#!/usr/bin/perl -w
#
# Written with reference to pandoc_markdown from Debian jessie
# We require atx-style headers
#
# usage:
#   pandoc -t json SUPPORT.md >j-unstable
#   git cat-file blob origin/staging-4.11:SUPPORT.md | pandoc -t json >j-4.11
#   docs/parse-support-md \
#            j-unstable https://xenbits/unstable/SUPPORT.html
#            j-4.11 https://xenbits/4.11/SUPPORT.html
# or equivalent

use strict;
use JSON;
use Tie::IxHash;
use IO::File;
use CGI qw(escapeHTML);
use Data::Dumper;
use POSIX;

#---------- accumulating input/output ----------

# This combines information from all of the input files.

sub new_sectlist () { { } };
our $toplevel_sectlist = new_sectlist();
# an $sectlist is
#   { }                 nothing seen yet
#   a tied hashref      something seen
# (tied $sectlist)    is an object of type Tie::IxHash
# $sectlist->{KEY} a $sectnode:
# $sectlist->{KEY}{Status}[VI] = absent or string or markdown content
# $sectlist->{KEY}{Children} = a further $sectlist
# $sectlist->{KEY}{Key} = KEY
# $sectlist->{KEY}{RealSectNode} = us, or our parent
# $sectlist->{KEY}{RealSectNode}{HasCaveat}[VI] = trueish iff other in a Para
# $sectlist->{KEY}{RealInSect} = containing real section in @insections, so
# $sectlist->{KEY}{RealInSect}{HasDescription} = VI for some Emph in Para
# $sectlist->{KEY}{RealInSect}{Anchor} = value for < id="" > in the pandoc html
# A $sectnode represents a single section from the original markdown
# document.  Its subsections are in Children.
#
# Also, the input syntax:
#    Status, something or other: Supported
# is treated as a $sectnode, is as if it were a subsection -
# one called `something or other'.  That is not a `real' section.
#
# KEY is the Anchor, or derived from the `something or other'.
# It is used to match up identical features in different versions.

#---------- state for this input file ----------

our $version_index;
our @version_urls;

our @insections;
# $insections[]{Key} = string
# $insections[]{Headline} = markdown content
# these next are only defined for real sections, not Status elements
# $insections[]{Anchor} = string
# $insections[]{HasDescription} VI, likewise

our $had_unknown;
our $had_feature;
# adding new variable ?  it must be reset in r_toplevel

#---------- parsing ----------

sub find_current_sectnode () {
    die unless @insections;

    my $sectnode;
    my $realinsect;
    my $realsectnode;
    foreach my $s (@insections) {
        my $sectlist = $sectnode
            ? $sectnode->{Children} : $toplevel_sectlist;
        my $key = $s->{Key};
        $realinsect = $s if $s->{Anchor};
        tie %$sectlist, 'Tie::IxHash' unless tied %$sectlist;
#print STDERR "FIND_CURRENT_SECTNODE ", Dumper($s);
        $sectlist->{$key} //=
            {
             Children => new_sectlist(),
             Headline => $s->{Headline},
             Key => $key,
             RealInSect => $realinsect,
             HasCaveat => [],
            };
        $sectnode = $sectlist->{$key};
        $realsectnode = $sectnode if $s->{Anchor};
        $sectnode->{RealSectNode} = $realsectnode;
    }
    die unless $sectnode;
    return $sectnode;
}

sub ri_Header {
    my ($c) = @_;
    my ($level, $infos, $hl) = @$c;
#print STDERR 'RI_HEADER ', Dumper($c, \@c);
    my ($id) = @$infos;
    die unless $level >= 1;
    die unless $level-2 <= $#insections;
    $#insections = $level-2;
    push @insections,
        {
         Key => $id,
         Anchor => $id,
         Headline => $hl,
         HasDescription => undef,
        };
#print STDERR Dumper(\@insections);
    $had_feature = 0;
}

sub ri_Para {
    return unless @insections;
    my $insection = $insections[$#insections];
#    print DEBUG "ri_Para ",
#        Dumper($version_index, $had_feature, $insection);

    if ($had_feature) {
        my $sectnode = find_current_sectnode();
        $sectnode->{RealSectNode}{HasCaveat}[$version_index] = 1;
    } else {
        $insection->{HasDescription} //= $version_index;
    }
};

sub parse_feature_entry ($) {
    my ($value) = @_;

    $had_feature = 1;
    my $sectnode = find_current_sectnode();
    $sectnode->{Status}[$version_index] = $value;
}

sub descr2key ($) {
    my ($descr) = @_;

    die unless @insections;
    my $insection = $insections[$#insections];

    my $key = lc $descr;
    $key =~ y/ /-/;
    $key =~ y/-0-9A-Za-z//cd;
    $key = $insection->{Anchor}.'--'.$key;
    return $key;
}

sub ri_CodeBlock {
    my ($c) = @_;
    my ($infos, $text) = @$c;

    if ($text =~ m{^(?: Functional\ completeness 
                   | Functional\ stability
                   | Interface\ stability
                   | Security\ supported ) \:}x) {
        # ignore this
        return;
    }
    die "$had_unknown / $text ?" if $had_unknown;

    my $toplevel = $text =~ m{^Xen-Version:};

    foreach my $l (split /\n/, $text) {
        $l =~ s/\s*$//;
        next unless $l =~ m/\S/;

        my ($descr, $value) =
            $toplevel
            ? $l =~ m{^([A-Z][-A-Z0-9a-z]+)\:\s+(\S.*)$}
            : $l =~ m{^(?:Status|Supported)(?:\,\s*([^:]+))?\:\s+(\S.*)$}
            or die ("$text\n^ cannot parse status codeblock line:".
                    ($toplevel and 'top').
                    "\n$l\n ?");

        if (length $descr) {
            push @insections,
                {
                 Key => descr2key($descr),
                 Headline => [{ t => 'Str', c => $descr }],
                };
        }
        parse_feature_entry $value;
        if (length $descr) {
            pop @insections;
        }
    }
}

sub ri_DefinitionList {
    my ($c) = @_;
    foreach my $defent (@$c) {
        my ($term, $defns) = @$defent;
        my $descr =
            join ' ',
            map { $_->{c} }
            grep { $_->{t} eq 'Str' }
            @$term;
        push @insections,
            {
             Key => descr2key($descr),
             Headline => $term,
            };
        die "multiple definitions in definition list definition"
            if @$defns > 1;
        my $defn = $defns->[0];
        die "multiple paras in definition list definition"
            if @$defn > 1;
        my $defnp = $defn->[0];
        die "only understand plain definition not $defnp->{t} ?"
            unless $defnp->{t} eq 'Plain';
        parse_feature_entry $defnp->{c};
        pop @insections;
    }
}

sub process_unknown {
    my ($c, $e) = @_;
    $had_unknown = Dumper($e);
}

sub r_content ($) {
    my ($i) = @_;
    foreach my $e (@$i) {
        my $f = ${*::}{"ri_$e->{t}"};
        $f //= \&process_unknown;
        $f->($e->{c}, $e);
    }
}

our $pandoc_toplevel_constructor;

sub r_toplevel ($) {
    my ($i) = @_;

    die unless defined $version_index;

    @insections = ();
    $had_unknown = undef;
    $had_feature = undef;

    # Pandoc's JSON output changed some time between 1.17.2 (stretch)
    # and 2.2.1 (buster).  I can find no documentation about this
    # change or about the compatibility rules.  (It seems that
    # processing the parse tree *is* supported upstream: they offer
    # many libraries to do this inside the pandoc process.)
    # Empirically, what has changed is just the top level structure.
    # Also pandoc wants the same structure back that it spat out,
    # when we ask it to format snippets.

    my $blocks;
    if (ref $i eq 'ARRAY') {
	$pandoc_toplevel_constructor = sub {
	    my ($blocks) = @_;
	    return [
		    { unMeta => { } },
		    $blocks,
		   ];
	};
	foreach my $e (@$i) {
	    next unless ref $e eq 'ARRAY';
	    r_content $e;
	}
    } elsif (ref $i eq 'HASH') {
	my $api_version = $i->{'pandoc-api-version'};
	$pandoc_toplevel_constructor = sub {
	    my ($blocks) = @_;
	    return {
		    blocks => $blocks,
		    meta => { },
		    'pandoc-api-version' => $api_version,
		   };
	};
	r_content $i->{blocks};
    } else {
	die;
    }
}

sub read_inputs () {
    $version_index = 0;

    local $/;
    undef $/;

    while (my $f = shift @ARGV) {
        push @version_urls, shift @ARGV;
        eval {
            open F, '<', $f or die $!;
            my $input_toplevel = decode_json <F>;
            r_toplevel $input_toplevel;
        };
        die "$@\nwhile processing input file $f\n" if $@;
        $version_index++;
    }
}

#---------- reprocessing ----------

# variables generated by analyse_reprocess:
our $maxdepth;

sub pandoc2html_inline ($) {
    my ($content) = @_;

    my $json_fh = IO::File::new_tmpfile or die $!;

    my $blocks = [{ t => 'Para', c => $content }];
    my $data = $pandoc_toplevel_constructor->($blocks);
    my $j = to_json($data) or die $!;
    print $json_fh $j;
    flush $json_fh or die $!;
    seek $json_fh,0,0 or die $!;

    my $c = open PD, "-|" // die $!;
    if (!$c) {
        open STDIN, "<&", $json_fh;
        exec qw(pandoc -f json) or die $!;
    }

    local $/;
    undef $/;
    my $html = <PD>;
    $?=$!=0;
    if (!close PD) {
        eval {
            seek $json_fh,0,0 or die $!;
            open STDIN, '<&', $json_fh or die $!;
            system 'json_pp';
        };
        die "$j \n $? $!";
    }

    $html =~ s{^\<p\>}{} or die "$html ?";
    $html =~ s{\</p\>$}{} or die "$html ?";
    $html =~ s{\n$}{};
    return $html;
}

sub reprocess_sectlist ($$);

sub reprocess_sectnode ($$) {
    my ($sectnode, $d) = @_;

    $sectnode->{Depth} = $d;

    if ($sectnode->{Status}) {
        $maxdepth = $d if $d > $maxdepth;
    }

    if ($sectnode->{Headline}) {
#            print STDERR Dumper($sectnode);
        $sectnode->{Headline} =
            pandoc2html_inline $sectnode->{Headline};
    }

    reprocess_sectlist $sectnode->{Children}, $d;
}

sub reprocess_sectlist ($$) {
    my ($sectlist, $d) = @_;
    $d++;

    foreach my $sectnode (values %$sectlist) {
        reprocess_sectnode $sectnode, $d;
    }
}

sub count_rows_sectlist ($);

sub count_rows_sectnode ($) {
    my ($sectnode) = @_;
    my $rows = 0;
    $sectnode->{RealInSect}{OwnRows} //= 0;
    if ($sectnode->{Status}) {
        $rows++;
        $sectnode->{RealInSect}{OwnRows}++;
    }
    $rows += count_rows_sectlist $sectnode->{Children};
    $sectnode->{Rows} = $rows;
    $sectnode->{RealInSect}{Rows} = $rows;
    return $rows;
}

# Now we have
#   $sectnode->{Rows}
#   $sectnode->{RealInSect}{Rows}
#   $sectnode->{RealInSect}{OwnRows}

sub count_rows_sectlist ($) {
    my ($sectlist) = @_;
    my $rows = 0;
    foreach my $sectnode (values %$sectlist) {
        $rows += count_rows_sectnode $sectnode;
    }
    return $rows;
}

# After reprocess_sectlist,
#    ->{Headline}   is in html
#    ->{Status}     is (still) string or markdown content

sub analyse_reprocess () {
    $maxdepth = 0;
    reprocess_sectlist $toplevel_sectlist, 0;
}

#---------- output ----------

sub o { print @_ or die $!; }

our @pending_headings;

sub docref_a ($$) {
    my ($i, $realinsect) = @_;
    return sprintf '<a href="%s#%s">',
        $version_urls[$i], $realinsect->{Anchor};
}

sub write_output_row ($) {
    my ($sectnode) = @_;
#    print STDERR 'WOR ', Dumper($d, $sectnode);
    o('<tr>');
    my $span = sub {
        my ($rowcol, $n) = @_;
        o(sprintf ' %sspan="%d"', $rowcol, $n) if $n != 1;
    };
    # This is all a bit tricky because (i) the input is hierarchical
    # with variable depth, whereas the output has to have a fixed
    # number of heading columns on the LHS; (ii) the HTML
    # colspan/rowspan system means that when we are writing out, we
    # have to not write table elements for table entries which have
    # already been written with a span instruction that covers what we
    # would write now.
    while (my $heading = shift @pending_headings) {
        o('<th valign="top"');
        o(sprintf ' id="%s"', $heading->{Key});
        $span->('row', $heading->{Rows});
        $span->('col', $maxdepth - $heading->{Depth} + 1)
            if !%{ $heading->{Children} };
        o(' align="left">');
        my $end_a = '';
        my $desc_i = $heading->{RealInSect}{HasDescription};
        if (defined $desc_i) {
            o(docref_a $desc_i, $heading->{RealInSect});
            $end_a= '</a>';
        }
        o($heading->{Headline});
        o($end_a);
        o('</th>');
    }
    if (%{ $sectnode->{Children} }) {
        # we suppressed the colspan above, but we do need to make the gap
        my $n = $maxdepth - $sectnode->{Depth};
        die 'XX '. Dumper($n, $sectnode) if $n<0;
        if ($n) {
            o('<td');
            $span->('col', $n);
            o('></td>');
        }
    }
    for (my $i=0; $i<@version_urls; $i++) {
        my $st = $sectnode->{Status}[$i];

        my $colspan = $sectnode->{RealInSect}{ColSpan}[$i];
        my $nextcell = '';
        if (!defined $colspan) { # first row of this RealInSect
            $colspan= ' colspan="2"';
            if ($sectnode->{RealSectNode}{HasCaveat}[$i] && $st
                && $sectnode->{RealInSect}{Anchor}) {
                my $rows = $sectnode->{RealInSect}{OwnRows};
                $nextcell = '<td';
                $nextcell .= sprintf ' rowspan=%d', $rows if $rows>1;
                $nextcell .= '>';
                $nextcell .= docref_a $i, $sectnode->{RealInSect};
                $nextcell .= '[*]</a>';
                $nextcell .= '</td>';
                $colspan = '';
            }
            $sectnode->{RealInSect}{ColSpan}[$i] = $colspan;
        }

        $st //= '-';
        o("\n<td$colspan>");
        my $end_a = '';
        if ($sectnode->{Key} eq 'release-support--xen-version') {
            o(sprintf '<a href="%s">', $version_urls[$i]);
            $end_a = '</a>';
        }
        if (ref $st) {
            $st = pandoc2html_inline $st;
        } else {
            $st = escapeHTML($st);
        }
        o($st);
        o($end_a);
        o('</td>');
        o($nextcell);
    }
    o("</tr>\n");
}      

sub write_output_sectlist ($);
sub write_output_sectlist ($) {
    my ($sectlist) = @_;
    foreach my $key (keys %$sectlist) {
        my $sectnode = $sectlist->{$key};
        push @pending_headings, $sectnode;
        write_output_row $sectnode if $sectnode->{Status};
        write_output_sectlist $sectnode->{Children};
    }
}

sub write_output () {
    o('<table rules="all">');
    write_output_sectlist $toplevel_sectlist;
    o('</table>');
}

#---------- main program ----------

open DEBUG, '>', '/dev/null' or die $!;
if (@ARGV && $ARGV[0] eq '-D') {
    shift @ARGV;
    open DEBUG, '>&2' or die $!;
}

die unless @ARGV;
die if $ARGV[0] =~ m/^-/;
die if @ARGV % 2;

read_inputs();

#use Data::Dumper;
#print DEBUG Dumper($toplevel_sectlist);

analyse_reprocess();
# Now Headline is in HTML

count_rows_sectlist($toplevel_sectlist);

#use Data::Dumper;
print DEBUG Dumper($toplevel_sectlist);

write_output();