sub clocks_poss( $ );
sub time12to24( $ );
sub add( $$$ );
sub scrub_empty( @ );
sub set_stop_time( $$ );
sub dump_pseudo_programme( $ );

# --no-create-sub-titles is an undocumented switch, affecting the
# splitting of multipart programmes only, to not break a title
# containing colons into title and sub-title, but always keep it as a
# single title containing a colon.  This is for consistency with some
# data sources that do this.
#
my ($opt_help, $opt_output, $opt_no_create_sub_titles);
GetOptions('help' => \$opt_help, 'output=s' => \$opt_output,
	   'no-create-sub-titles' => \$opt_no_create_sub_titles)
  or usage(0);
usage(1) if $opt_help;
@ARGV = ('-') if not @ARGV;

####
# Language selection stuff.
#
my $LANG = 'en';

# bn(): wrapper for XMLTV::best_name().  Does some memoizing (so
# assumes that the languages in a list of pairs will not change).
#
my %bn;
sub bn( $ ) {
    my $pairs = shift;
    return undef if not defined $pairs;
    die 'bn(): expected ref to list of [text,lang] pairs'
      if ref $pairs ne 'ARRAY';
    for ($bn{$pairs}) {
	return $_ if defined;
	foreach (@$pairs) {
	    carp "found bad [text,lang] pair: $_" if ref ne 'ARRAY';
	}
	return $_ = XMLTV::best_name([ $LANG ], $pairs);
    }
}

# pair_ok(): returns whether a [ text, lang ] pair is usable.
sub pair_ok( $ ) {
    not defined $_->[1] or $_->[1] =~ /^$LANG(?:_\w+)?$/o;
}

####
# Human name stuff.
#

# Regular expression to match a name
my $UC = '[A-Z]';         # upper case
my $LC = "[a-z]";         # lower case
my $AC_P = "[\'A-Za-z-]"; # any case with punctuation
my $NAME_RE;
{
    # One word of a name.  Uppercase, anycase then ending in at least
    # two lowercase.  Alternatively, uppercase then lowercase (eg
    # 'Lee'), all uppercase ('DJ') or just uppercase and an optional dot (for
    # initials).
    #
    my $name_comp_re = "(?:$UC(?:(?:$AC_P+$LC$LC)|(?:$LC+)|(?:$UC+)|\\.?))";
    foreach ('Simon', 'McCoy') {
	die "cannot match name component $_" unless /^$name_comp_re$/o;
    }
    foreach ("Valentine's") {
	die "wrongly matched name component $_" if /^$name_comp_re$/o;
    }

    # Additional words valid in the middle of names.
    my $name_join_re = "(?:von|van|de|di|da|van\\s+den|bin|ben|al)";

    # A name must have at least two components.  This excludes those
    # celebrities known only by first name but it's a reasonable
    # heuristic for distinguishing between the names of actors and the
    # names of characters.
    #
    my $name_re = "(?:$name_comp_re\\s+(?:(?:(?:$name_comp_re)|$name_join_re)\\s+)*$name_comp_re)";
    foreach ('Simon McCoy', 'Annie Wu') {
	die "cannot match $_" unless /^$name_re$/o;
    }

    # Special handling for some titles.  But others beginning 'the'
    # are specifically excluded (to avoid 'the Corornation Street
    # star' parsing as '$NAME_RE star').
    #
    $NAME_RE = "(?<!the\\s)(?:(?:[Tt]he\\s+Rev(?:\\.|erend)\\s+)?$name_re)";
}

# Regexp to match a list of names: 'Tom, Dick, and Harry'
my $NAMES_RE = "(?:$NAME_RE(?:(?:\\s*,\\s*$NAME_RE)*(?:\\s*,?\\s*\\band\\s+$NAME_RE))?(?!\\s*(?:and\\s+$UC|from|[0-9])))";

# Subroutine to extract the names from this list
sub list_names( $ ) {
    die 'usage: list_names(English string listing names)'
      if @_ != 1;
    local $_ = shift; die if not defined;
    t 'list_names() processing string: ' . d $_;
    my @r;

    s/^($NAME_RE)\s*// or die "bad 'names' '$_'";
    push @r, $1;

    while (s/^,?\s*(?:and\s+)?($NAME_RE)\s*//) {
	push @r, $1;
    }
    die "unmatched bit of names $_" unless $_ eq '';

    return @r;
}

my @tests =
  (
   [ 'Richard Whiteley and Carol Vorderman', [ 'Richard Whiteley', 'Carol Vorderman' ] ],
   [ 'show presented by Jonathan Ross, with', [ 'Jonathan Ross' ] ],
   [ 'Shane Richie, Michael Starke and Scott Wright',
     [ 'Shane Richie', 'Michael Starke', 'Scott Wright' ] ],
   [ 'Basil Brush,Barney Harwood and Ugly Yetty present',
     [ 'Basil Brush', 'Barney Harwood', 'Ugly Yetty'] ],
  );
foreach (@tests) {
    my ($in, $expected) = @$_;
    for ($in) {
	/($NAMES_RE)/o or die "$in doesn't match \$NAMES_RE";
	my @out = list_names($1);
	local $Log::TraceMessages::On = 1;
	if (d(\@out) ne d($expected)) {
	    die "$in split into " . d(\@out);
	}
    }
}


####
# Date handling stuff.
#
# This loses any information on partial dates (FIXME).
#
sub pd( $ ) {
    for ($_[0]) {
	return undef if not defined;
	return parse_date($_);
    }
}
sub ud( $ ) {
    for ($_[0]) {
	return undef if not defined;
	return UnixDate($_, '%q');
    }
}
sub nd( $ ) {
    for ($_[0]) {
	return undef if not defined;
	return ud(pd($_));
    }
}

# Memoize some subroutines if possible.  FIXME commonize to
# XMLTV::Memoize.
#
eval { require Memoize };
unless ($@) {
    foreach (qw(parse_date UnixDate Date_Cmp
		clocks_poss time12to24)) {
	Memoize::memoize($_) or die "cannot memoize $_: $!";
    }
}

my $encoding;
my $credits;
my %ch;
my @progs;
XMLTV::parsefiles_callback(sub( $ ) { $encoding = shift },
			   sub( $ ) { $credits = shift },
			   sub( $ ) { my $c = shift; $ch{$c->{id}} = $c },
			   sub( $ ) { push @progs, shift },
			   @ARGV);
# Assume encoding is a superset of ASCII, and that Perl's regular
# expressions work with it in the current locale.
#

my $related = clump_relation(\@progs);

# Apply all handlers.  We just haphazardly
# run one after the other; when a programme has been run
# through all of them in sequence without any changes, we
# move it to @progs_done.
#
# The reason for using _lists_ is that some handlers turn
# a single programme into several.
#
my @progs_done = ();
my $bar = new Term::ProgressBar('munging programmes', scalar @progs)
  if Have_bar;
while (@progs) {
    # Deal with one more programme from the input, it may transform
    # itself into one or more programmes which need processing in
    # turn.  When all the offspring are dealt with we have finally
    # finished with that input and can update the progress bar.
    #
    no_shared_scalars(\@progs) if $SLOW;
    push @progs_done, munge(shift @progs);
    update $bar if Have_bar;
}
if ($SLOW) { cst $_ foreach @progs_done }

my %w_args = ();
if (defined $opt_output) {
    my $fh = new IO::File ">$opt_output";
    die "cannot write to $opt_output\n" if not $fh;
    %w_args = (OUTPUT => $fh);
}
XMLTV::write_data([ $encoding, $credits, \%ch, \@progs_done ], %w_args);
exit();

# Take a programme, munge it and return a list of programmes (empty if
# the programme was deleted).  Uses the global $related to fiddle with
# other programmes in the same clump.
#
sub munge( $ ) {
#    local $Log::TraceMessages::On = 1;
    t 'munge() ENTRY';
    my @todo = (shift);
    my @done;
    t 'todo list initialized to: ' . d \@todo;
    t 'done list initialized to: ' . d \@done;
    t 'relatives of todo programme: ' . d relatives($related, $todo[0]);

    # Special-case mungers for various programme types.  Each of these
    # should take a single programme and return: a reference to a list of
    # programmes, if successful; undef, if the programme is to be left
    # alone.  Most special-case handlers will not break up a programme
    # into several others, so the returned list will have only one
    # element.
    #
    # A handler may modify the programme passed in iff it returns a list
    # of munged programmes.
    #
    # Ones earlier in the list get the chance to run first, so in general
    # things like splitting programmes (which may be relied on by other
    # handlers) should come at the top and last-chance guesswork (such as
    # parsing English text) at the bottom.
    #
    my @special_handlers =
      (
       \&special_multipart,
       \&special_category,
       \&special_credits,
       \&special_new_series,
       \&special_year,
       \&special_tv_movie,
       \&special_teletext_subtitles,
       \&special_useless,

       # There are three handlers specific to Ananova / UK listings.  I
       # haven't yet decided what to do with them: should they be in this
       # program and enabled with a special flag, or moved into the
       # Ananova grabber?
       #
       # They haven't been ported to the new XMLTV.pm data structures, so
       # leave them commented for now.
       #
       #   \&special_radio4,
       #   \&special_split_title,
       #   \&special_film,
      );

  PROG: while (@todo) {
	my $prog = shift @todo;
	t('running handlers for prog: ' . d($prog));
	my $prog_length;
	if (defined $prog->{stop}) {
	    # Get the programme length in seconds.
	    my $delta = DateCalc($prog->{start}, $prog->{stop});
	    $prog_length = Delta_Format($delta, 0, '%st') if defined $delta;
	}
	foreach (@special_handlers) {
	    t('running handler: ' . d($_));
	    my $out = $_->($prog);
	    if (defined $out) {
		t('gave new list of progs: ' . d($out));
		die "handler didn't return list of progs"
		  if ref($out) ne 'ARRAY';
		if ($SLOW) { cst $_ foreach @$out }
		check_time_fits($prog, $out);
		if ($SLOW) { cst $_ foreach @$out }
		fix_clumps($prog, $out, $related);
		foreach (@$out) {
		    cst $_ if $SLOW;
		    # Sanity check that length hasn't increased.
		    if (defined $_->{stop}) {
			my $delta = DateCalc($_->{start}, $_->{stop});
			if (defined $prog_length) {
			    my $new_length = Delta_Format($delta, 0, '%st');
			    if ($new_length > $prog_length) {
				local $Log::TraceMessages::On = 1;
				t 'original programme (after handlers run): ' . d $prog;
				t 'split into: ' . d $out;
				t 'offending result: ' . d $_;
				t 'length of result: ' . d $new_length;
				t 'length of original programme: ' . d $prog_length;
				die 'split into programme longer than the original';
			    }
			}
		    }
		}
		push @todo, @$out;
		next PROG;
	    }
	    t('gave undef');
	}
	cst $prog if $SLOW;
	t 'none of the handlers fired, finished with this prog';
	cst $prog if $SLOW;
	push @done, $prog;
    }
    return @done;
}

# All the special handlers

# special_category()
#
# Some descriptions have the last word as the category: 'blah blah
# blah.  Western' (or 'Western series').  Remove this to the <category>
# element.
#
# Also look for magic words like 'news' or 'interview' and add them as
# categories.  This is mostly so that other handlers can then fire.
#
sub special_category( $ ) {
    t 'special_category() ENTRY';
    my $p = shift;
    my $changed = 0;

    # First, non-destructively look for 'news' in title or desc.
    foreach (qw(title desc)) {
	foreach my $pair (grep { pair_ok($_) } @{$p->{$_}}) {
	    t "pair for $_: " . d $pair;
	    if ($pair->[0] =~ /\bnews/i) {
		t 'matches "news"';

		if (hasp($p, 'category', sub { $_[0] =~ /\b(?:soap|drama|game show)\b/i })) {
		    t '...but clearly not a news programme';
		}
		else {
		    $changed |= add($p, 'category', 'news');
		    cst $p if $SLOW;
		}
	    }
	    if ($pair->[0] =~ /\binterviews\b/i) {
		t 'matches "interviews"';
		$changed |= add($p, 'category', 'talk');
		cst $p if $SLOW;
	    }
	}
    }

    # Now try the last-word-of-desc munging.
    my $replacement = sub( $$$$ ) {
	my ($punct, $adj, $country, $genre) = @_;
	$changed |= add($p, 'category', lc $genre);
	if (length $adj or length $country) {
	    return "$punct $adj$country$genre";
	}
	else {
	    $changed = 1;
	    return $punct;
	}
    };
    foreach (grep { pair_ok($_) } @{$p->{desc}}) {
	# 'Western'         -> ''
	# 'Western series'  -> ''
	# 'Classic Western' -> 'Classic Western'
	# etc.
	#
	$_->[0] =~ s/(^|\.|\?)\s*
	  (Classic\s+|Award-winning\s+|)
	    (Australian\s+|)
	      ([aA]dventure|[aA]nimation|[bB]iopic|[cC]hiller
	       |[cC]omedy|[dD]ocumentary|[dD]rama|[fF]antasy
	       |[hH]eadlines|[hH]ighlights|[hH]orror|[mM]agazine
	       |[mM]elodrama|[mM]usical|[mM]ystery|[oO]mnibus
	       |[qQ]uiz|[rR]omance|[sS]itcom|[tT]earjerker
	       |[tT]hriller|[wW]estern)\s*(?:series\s*|)$/$replacement->($1, $2, $3, $4)/xe;
    }

    if ($changed) {
	t 'some categories found, programme: ' . d $p;
	scrub_empty($p->{title}, $p->{desc});
	t 'after removing empty titles and descs, programme: ' . d $p;
	return [ $p ];
    }
    else {
	return undef;
    }
}


# special_multipart()
#
# Often TV listings contain several programmes stuffed into one entry,
# which might have made sense for a printed guide to save space, but
# is stupid for electronic data.  This special handler looks at the
# programme description and haphazardly attempts to split the
# programme into its components.
#
# Parameters: a 'programme' hash reference
# Returns: reference to list of sub-programmes, or undef if programme
#          was not split
#
# We find the title using bn(), in other words we look only at
# the first title.  Similarly we use only the first description.  But
# it should work.  FIXME should split the secondary title as well!
#
sub special_multipart( $ ) {
#    local $Log::TraceMessages::On = 1;
    die "usage: special_multipart(hashref of programme details)"
      if @_ != 1;
    my $p = shift;
    cst $p if $SLOW;
    t 'special_multipart() ENTRY';
    t 'checking programme descs: ' . d $p->{desc};
    my $best_desc = bn($p->{desc});
    t 'got best desc: ' . d $best_desc;
    return undef if not $best_desc;
    my ($desc, $desc_lang) = @$best_desc;
    t 'testing description for multipart: ' . d $desc;
    local $_ = $desc;
    my @words = split;
    my @poss_times = split /[ ,;-]/;
    my @r;
    my ($p_start, $p_stop) = (pd($p->{start}), pd($p->{stop}));
    # Assume that the timezone for every time listed in the
    # description is the same as the timezone for the programme's
    # start.  FIXME will fail when timezone changes partway through.
    #
    my $tz = gettz($p->{start});

    my $day;
    if (defined $tz) {
	# Find the base day taking into account timezones.  Eg if a
	# programme starts at 00:45 BST on the 20th and then lists
	# times as '01:00' etc, the base date for these times is the
	# 20th, even though the real start time is 23:45 UTC on the
	# 19th.
	#
	$day = pd(UnixDate(Date_ConvTZ($p_start, 'UTC', offset_to_gmt($tz)), '%Q'));
    }
    else {
	$day = pd(UnixDate($p_start, '%q'));
    }
    t "day is $day";
    # FIXME won't be correct when split programme spans days.

    # Sanity check for a time, that it is within the main programme's
    # timespan.
    #
    my $within_time_period = sub {
	my $t = shift;
	t("checking whether $t is in time period $p_start.."
	  . (defined $p_stop ? $p_stop : ''));
	if (Date_Cmp($t, $p_start) < 0) {
	    # Before start of programme, that makes no sense.
	    return 0;
	}
	if (defined $p_stop and Date_Cmp($p_stop, $t) < 0) {
	    # After end of programme, likewise.
	    return 0;
	}
	return 1;
    };

    # Three different ways of interpreting a time.  Return undef if
    # not valid under that system, a 24 hour hh:mm otherwise.
    #
    # FIXME doesn't handle multiparts bridging noon or midnight.
    #
    my $as_12h_am = sub {
	my $w = shift;
	$w =~ s/[,;.]$//;
	t "trying $w as 12 hour am time";
	clocks_poss($w)->[0] || return undef;
	return time12to24("$w am");
    };
    my $as_12h_pm = sub {
	my $w = shift;
	$w =~ s/[,;.]$//;
	t "trying $w as 12 hour pm time";
	clocks_poss($w)->[0] || return undef;
	return time12to24("$w pm");
    };
    my $as_24h = sub {
	my $w = shift;
	$w =~ s/[,;.]$//;
	t "trying $w as 24 hour time";
	clocks_poss($w)->[1] || return undef;
	$w =~ tr/./:/;
	return $w;
    };

    if (defined $tz) { t "using timezone $tz for interpreting times" }
    else { t "interpreting times with no timezone (ie UTC)" }

    my ($best_interp, $best_count,
	$best_first_word_is_time, $best_including_at_time);
  INTERP: foreach my $interp ($as_24h, $as_12h_am, $as_12h_pm) {
	t 'testing an interpretation of times';
	my $count = 0;
	my $first_word_is_time = 0;
	my $including_at_time = 0;
	my $prev;
	for (my $pos = 0; $pos < @poss_times; $pos++) {
	    t "testing word $poss_times[$pos] at position $pos";
	    my $w = $poss_times[$pos];
	    t "word is '$w'";
	    my $i = $interp->($w);
	    if (not defined $i) {
		t "doesn't parse to a time with this interp.";
		next;
	    }
	    warn "bad 24h returned time: $i" unless $i =~ /^\d?\d:\d\d$/;
	    t "found a time that interprets: $i";
	    my $t = Date_SetTime($day, $i);
	    die if not defined $t;
	    t "taken as day $day, gets time $t";
	    $t = Date_ConvTZ($t, offset_to_gmt($tz), 'UTC') if defined $tz;
	    t "after converting to UTC, $t";
	    if (not $within_time_period->($t)) {
		# Obviously wrong.  One bad time is enough to abandon
		# this whole interpretation and try another.
		#
		t "not within time period, whole interpretation wrong";
		next INTERP;
	    }
	    # Don't insist that times be in order, this isn't the case
	    # for all listings (eg 'News at 0700 and 0730; Weather at
	    # 0715').
	    #

	    $prev = $t;
	    ++ $count;
	    if ($pos == 0) {
		$first_word_is_time = 1;
	    }
	    if ($pos >= 2
		and $poss_times[$pos - 2] =~ /^[Ii]ncluding$/
		and $poss_times[$pos - 1] eq 'at') {
		$including_at_time = 1;
		t 'previous words are "including at", setting $including_at_time true';
	    }
	}
	t "found $count matching times and nothing badly wrong";

	if (not defined $best_interp
	    or $count > $best_count) {
	    t 'best so far';
	    $best_interp = $interp;
	    $best_count = $count;
	    $best_first_word_is_time = $first_word_is_time;
	    $best_including_at_time = $including_at_time;
	}
    }

    if (defined $best_interp) {
	t "best result found: count $best_count";
	t "first word? $best_first_word_is_time";
	t "best includes 'at time'? $best_including_at_time";
    }
    else {
	t "couldn't find any interpretation that worked at all";
    }

    # Heuristic.  We require at least three valid times to split - or
    # when the programme description begins with a time, that's also
    # good enough.  Also when the description contains 'including at'
    # followed by a time.
    #
    return undef if not defined $best_interp;
    return undef unless ($best_count >= 3
			 or $best_first_word_is_time
			 or $best_including_at_time);

#    local $Log::TraceMessages::On = 1;
    t 'looks reasonable, proceed';

    t 'calling multipart_split_desc() with words and interpretation fn';
    my $split = multipart_split_desc(\@words, $best_interp);
    t 'got result from multipart_split_desc(): ' . d $split;
    die if not defined $split->[0];
    die if not defined $split->[2];
    our @pps; local *pps = $split->[0];
    t 'got list of pseudo-programmes: ' . d \@pps;
    if (not @pps) {
	warn "programme looked like a multipart, but couldn't grok it";
	return undef;
    }
    if (@pps == 1) {
	# Didn't really split, perhaps it wasn't a multipart.
	t 'split into only one, leave unchanged';
	return undef;
    }

    foreach (@pps) {
	die if not defined;
	die if not keys %$_;
    }
    my $common = $split->[1];
    our @errors; local *errors = $split->[2];

    # We split the first description, and only after checking it did
    # look like a plausible multipart.  So if anything went wrong we
    # should warn about it.
    #
    foreach (@errors) {
	warn $_;
    }

    # What was returned is a list of pseudo-programmes, these have
    # main_desc instead of real [text, lang] descriptions, and hh:mm
    # 'time' instead of real start time+date.
    #
    # At most one of them is allowed to have time undefined; this is
    # the 'rump' of the parent programme.  Whether such a rump exists
    # depends on what kind of splitting was done.
    #
    my $seen_rump = 0;
    foreach (@pps) {
	my $time = delete $_->{time};
	die if not defined $time and $seen_rump++;
	if (defined $time) {
	    my $start = Date_SetTime($day, $time);
	    die if not defined $start;
	    $start = Date_ConvTZ($start, offset_to_gmt($tz), 'UTC') if defined $tz;
	    if (Date_Cmp($start, $p->{start}) < 0) {
		my $dump = dump_pseudo_programme($_);
		die "subprogramme ($dump, has 'time' $time) "
		  . "starts before main programme ($p->{start}, $p->{title}->[0]->[0])";
	    }
	    if (defined $p->{stop} and Date_Cmp($p->{stop}, $start) < 0) {
		my $dump = dump_pseudo_programme($_);
		die "subprogramme ($dump, has 'time' $time) starts after main one stops";
	    }

	    # Now we store the time in the official 'start' key.  But
	    # convert back to the original timezone to look nice.
	    #
	    if (defined $tz) {
		$_->{start} = ud(Date_ConvTZ($start, 'UTC', offset_to_gmt($tz))) . " $tz";
	    }
	    else {
		$_->{start} = ud($start);
	    }
	}
	else {
	    $_->{start} = $p->{start};
	}

	if (not defined $_->{main_title}) {
	    # A title is needed, normally splitting will find one, but
	    # in case it didn't...
	    #
	    $_->{title} = $p->{title};
	}

	# Now deal with each of the main_X fields turning them into
	# real X.
	#
	foreach my $key (qw(desc title sub-title)) {
	    my $v = delete $_->{"main_$key"};
	    next if not defined $v;
	    $_->{$key} = [ [ $v, $desc_lang ] ];
	}

	if (defined $common) {
	    # Add the common text to this programme.  So far it has at
	    # most one description in language $desc_lang.
	    #
	    for ($_->{desc}->[0]->[0]) {
		if (defined and length) {
		    $_ .= '. ' if $_ !~ /[.?!]\s*$/;
		    $_ .= " $common";
		}
		else {
		    $_ = $common;
		}
	    }
	    $_->{desc}->[0]->[1] = $desc_lang;
	}

	$_->{channel} = $p->{channel};
	t "set channel of split programme to $_->{channel}";
    }

    # The last subprogramme should stop at the same time as the
    # multipart programme stopped.
    #
    if (defined $p->{stop}) {
	t "setting stop time of last subprog to stop time of main prog ($p->{stop})";
	set_stop_time($pps[-1], $p->{stop});
    }
    else { t 'main prog had no stop time, not adding to last subprog' }

    # And similarly, the first should start at the same time as the
    # multipart programme.  Add a dummy prog to fill the gap if
    # necessary.
    #
    my $first_sub_start = $pps[0]->{start};
    my $cmp = Date_Cmp(pd($first_sub_start), $p_start);
    if ($cmp < 0) {
	# Should have caught this already.
	die 'first subprogramme starts _before_ main programme';
    }
    elsif ($cmp == 0) {
	# Okay.
    }
    elsif ($cmp > 0) {
	my $dummy = { title   => $p->{title},
		      channel => $p->{channel},
		      start   => $p->{start},
		      stop    => $first_sub_start };
	t 'inserting dummy subprogramme: ' . d $dummy;
	cst $dummy if $SLOW;
	unshift @pps, $dummy;
    }
    else { die }

    if ($SLOW) { cst $_ foreach @pps }
    scrub_empty($_->{title}, $_->{"sub-title"}, $_->{desc}) foreach @pps;
    t 'returning new list of programmes: ' . d \@pps;
    return \@pps;
}
# Given a programme description split into a list of words, and a
# subroutine to interpret times, return a list of the subprogrammes
# (assuming it is a multipart).
#
# Returns [pps, common, errs] where pps is a list of 'pseudo-programmes',
# hashes containing some of:
#
#  time: 24h time within the main programme's day,
#  main_title, main_desc, main_sub-title: text in the same language as
#  the desc passed in,
#
# and where common is text which belongs to the description of every
# subprogramme, and errs is a list of errors found (probably quite
# large if the description was not multipart).
#
sub multipart_split_desc( $$ ) {
    our @words; local *words = shift;
    my $interp = shift;

    # We need to decide what style of multipart listing this is.
    # There's the kind that has time - title - description for each
    # subprogramme.  There's the kind that has 'News at time0, time1,
    # time2; Weather at time3, time4'.  And then something more like a
    # normal English sentence, which of course is the hardest to
    # parse.  We use some heuristics to work out which it is and call
    # the appropriate 'parsing' routine.
    #
    t "testing for 'Including at'";
    foreach my $i (0 .. $#words - 1) {
	t "looking at pos $i, word is $words[$i]";
	if ($words[$i] =~ /^[Ii]ncluding$/ and $words[$i + 1] eq 'at') {
	    t 'yup, calling multipart_split_desc_including_at()';
	    return multipart_split_desc_including_at(\@words, $interp);
	}
    }

    t "testing for 'With X at T0, T1; ...'";
    if (@words >= 4 and $words[0] =~ /^with$/i) {
	my $first_lc_word;
	foreach (@words) {
	    if (not tr/[A-Z]//) {
		$first_lc_word = $_;
		last;
	    }
	}
	if (defined $first_lc_word and $first_lc_word eq 'at') {
	    return multipart_split_desc_rt(\@words, $interp);
	}
    }

    t "looking for two times in a row, or separated only by 'and'";
    my $prev_was_time = 0;
    foreach (@words) {
	if (defined $interp->($_)) {
	    # Found a time.
	    if ($prev_was_time) {
		t 'found two times in a row, using multipart_split_desc_simple()';
		return multipart_split_desc_simple(\@words, $interp);
	    }
	    $prev_was_time = 1;
	}
	elsif ($_ eq 'and') {
	    # Skip.
	}
	else {
	    $prev_was_time = 0;
	}
    }

    t "looking for pairs of times 'from-to'";
    foreach (@words) {
	if (/^([0-9.:]+)-([0-9.:]+)$/) {
	    my ($from, $to) = ($1, $2);
	    if (defined $interp->($from) and defined $interp->($to)) {
		return multipart_split_desc_fromto(\@words, $interp);
	    }
	}
    }

    t "must be old style of 'time title. description'";
    return multipart_split_desc_ananova(\@words, $interp);
}
# And these routines handle the different styles.
sub multipart_split_desc_ananova( $$ ) {
    our @words; local *words = shift;
    my $interp = shift;
    my @r;
    my @errors;

    # First extract any 'common text' at the start of the programme,
    # before any sub-programmes.
    #
    my $common;
    while (@words) {
	my $first = shift @words;
	if (defined $interp->($first)) {
	    unshift @words, $first;
	    last;
	}
	if (defined $common and length $common) {
	    $common .= " $first";
	}
	else {
	    $common = $first;
	}
    }
    t 'common text: ' . d $common;

    while (@words > 1) { # At least one thing after the time
	my $time = shift @words;
	my $i = $interp->($time);
	if (defined $i) {
	    my (@title_words, @desc_words);

	    # Build up a current 'pseudo-programme' with title,
	    # description and time.  It's up to our caller to
	    # manipulate these simple data structures into real
	    # programmes.
	    #
	    my $curr_pp;
	    $curr_pp->{time} = $i;
	    my $done_title = 0;
	    my @words_orig = @words;
	    while (@words) {
		my $word = shift @words;

		if (defined $interp->($word)) {
		    # Finished this bit of multipart.
		    unshift @words, $word;
		    last;
		}
		elsif (not $done_title) {
		    if ($word =~ s/[.]$// or $word =~ s/([!?])$/$1/) {
			# Finished the title, move on to description.
			$done_title = 1;
		    }
		    push @title_words, $word;
		}
		else {
		    push @desc_words, $word;
		}
	    }
	    if (not @title_words) {
		warn "trouble finding title in multipart";
		if (not @desc_words) {
		    warn "cannot find title or description in multipart";
		    @title_words = ('???');
		}
		else {
		    # Use the description so far as the title.
		    if ($desc_words[-1] eq 'at') {
			pop @desc_words;
		    }
		    @title_words = @desc_words;
		    @desc_words = ();
		}
	    }

	    # The title sometimes looks like 'History in Action: Women
	    # in the 20th Century'; this should be broken into main
	    # title and secondary title.  But not 'GNVQ: Is It For You
	    # 2'.  So arbitrarily we check that the main title has at
	    # least two words.
	    #
	    if (@title_words) {
		my (@main_title_words, @sub_title_words);

		while (@title_words) {
		    my $word = shift @title_words;
		    my $main_title_length = @main_title_words + 1;

		    # Split at colon, sometimes
		    if ((not $opt_no_create_sub_titles)
			and $main_title_length >= 2 and $word =~ s/:$//) {
			push @main_title_words, $word;
			@sub_title_words = @title_words;
			last;
		    }
		    else {
			push @main_title_words, $word;
		    }
		}

		$curr_pp->{main_title} = join(' ', @main_title_words);
		$curr_pp->{'main_sub-title'} = join(' ', @sub_title_words)
		  if @sub_title_words;
	    }

	    $curr_pp->{main_desc} = join(' ', @desc_words) if @desc_words;
	    t 'built sub-programme: ' . d $curr_pp;
	    push @r, $curr_pp;
	}
	else {
	    push @errors, "expected time in multipart description, got $time";
	    # Add it to the previous programme, so it doesn't get lost
	    if (@r) {
		my $prev = $r[-1];
		$prev->{main_desc} = '' if not defined $prev->{main_desc};
		$prev->{main_desc} .= $time;
	    }
	    else {
		# Cannot happen.  If @r is empty, this must be the
		# first word.
		#
		warn 'first word of desc is not time, but checked this before';
		# Not worthy of @errors, this is a bug in the code.
	    }
	}
    }
    foreach (@r) {
	die if not keys %$_;
	die if not defined $_->{main_title};
    }
    t 'returning list of pseudo-programmes: ' . d \@r;
    t '...and common text: ' . d $common;
    t '...and errors: ' . d \@errors;
    return [\@r, $common, \@errors];
}
sub multipart_split_desc_rt( $$ ) {
    our @words; local *words = shift;
    my $interp = shift;
    my @errors;

    my $with = shift @words;
    die if not defined $with;
    die if $with !~ /^with$/i;

    my @got;
    my @title = ();
    my @times = ();
    my $done_title = 0;
    while (@words) {
	my $w = shift @words;
	if ($w eq 'at') {
	    $done_title = 1;
	    next;
	}

	my $i = $interp->($w);
	if (defined $i) {
	    # It's a time.
	    if (not $done_title) {
		warn "unexpected time $w in multipart description, before 'at'";
		push @errors, $w;
	    }
	    else {
		push @times, $i;
	    }

	    if ($w =~ /[.;]$/) {
		# End of the list of times for this programme.
		push @got, [ [ @title ], [ @times ] ];
		@title = ();
		@times = ();
		$done_title = 0;
	    }
	    elsif ($w =~ /,$/) {
		# List continues.
	    }
	    else {
		warn "strange time $w";
	    }

	    next;
	}

	# Not a time, should be part of the title.
	if ($done_title) {
	    warn "strange word $w in multipart description, expected a time";
	    push @errors, $w;
	}
	else {
	    push @title, $w;
	}
    }

    my @r;
    foreach (@got) {
	my ($title, $times) = @$_;
	foreach (@$times) {
	    push @r, { main_title => join(' ', @$title), time => $_ };
	}
    }

    # There is no 'common text' with this splitter.
    return [\@r, undef, \@errors];
}
# Split the programme by looking for times, but each new programme has
# the same words (except times).
#
sub multipart_split_desc_simple( $$ ) {
    our @words; local *words = shift;
    my $interp = shift;

    my @common;
    my @times;
    foreach (@words) {
	die if not defined;
	my $i = $interp->($_);
	if (defined $i) {
	    push @times, $i;
	    if (@common and ($common[-1] eq 'at' or $common[-1] eq 'and')) {
		pop @common;
	    }
	}
	else {
	    push @common, $_;
	}
    }

    my @r;
    foreach (@times) {
	die if not defined;
	push @r, { time => $_ };
    }

    # No 'errors' but lots of 'common text'.
    return [ \@r, join(' ', @common), [] ];
}
sub multipart_split_desc_fromto( $$ ) {
    our @words; local *words = shift;
    my $interp = shift;
    my @r;
    my @errors;

    # This routine is limited a bit because it's expected to return
    # hashes with just 'time'.  But we know more than that, we know
    # both start time and stop time for each subprogramme.  That
    # information would be thrown away.
    #
    # For now, it seems that this kind of multipart programme always
    # has one part beginning when the previous one ended, so we can
    # just check that this property holds.  Then there will be no loss
    # of stop-time information.
    #
    my ($last_start, $last_stop);
    my @title = ();
    my $done_title = 0;
    my @desc = ();
    foreach (@words) {
	if (/^([0-9.:]+)-([0-9.:]+)$/
	    and defined(my $istart = $interp->($1))
	    and defined(my $istop = $interp->($2))) {
	    # It's a pair of times.
	    if (defined $last_start) {
		# Deal with the previous subprogramme.
		warn "mismatch between stop time $last_stop and start time $istart"
		  if $last_stop ne $istart;
		my %p = ( time => $last_start, main_title => join(' ', @title) );
		$p{main_desc} = join(' ', @desc) if @desc;
		push @r, \%p;
	    }
	    ($last_start, $last_stop) = ($istart, $istop);
	    @title = ();
	    $done_title = 0;
	    @desc = ();
	}
	elsif (/:$/) {
	    # A colon ends the title.
	    if (not $done_title) {
		(my $tmp = $_) =~ s/:$//;
		push @title, $tmp;
		$done_title = 1;
	    }
	    else {
		warn "seen colon in description: '$_'";
		push @desc, $_;
	    }
	}
	elsif ($_ eq 'with') {
	    # Also 'with' can end a title, as in 'News with...'.  This
	    # is probably the only time I've seen a use for the
	    # convention that words in titles should be capitalized.
	    #
	    # The 'with' stuff goes into the description, where some
	    # other handler can pick it up.
	    #
	    $done_title = 1;
	    push @desc, $_;
	}
	else {
	    if (not $done_title) {
		push @title, $_;
	    }
	    else  {
		push @desc, $_;
	    }
	}
    }
    if (defined $last_start) {
	my %p = ( time => $last_start, main_title => join(' ', @title) );
	$p{main_desc} = join(' ', @desc) if @desc;
	push @r, \%p;
    }

    return [ \@r, undef, [] ];
}
# Really an 'including at' programme should be sandwiched in the
# middle of its parent, but the format doesn't allow that so for
# simplicity we treat as a multipart.
#
sub multipart_split_desc_including_at( $$ ) {
    our @words; local *words = shift;
    my $interp = shift;
    my @r;
    my @rump;

    while (@words) {
	my $t;
	if (@words >= 4
	    and $words[0] =~ /^[Ii]ncluding$/
	    and $words[1] eq 'at'
	    and defined ($t = $interp->($words[2]))
	    and $words[3] =~ /^[A-Z]/) {
	    shift @words; shift @words; shift @words;
	    my @title;
	    while (@words and $words[0] =~ /^[A-Z]/) {
		my $w = shift @words;
		if ($w =~ s/[.,;]$//) {
		    push @title, $w;
		    last;
		}
		else {
		    push @title, $w;
		}
	    }
	    push @r, { time => $t, main_title => join(' ', @title) };
	}
	else {
	    push @rump, shift @words;
	}
    }

    unshift @r, { main_desc => join(' ', @rump) };
    return [ \@r, '', [] ];
}
# Is a time string using the 12 hour or 24 hour clock?  Returns a pair
# of two booleans: the first means it could be 12h, the seecond that
# it could be 24h.  Expects an h.mm or hh.mm time string.  If the
# string is not a valid time under either clock, returns [0, 0].
#
# Allows eg '5.30' to be a 24 hour time (05:30).
#
sub clocks_poss( $ ) {
    local $_ = shift;
    if (not /^(\d\d?)\.(\d\d)$/) {
	return [0, 0];
    }
    my ($hh, $mm) = ($1, $2);
    return [0, 0] if $mm > 59;
    return [0, 1] if $hh =~ /^0/;
    return [1, 1] if 1 <= $hh && $hh < 13;
    return [0, 1] if 13 <= $hh && $hh < 24;

    # Do not accept '24:00', '24:01' etc - not until it's proved we
    # need to.
    #
    return [0, 0];
}
# Debugging stringification.
sub dump_pseudo_programme( $ ) {
    my @r;
    my $pp = shift;
    foreach (qw(time main_title main_desc)) {
	push @r, $pp->{$_} if defined $pp->{$_};
    }
    return join(' ', @r);
}


# time12to24()
#
# Convert a 12 hour time string to a 24 hour one, without anything too
# fancy.  In particular the timezone is passed through unchanged.
#
sub time12to24( $ ) {
    die 'usage: time12to24(12 hour time string)' if @_ != 1;
    local $_ = shift;
    die if not defined;

    # Remove the timezone and stick it back on afterwards.
    my $tz = gettz($_);
    s/\Q$tz\E// if defined $tz;

    s/\s+//;
    my ($hours, $mins, $ampm) = /^(\d\d?)[.:]?(\d\d)\s*(am|pm)$/
      or die "bad time $_";
    if ($ampm eq 'am') {
	if (1 <= $hours and $hours < 12) {
	    $hours = $hours;      # 5am = 05:00
	}
	elsif ($hours == 12) {
	    $hours = 0;           # 12am = 00:00
	}
	else { die "bad number of hours $hours" }
    }
    elsif ($ampm eq 'pm') {
	if ($hours == 12) {
	    $hours = 12;          # 12pm = 12:00
	}
	elsif (1 <= $hours and $hours < 12) {
	    $hours = 12 + $hours; # 5pm = 17:00
	}
	else { die "bad number of hours $hours" }
    }
    else { die }

    my $r = sprintf('%02d:%02d', $hours, $mins);
    $r .= " $tz" if defined $tz;
    return $r;
}




# special_credits()
#
# Try to sniff out presenter, actor or guest info from the start of the
# description and put it into the credits section instead.
#
# Parameters: one programme (hashref)
# Returns: [ modified programme ], or undef
#
# May modify the programme passed in, if return value is not undef.
# But that's okay for a special-case handler.
#
sub special_credits( $ ) {
#    local $Log::TraceMessages::On = 1;
    die 'usage: special_credits(programme hashref)' if @_ != 1;
    my $prog = shift;
    t 'special_credits() ENTRY';

    # Caution: we need to make sure $_ is 'live' so updates to it
    # change the programme, when calling the extractors.
    #
    foreach my $pair (grep { pair_ok($_) } @{$prog->{desc}}) {
	die if not defined;
	t "testing desc: $pair->[0]";
	if (not length $pair->[0]) {
	    local $Log::TraceMessages::On = 1;
	    t 'programme with empty desc:' . d $prog;
	}

	if (s/\b([pP])resenteed\b/$1resented/g) {
	    t "fixing spelling mistake!";
	    return [ $prog ];
	}

	# Regexps to apply to the description (currently only the
	# first English-language description is matched).  The first
	# element is a subroutine which should alter $_ and return a
	# name or string of names if it succeeds, undef if it fails to
	# match.
	#
	# The first argument of the subroutine is the programme
	# itself, but this usually isn't used.  In any case, it should
	# not be modified except by altering $_.
	#
 	my @extractors =
 	  (
	   # Definitely presenter
	   [ sub {
		 s{(\b[a-z]\w+\s+)(?:(?:guest|virtual|new\s+)?presenters?)\s+($NAMES_RE)}{$1$2}o
		   && return $2;
		 s{((?:^|\.|\?)\s*)($NAMES_RE)\s+(?:(?:presents?)|(?:rounds?\s+up)|(?:introduces?))\b\s*(\.|,|\w|\Z)}
		   {$1 . uc $3}oe
		     && return $2;
		 s{Presenters?\s+($NAMES_RE)}{$1}o
		   && return $1;
		 s{,?\s*[cC]o-?presenters?\s+($NAMES_RE)}{}o
		   && return $1;
		 s{,?\s*[pP]resented by\s+($NAMES_RE)\b\s*(.|,?\s+\w|\Z)}{uc $2}oe
		   && return $1;
		 s{^\s*([hH]eadlines?(?:\s+\S+)?),?\s*[wW]ith\s+($NAMES_RE)\b(?:\.\s*)?}{$1}o
		   && return $2;
		 s{,?\s*(?:[iI]ntroduced|[cC]haired)\s+by\s+($NAMES_RE)(?:\.\s*)?}{}o
		   && return $1;

		 # This last one is special: it adds 'Last in series'
		 # which some other handler might pick up.
		 #
		 s{((?:^|\.|\?)\s*)($NAMES_RE)\s+concludes?\s+the\s+series\b\s*(?:with\b\s*)?(\.|,|\w|\Z)}
		   {$1 . 'Last in series.  ' . uc $3}oe
		     && return $2;

		 return undef;
	     }, 'presenter' ],
	   # Leave 'virtual presenter', 'aquatic presenter',
	   # 'new presenter' alone for now
	   #

	   # Might be presenter depending on type of show
 	   [ sub {
		 if (hasp($_[0], 'category',
			  sub { $_[0] =~ /\b(?:comedy|drama|childrens?)\b/i })
		     and not $prog->{credits}->{presenter}) {
		     return undef;
		 }
		 s{^\s*,?\s*[wW]ith\s+($NAMES_RE)\b(?:(?:\.\s*)?$)?}{}o
		   && return $1;
		 s{^\s*(?:[hH]ost\s+)?($NAME_RE) (?:introduces|conducts) (\w)(?![^.,;:!?]*\bto\b)}                    {uc $2}oe
		   && return $1;
		 s{^\s*(?:[hH]ost\s+)?($NAME_RE)\s+(?:explores|examines)\s*}{}o
		   && return $1;
		 return undef;
	     }, 'presenter' ],

	   [ sub {
		 s{((?:^|\.|\?)\s*)($NAME_RE)\s+interviews\b\s*(\.|,|\w|\Z)}{$1 . uc $3}oe
		   && return $2;
		 return undef;
	     }, 'presenter' ], # FIXME should be 'host' or 'interviewer'

	   # 'with' in quiz shows is guest (maybe)
	   [ sub {
		 return undef unless hasp($_[0], 'category',
					  sub { $_[0] =~ /\b(?:quiz|sports?)\b/i });
		 s{((?:^|,|\.|\?)\s*)[wW]ith\s*($NAMES_RE)\b(?!\s+among)(\.\s*\S)}
		   {$1 ne ',' ? "$1$2" : $2}oe
		     && return $2;
		 s{((?:^|,|\.|\?)\s*)[wW]ith\s*($NAMES_RE)\b(?!\s+among)(?:\.\s*$)?}
		   {$1 ne ',' ? $1 : ''}oe
		     && return $2;
		 return undef;
	     }, 'guest' ],

	   # 'with' in news/children shows is presenter (equally
	   # dubious).  Also a 'with' in a talk show might be
	   # presenter or might be guest, but at least we know it's
	   # not actor.
	   #
	   [ sub {
		 return undef
		   unless hasp($_[0], 'category',
			       sub { $_[0] =~ /\b(?:news|business|economics?|political|factual|talk|childrens?|game show)\b/i });
		 s{(?:^|,|\.|\?)\s*[wW]ith\s*($NAMES_RE)\b(?:\.\s*)?}{}o && return $1;
		 return undef;
	     }, 'presenter' ],

	   [ sub {
		 # Anything with a 'presenter' does not have actors.
		 return undef if $prog->{credits}->{presenter};
		 s{(?:[Ww]ith\s+)?[gG]uest\s+star\s+($NAMES_RE)\b\s*[,;.]?\s*}{}o
		   && return $1;
		 s{^($NAMES_RE) (?:co-)?stars? in\s+(?:this\s+)?}{uc $2}oe
		   && return $1;
		 s{\s*($NAMES_RE) (?:co-)?stars?\.?\s*$}{}o
		   && return $1;
		 s{(?:^|\.|\?)\s*($NAMES_RE)\s+(?:co-)?stars?\s+as\s*$}{}o
		   && return $1;
		 return undef;
	     }, 'actor' ],

	   [ sub {
		 # A discussion of 'a film starring Robin Williams'
		 # does not itself feature that actor.
		 #
		 return undef if $prog->{credits}->{presenter};
		 return undef if hasp($_[0], 'category', sub { $_[0] =~ /\barts\b/i });
		 s{(?:^|,|\.|\?)\s*[wW]ith\s*($NAMES_RE)\b(?:,|\.|;|$)?}{}o
		   && return $1;
		 s{,?\s*(?:(?:[Aa]lso|[Aa]nd)\s+)?(?:[Cc]o-|[Gg]uest-|[Gg]uest\s+)?[Ss]tarring\s+($NAMES_RE)\s*$}{}o
		   && return $1;
		 return undef;
	     }, 'actor' ],

	   [ sub {
		 s{,?\s*[wW]ith\s+guests?\s+($NAMES_RE)\b(?:\.\s*)?}{}o
		   && return $1;
		 s{((?:^|\.|!|\?)\s*)($NAME_RE)\s+guests(?:$|(?:\s+)|(?:.\s*))}{$1}o
		   && return $2;
		 return undef;
	     }, 'guest' ],

	   [ sub {
		 s{(?:^|\.|!|\?|,)(?:[Ww]ritten\s+)?\s*by\s+($NAMES_RE)\b($|\.)}{$2}o
		   && return $1;
		 return undef;
	     }, 'writer' ],
	  );

	# Run our own little hog-butchering algorithm to match each of
	# the subroutines in turn.
	#
	my $matched = 0;
	EXTRACTORS: foreach my $e (@extractors) {
	    my ($sub, $person) = @$e;
	    t "running extractor for role $person";
	    my $old_length = length $pair->[0];
	    my $match;
	    for ($pair->[0]) { $match = $sub->($prog) }
	    if (defined $match) {
		# Found one or more $person called $match.  We add them to
		# the list unless they're already in there.  We use a
		# per-programme cache of this information to avoid
		# going through the list each time (basically because
		# hashes are more Perlish).
		#
		t "got list of people: $match";
		my @names = list_names($match);
		t 'that is, names: ' . d \@names;
		t 'by shortening desc, programme updated to: ' . d $prog;
		for my $credits ($prog->{credits}) {
		    my %seen;
		    if (lc $person eq 'guest') {
			# Impossible for someone to be guest as well
			# as another part, so don't add it if already
			# listed anywhere.
			#
			foreach (keys %$credits) {
			    $seen{$_}++ foreach @{$credits->{$_}};
			}
		    }
		    else {
			# Cannot add this person if they are already
			# given in the same job, or as a guest.
			#
			foreach (@{$credits->{$person}}, @{$credits->{guest}}) {
			    $seen{$_}++ && warn "person $_ seen twice";
			}
		    }

		    t 'people already known (or ineligible): ' . d \%seen;
		    foreach (@names) {
			t "maybe adding $_ as a $person";
			push @{$credits->{$person}}, $_ unless $seen{$_}++;
		    }
		    t '$credits->{$person}=' . d $credits->{$person};
		}

		if (length $pair->[0] >= $old_length) {
		    warn "extractor failed to shorten text: now $pair->[0]";
		}

		t 'by adding people, programme updated to: ' . d $prog;
		$matched = 1;
		goto EXTRACTORS; # start again from beginning of loop
	    }
	}
	if ($matched) {
	    t 'some handlers matched, programme: ' . d $prog;
	    scrub_empty($prog->{desc});
	    t 'after removing empty things, programme: ' . d $prog;
            return [ $prog ];
        }
    }
    return undef;
}
# has()
#
# Check whether some attribute of a programme matches a particular
# string.  For example, does the programme have the category 'quiz'?
# This means checking all categories of acceptable language.
#
#   has($programme, 'category', 'quiz');
#
sub has( $$@ ) {
#    local $Log::TraceMessages::On = 1;
    my ($p, $attr, @allowed) = @_;
    t 'testing whether programme: ' . d $p;
    t "has attribute $attr in the list: " . d \@allowed;
    my $list = $p->{$attr};
    t 'all [text, lang] pairs for this attr: ' . d $list;
    return 0 if not defined $list;
    foreach (grep { pair_ok($_) } @$list) {
	my ($text, $lang) = @$_;
	foreach (@allowed) {
	    t "testing if $text matches $_ (nocase)";
	    return 1 if lc $text eq $_;
	}
    }
    t 'none of them matched, returning false';
    return 0;
}
# hasp()
#
# Like has() but instead of a list of strings to compare against,
# takes a subroutine reference.  This subroutine will be run against
# all the text strings of suitable language in turn until it matches
# one, when true is returned.  If none match, returns false.
#
# Parameters:
#   ref to programme hash
#   name of key to look under
#   subroutine to apply to each value of key with acceptable language
#
# Returns: whether subroutine gives true for at least one value.
#
# The subroutine will get the text value passed in $_[0].
#
sub hasp( $$$ ) {
#    local $Log::TraceMessages::On = 1;
    my ($p, $attr, $sub) = @_;
    die "expected programme hash as first argument, not $p"
      if ref $p ne 'HASH';
    t 'testing whether programme: ' . d $p;
    t "has a value for attribute $attr that makes sub return true";

    # FIXME commonize this with has().
    my $list = $p->{$attr};
    t 'all [text, lang] pairs for this attr: ' . d $list;
    return 0 if not defined $list;
    foreach (grep { pair_ok($_) } @$list) {
	my ($text, $lang) = @$_;
	t "testing if $text matches";
	return 1 if $sub->($text);
    }
    t 'none of them matched, returning false';
    return 0;
}


# special_new_series()
#
# Contrary to first appearances, the <new /> element in the XML isn't
# to indicate a new series - it means something stronger, a whole new
# show (not a new season of an existing show).  But you can represent
# part of the meaning of 'new series' within the episode-num
# structure, because obviously a new series means that this is the
# first episode of the current series.
#
# This handler is mostly here to get rid of the 'New series' text at
# the start of descriptions, to try and make output from different
# grabbers look the same.
#
sub special_new_series( $ ) {
    die 'usage: special_new_series(programme)' if @_ != 1;
    my $p = shift;

    # Just assume that if it contains 'New series' at the start then
    # it's English.
    #
    my $is_new_series = 0;
    foreach (@{$p->{desc}}) {
	for ($_->[0]) {
	    if (s/^New series(?:\.\s*|$)//
		or s/^New series (?:of (?:the )?)?(\w)/uc $1/e
	       ) {
		$is_new_series = 1;
	    }
	}
    }

    return undef if not $is_new_series;
    if (defined $p->{'episode-num'}) {
	foreach (@{$p->{'episode-num'}}) {
	    my ($content, $system) = @$_;
	    next unless $system eq 'xmltv_ns';
	    $content =~ m!^\s*(\d+/\d+|\d+|)\s*\.\s*(\d+/\d+|\d+|)\s*\.\s*(\d+/\d+|\d+|)\s*$!
		or warn("badly formed xmltv_ns episode-num: $content"), return [ $p ];
	    my ($season, $episode, $part) = ($1, $2, $3);
	    if ($episode ne '' and $episode !~ /^0/) {
		warn "new series, but episode number $episode";
	    }
	    elsif ($episode eq '') {
		# We now know the information that this is the first
		# episode of the series.
		#
		$episode = '0';
		$content = "$season . $episode . $part";
		$_ = [ $content, $system ];
		last;
	    }
	}
    }
    else {
	# Make a dummy episode num which says nothing other than
	# this is the first episode of the series.
	#
	$p->{'episode-num'} = [ [ ' . 0 . ', 'xmltv_ns' ] ];
    }
    scrub_empty($p->{desc});
    return [ $p ];
}


# special_year(): take a year at the start of the description and move
# it to the 'date' field.
#
sub special_year( $ ) {
    die 'usage: special_new_series(programme)' if @_ != 1;
    my $p = shift;

    my $year;
    foreach (@{$p->{desc}}) {
	if ($_->[0] =~ s/^(\d{4})\s+//) {
	    my $got = $1;
	    if (defined $year and $got ne $year) {
		warn "found different years $year and $got";
		return [ $p ];
	    }
	    $year = $got;
	}
    }
    return undef if not defined $year;
    if (defined $p->{date}) {
	if ($p->{date} !~ /^\s*$year/) {
	    warn "found year $year in programme description, but date $p->{date}";
	}
    }
    else {
	$p->{date} = $year;
    }
    scrub_empty($p->{desc});
    return [ $p ];
}


# 'TVM' at start of description means TV movie.
sub special_tv_movie( $ ) {
    die 'usage: special_tv_movie(programme)' if @_ != 1;
    my $p = shift;
    my $is_tv_movie = 0;
    foreach (@{$p->{desc}}) {
	my $lang = $_->[1];
	if (not defined $lang or $lang =~ /^en/) {
	    if ($_->[0] =~ s/^TVM\b\s*//) {
		$is_tv_movie = 1;
	    }
	}
    }
    return undef if not $is_tv_movie;
    add($p, 'category', 'TV movie');
    scrub_empty($p->{desc});
    return [ $p ];
}


# '(T)' in description means teletext subtitles.  But this should run
# after doing any splitting and other stuff.
#
sub special_teletext_subtitles( $ ) {
    die 'usage: special_teletext_subtitles(programme)' if @_ != 1;
    my $p = shift;
    my $has_t = 0;
    foreach (@{$p->{desc}}) {
	if ($_->[0] =~ s/\s*\(T\)\s*$//) {
	    $has_t = 1;
	}
    }
    return undef if not $has_t;
    if (defined $p->{subtitles}) {
	foreach (@{$p->{subtitles}}) {
	    return [ $p ] if defined $_->{type} and $_->{type} eq 'teletext';
	}
    }
    push @{$p->{subtitles}}, { type => 'teletext' };
    scrub_empty($p->{desc});
    return [ $p ];
}


# Remove stock phrases that have no meaning.
sub special_useless( $ ) {
    die 'usage: special_useless(programme)' if @_ != 1;
    my $p = shift;

    # FIXME need to commonize hog-butchering with special_credits().
    my $changed = 0;
    foreach (@{$p->{desc}}) {
	for ($_->[0]) {
	    $changed |= s/^(?:a\s+|)round-up\s+of\s+(\w)/uc $1/ie;
	    $changed |= s/^(\w+[^s])\s+round-up\.?\s*$/$1 . 's'/ie;
	    $changed |= s/((?:^|\.|!|\?)\s*)Coverage\s+of\s+(\w)/$1 . uc $2/e;
	}
    }

    return [ $p ] if $changed;
    return undef;
}


# special_radio4()
#
# Split Radio 4 into FM and LW.
#
sub special_radio4( $ ) {
    die 'usage: special_radio4(programme)' if @_ != 1;
    my $p = shift;
    return undef if $p->{channel} ne 'radio4';

    for ($p->{title}) {
	if (s/^\(FM\)\s+//) {
	    $p->{channel} = 'radio4-fm';
	    return [ $p ];
	}
	if (s/^\(LW\)\s+//) {
	    $p->{channel} = 'radio4-lw';
	    return [ $p ];
	}

	my %fm = ( %$p, channel => 'radio4-fm' );
	my %lw = ( %$p, channel => 'radio4-lw' );
	return [ \%fm, \%lw ];
    }
}


# special_split_title()
#
# In addition to the 'programme tacked onto the end of another'
# handled by add_clumpidx, you also sometimes see two programmes
# totally sharing an entry.  For example 'News; Shipping Forecast'.
#
sub special_split_title( $ ) {
    die 'usage: special_split_title(programme)' if @_ != 1;
    my $p = shift;
    return undef if $p->{title} !~ tr/;//;

    # Split the title at ; and make N identical programmes one with
    # each title.  The programme details are given to only the last of
    # the programmes - in the listings data we're getting, normally
    # the insignificant programme comes first with the main feature
    # last, as in 'News; Radio 3 Lunchtime Concert'.
    #

    # List of elements which are meta-data and should be kept for all
    # the programmes we split into - the rest are given only to the
    # last programme.
    #
    my %meta = (start => 1, stop => 1, 'pdc-start' => 1,
		'vps-start' => 1, showview => 1, videoplus => 1,
		channel => 1);
    # but not clumpidx!

    my %p_meta;
    foreach (grep { $meta{$_} } keys %$p) {
	$p_meta{$_} = $p->{$_};
    }

    my @r;
    my @titles = split /\s*;+\s*/, $p->{title};
    for (my $i = 0; $i < @titles - 1; $i++) {
	push @r, { %p_meta,
		title => $titles[$i],
		clumpidx => ( "$i/" . scalar @titles ) };
    }
    push @r, { %$p,
		title => $titles[-1],
		clumpidx => ("$#titles/" . scalar @titles) };

    return \@r;
}


# special_film()
#
sub special_film( $ ) {
    die 'usage: special_film(programme)' if @_ != 1;
    my $p = shift;
    if (not defined $p->{'sub-title'} or $p->{'sub-title'} ne '(Film)') {
	return undef;
    }

    warn "replacing category $p->{category} with 'film'"
      if defined $p->{category};
    $p->{category} = 'film';
    undef $p->{'sub-title'};

    if (defined $p->{desc} and $p->{desc} =~ s/^(\d{4})\s+//) {
	warn "found year $1 in description, replacing date $p->{date}"
	  if defined $p->{date};
	$p->{date} = $1;
    }

    return [ $p ];
}


# add()
#
# Add a piece of human-readable information to a particular slot, but
# only if it isn't there already.  For example add the category
# 'music', but only if that category isn't already set.  This is for
# keys that take multiple values and each value is a [ text, lang ]
# pair.  The language is assumed to be English.
#
# Parameters:
#   programme hash to add to
#   name of key
#   textual value to add
#
# Returns: whether the programme was altered.
#
sub add( $$$ ) {
    my ($p, $k, $v) = @_;
    if (defined $p->{$k}) {
	foreach (@{$p->{$k}}) {
	    return 0 if $_->[0] eq $v;
	}
    }
    push @{$p->{$k}}, [ $v, $LANG ];
    return 1;
}


# scrub_empty(): remove empty strings from a list of [text, lang]
# pairs.
#
# Parameters: zero or more listrefs
#
# Modifies lists passed in, removing all [ '', whatever ] pairs.
#
sub scrub_empty( @ ) {
    foreach (@_) {
	@$_ = grep { length $_->[0] } @$_;
    }
}


# Make sure that a programme altered by a special handler does not
# spill outside its alotted timespan.  This is just a sanity check
# before fix_clumps() does its stuff.  In a future version we might
# remove this restriction and allow special handlers to move
# programmes outside their original timeslot.
#
# Parameters:
#   original programme
#   ref to list of new programmes
#
sub check_time_fits( $$ ) {
    my $orig = shift;
    my @new = @{shift()};
    my $o_start = pd($orig->{start});
    die if not defined $o_start;
    my $o_stop = pd($orig->{stop});
    foreach (@new) {
	my $start = pd($_->{start});
	die if not defined $start;
	if (Date_Cmp($start, $o_start) < 0) {
	    die "programme starting at $o_start was split into one starting at $start";
	}

	if (defined $o_stop) {
	    my $stop = pd($_->{stop});
	    if (defined $stop and Date_Cmp($o_stop, $stop) < 0) {
		die "programme stopping at $o_stop was split into one stopping at $stop";
	    }
	}
    }
}



# Another sanity check.
sub check_same_channel( $ ) {
    my $progs = shift;
    my $ch;
    foreach my $prog (@$progs) {
	for ($prog->{channel}) {
	    if (not defined) {
		t 'no channel! ' . d $prog;
		croak 'programme has no channel';
	    }
	    if (not defined $ch) {
		$ch = $_;
	    }
	    elsif ($ch eq $_) {
		# Okay.
	    }
	    else {
		# Cannot use croak() due to this error message:
		#
		# Bizarre copy of ARRAY in aassign at /usr/lib/perl5/5.6.0/Carp/Heavy.pm line 79.
		#
		local $Log::TraceMessages::On = 1;
		t 'same clump, different channels: ' . d($progs->[0]) . ' and ' . d($prog);
		die "programmes in same clump have different channels: $_, $ch";
	    }
	}
    }
}


# There is a very hard to track down bug where stop times mysteriously
# get set to something ridiculous.  It varies from one perl version to
# another (hence the version check at the top) but still occurs even
# with 5.6.1.  To track it down I have isolated all code that sets
# stop times in this subroutine.
#
sub set_stop_time( $$ ) {
    my $p = shift;
    my $s = shift;

    if ($SLOW) {
	# Another mysterious-bug-preventing line, see no_shared_scalars().
	my $dummy = "$s";

	$p->{stop} = $s;
    }
    else {
	# Don't set stop times at all.
	delete $p->{stop};
    }
}

# More debugging aids.
sub cst( $ ) {
    my $p = shift;
    croak "prog $p->{title}->[0]->[0] has bogus stop time"
      if exists $p->{stop} and $p->{stop} eq 'boogus FIXME XXX';
}

sub no_shared_scalars( $ ) {
    my %seen;
    foreach my $h (@{$_[0]}) {
	foreach my $k (keys %$h) {
	    my $ref = \ ($h->{$k});
	    my $addr = "$ref";
	    $seen{$addr}++ && die "scalar $addr seen twice";
	}
    }
}