autodoc.pl

#!/usr/bin/perl -w

use Text::Tabs;
#
# Unconditionally regenerate:
#
#    pod/perlintern.pod
#    pod/perlapi.pod
#
# from information stored in
#
#    embed.fnc
#    plus all the core .c, .h, and .pod files listed in MANIFEST
#
# Has an optional arg, which is the directory to chdir to before reading
# MANIFEST and the files
#
# This script is invoked as part of 'make all'
#
# The generated pod consists of sections of related elements, functions,
# macros, and variables.  The keys of %valid_sections give the current legal
# ones.  Just add a new key to add a section.
#
# Throughout the files read by this script are lines like
#
# =for apidoc_section Section Name
#
# "Section Name" (after having been stripped of leading space) must be one of
# the legal section names, or an error is thrown.  All API elements defined
# between this line and the next 'apidoc_section' line will go into the
# section "Section Name", sorted by dictionary order within it.  perlintern
# and perlapi are parallel documents, each potentially with a section "Section
# Name".  Each element is marked as to which document it goes into.  If there
# are none for a particular section in perlapi, that section is
# omitted.
#
# Also, in .[ch] files, there may be
#
# =head1 Section Name
#
# lines in comments.  These are also used by this program to switch to section
# "Section Name".  The difference is that if there are any lines after the
# =head1, inside the same comment, and before any =for apidoc-ish lines, they
# are used as a heading for section "Section Name" (in both perlintern and
# perlapi).  This includes any =head[2-5].  If more than one '=head1 Section
# Name' line has content, they appear in the generated pod in an undefined
# order.
#
# The next =head1, =for apidoc_section, or file end terminates what goes into
# the current section
#
# The %valid_sections hash below also can have header content, which will
# appear before any =head1 content.  The hash can also have footer content
# content, which will appear at the end of the section, after all the
# elements.
#
# The lines that define the actual functions, etc are documented in embed.fnc,
# because they have flags which must be kept in sync with that file.

use strict;
use warnings;

# 80 column terminal - 1 for pager adding a column; -7 for nroff
# indent;
my $max_width = 80 - 1 - 7;

if (@ARGV) {
    my $workdir = shift;
    chdir $workdir
        or die "Couldn't chdir to '$workdir': $!";
}
require './regen/regen_lib.pl';
require './regen/embed_lib.pl';

my %described_elsewhere;

#
# See database of global and static function prototypes in embed.fnc
# This is used to generate prototype headers under various configurations,
# export symbols lists for different platforms, and macros to provide an
# implicit interpreter context argument.
#

my %docs;
my %seen;
my %funcflags;
my %missing;
my %missing_macros;

my $link_text = "Described in";

my $description_indent = 4;
my $usage_indent = 3;   # + initial blank yields 4 total

my $av_scn = 'AV Handling';
my $callback_scn = 'Callback Functions';
my $casting_scn = 'Casting';
my $casing_scn = 'Character case changing';
my $classification_scn = 'Character classification';
my $names_scn = 'Character names';
my $scope_scn = 'Compile-time scope hooks';
my $compiler_scn = 'Compiler and Preprocessor information';
my $directives_scn = 'Compiler directives';
my $concurrency_scn = 'Concurrency';
my $COP_scn = 'COP Hint Hashes';
my $CV_scn = 'CV Handling';
my $custom_scn = 'Custom Operators';
my $dump_scn = 'Display and Dump functions';
my $embedding_scn = 'Embedding and Interpreter Cloning';
my $errno_scn = 'Errno';
my $exceptions_scn = 'Exception Handling (simple) Macros';
my $filesystem_scn = 'Filesystem configuration values';
my $floating_scn = 'Floating point configuration values';
my $formats_scn = 'Formats';
my $genconfig_scn = 'General Configuration';
my $globals_scn = 'Global Variables';
my $GV_scn = 'GV Handling';
my $hook_scn = 'Hook manipulation';
my $HV_scn = 'HV Handling';
my $io_scn = 'Input/Output';
my $integer_scn = 'Integer configuration values';
my $lexer_scn = 'Lexer interface';
my $locales_scn = 'Locales';
my $magic_scn = 'Magic';
my $memory_scn = 'Memory Management';
my $mro_scn = 'MRO';
my $multicall_scn = 'Multicall Functions';
my $numeric_scn = 'Numeric Functions';
my $optree_construction_scn = 'Optree construction';
my $optree_manipulation_scn = 'Optree Manipulation Functions';
my $pack_scn = 'Pack and Unpack';
my $pad_scn = 'Pad Data Structures';
my $password_scn = 'Password and Group access';
my $paths_scn = 'Paths to system commands';
my $intrpvar_scn = 'Per-Interpreter Variables';
my $prototypes_scn = 'Prototype information';
my $regexp_scn = 'REGEXP Functions';
my $signals_scn = 'Signals';
my $site_scn = 'Site configuration';
my $sockets_scn = 'Sockets configuration values';
my $filters_scn = 'Source Filters';
my $stack_scn = 'Stack Manipulation Macros';
my $string_scn = 'String Handling';
my $SV_flags_scn = 'SV Flags';
my $SV_scn = 'SV Handling';
my $time_scn = 'Time';
my $typedefs_scn = 'Typedef names';
my $unicode_scn = 'Unicode Support';
my $utility_scn = 'Utility Functions';
my $versioning_scn = 'Versioning';
my $warning_scn = 'Warning and Dieing';
my $xs_scn = 'XS';

my %valid_sections = (
    $av_scn => {},
    $callback_scn => {},
    $casting_scn => {},
    $casing_scn => {},
    $classification_scn => {},
    $scope_scn => {},
    $compiler_scn => {},
    $directives_scn => {},
    $concurrency_scn => {},
    $COP_scn => {},
    $CV_scn => {},
    $custom_scn => {},
    $dump_scn => {},
    $embedding_scn => {},
    $errno_scn => {},
    $exceptions_scn => {},
    $filesystem_scn => {
        header => <<~'EOT',
            Also see L</List of capability HAS_foo symbols>.
            EOT
        },
    $floating_scn => {
        header => <<~'EOT',
            Also L</List of capability HAS_foo symbols> lists capabilities
            that arent in this section.  For example C<HAS_ASINH>, for the
            hyperbolic sine function.
            EOT
        },
    $formats_scn => {
        header => <<~'EOT',
            These are used for formatting the corresponding type For example,
            instead of saying

             Perl_newSVpvf(pTHX_ "Create an SV with a %d in it\n", iv);

            use

             Perl_newSVpvf(pTHX_ "Create an SV with a " IVdf " in it\n", iv);

            This keeps you from having to know if, say an IV, needs to be
            printed as C<%d>, C<%ld>, or something else.
            EOT
      },
    $genconfig_scn => {
        header => <<~'EOT',
            This section contains configuration information not otherwise
            found in the more specialized sections of this document.  At the
            end is a list of C<#defines> whose name should be enough to tell
            you what they do, and a list of #defines which tell you if you
            need to C<#include> files to get the corresponding functionality.
            EOT

        footer => <<~'EOT',

            =head2 List of capability C<HAS_I<foo>> symbols

            This is a list of those symbols that dont appear elsewhere in ths
            document that indicate if the current platform has a certain
            capability.  Their names all begin with C<HAS_>.  Only those
            symbols whose capability is directly derived from the name are
            listed here.  All others have their meaning expanded out elsewhere
            in this document.  This (relatively) compact list is because we
            think that the expansion would add little or no value and take up
            a lot of space (because there are so many).  If you think certain
            ones should be expanded, send email to
            L<perl5-porters@perl.org|mailto:perl5-porters@perl.org>.

            Each symbol here will be C<#define>d if and only if the platform
            has the capability.  If you need more detail, see the
            corresponding entry in F<config.h>.  For convenience, the list is
            split so that the ones that indicate there is a reentrant version
            of a capability are listed separately

            __HAS_LIST__

            And, the reentrant capabilities:

            __HAS_R_LIST__

            Example usage:

            =over

             #ifdef HAS_STRNLEN
               use strnlen()
             #else
               use an alternative implementation
             #endif

            =back

            =head2 List of C<#include> needed symbols

            This list contains symbols that indicate if certain C<#include>
            files are present on the platform.  If your code accesses the
            functionality that one of these is for, you will need to
            C<#include> it if the symbol on this list is C<#define>d.  For
            more detail, see the corresponding entry in F<config.h>.

            __INCLUDE_LIST__

            Example usage:

            =over

             #ifdef I_WCHAR
               #include <wchar.h>
             #endif

            =back
            EOT
      },
    $globals_scn => {},
    $GV_scn => {},
    $hook_scn => {},
    $HV_scn => {},
    $io_scn => {},
    $integer_scn => {},
    $lexer_scn => {},
    $locales_scn => {},
    $magic_scn => {},
    $memory_scn => {},
    $mro_scn => {},
    $multicall_scn => {},
    $numeric_scn => {},
    $optree_construction_scn => {},
    $optree_manipulation_scn => {},
    $pack_scn => {},
    $pad_scn => {},
    $password_scn => {},
    $paths_scn => {},
    $intrpvar_scn => {},
    $prototypes_scn => {},
    $regexp_scn => {},
    $signals_scn => {},
    $site_scn => {
        header => <<~'EOT',
            These variables give details as to where various libraries,
            installation destinations, I<etc.>, go, as well as what various
            installation options were selected
            EOT
      },
    $sockets_scn => {},
    $filters_scn => {},
    $stack_scn => {},
    $string_scn => {
        header => <<~'EOT',
            See also C<L</Unicode Support>>.
            EOT
      },
    $SV_flags_scn => {},
    $SV_scn => {},
    $time_scn => {},
    $typedefs_scn => {},
    $unicode_scn => {},
    $utility_scn => {},
    $versioning_scn => {},
    $warning_scn => {},
    $xs_scn => {},
);

# Somewhat loose match for an apidoc line so we can catch minor typos.
# Parentheses are used to capture portions so that below we verify
# that things are the actual correct syntax.
my $apidoc_re = qr/ ^ (\s*)            # $1
                      (=?)             # $2
                      (\s*)            # $3
                      for (\s*)        # $4
                      apidoc (_item)?  # $5
                      (\s*)            # $6
                      (.*?)            # $7
                      \s* \n /x;
# Only certain flags, dealing with display, are acceptable for apidoc_item
my $display_flags = "fFnDopsT";

sub check_api_doc_line ($$) {
    my ($file, $in) = @_;

    return unless $in =~ $apidoc_re;

    my $is_item = defined $5;
    my $is_in_proper_form = length $1 == 0
                         && length $2 > 0
                         && length $3 == 0
                         && length $4 > 0
                         && length $6 > 0
                         && length $7 > 0;
    my $proto_in_file = $7;
    my $proto = $proto_in_file;
    $proto = "||$proto" if $proto !~ /\|/;
    my ($flags, $ret_type, $name, @args) = split /\s*\|\s*/, $proto;

    $name && $is_in_proper_form or die <<EOS;
Bad apidoc at $file line $.:
  $in
Expected:
  =for apidoc flags|returntype|name|arg|arg|...
  =for apidoc flags|returntype|name
  =for apidoc name
(or 'apidoc_item')
EOS

    die "Only [$display_flags] allowed in apidoc_item:\n$in"
                            if $is_item && $flags =~ /[^$display_flags]/;

    return ($name, $flags, $ret_type, $is_item, $proto_in_file, @args);
}

sub embed_override($) {
    my ($element_name) = shift;

    # If the entry is also in embed.fnc, it should be defined
    # completely there, but not here
    my $embed_docref = delete $funcflags{$element_name};

    return unless $embed_docref and %$embed_docref;

    my $flags = $embed_docref->{'flags'};
    warn "embed.fnc entry '$element_name' missing 'd' flag"
                                            unless $flags =~ /d/;

    return ($flags, $embed_docref->{'ret_type'}, $embed_docref->{args}->@*);
}

sub autodoc ($$) { # parse a file and extract documentation info
    my($fh,$file) = @_;
    my($in, $line_num, $header, $section);

    my $file_is_C = $file =~ / \. [ch] $ /x;

    # Count lines easier
    my $get_next_line = sub { $line_num++; return <$fh> };

    # Read the file
    while ($in = $get_next_line->()) {
        last unless defined $in;

        next unless (    $in =~ / ^ =for [ ]+ apidoc /x
                                      # =head1 lines only have effect in C files
                     || ($file_is_C && $in =~ /^=head1/));

        # Here, the line introduces a portion of the input that we care about.
        # Either it is for an API element, or heading text which we expect
        # will be used for elements later in the file

        my ($text, $element_name, $flags, $ret_type, $is_item, $proto_in_file);
        my (@args, @items);

        # If the line starts a new section ...
        if ($in=~ /^ = (?: for [ ]+ apidoc_section | head1 ) [ ]+ (.*) /x) {

            $section = $1;
            die "Unknown section name '$section' in $file near line $.\n"
                                    unless defined $valid_sections{$section};

        }
        elsif ($in=~ /^ =for [ ]+ apidoc \B /x) {   # Otherwise better be a
                                                    # plain apidoc line
            die "Unkown apidoc-type line '$in'" unless $in=~ /^=for apidoc_item/;
            die "apidoc_item doesn't immediately follow an apidoc entry: '$in'";
        }
        else {  # Plain apidoc

            ($element_name, $flags, $ret_type, $is_item, $proto_in_file, @args)
                                                = check_api_doc_line($file, $in);
            # Override this line with any info in embed.fnc
            my ($embed_flags, $embed_ret_type, @embed_args)
                                                = embed_override($element_name);
            if ($embed_ret_type) {
                warn "embed.fnc entry overrides redundant information in"
                    . " '$proto_in_file' in $file"
                                               if $flags || $ret_type || @args;
                $flags = $embed_flags;
                $ret_type = $embed_ret_type;
                @args = @embed_args;
            }
            elsif ($flags !~ /[my]/)  { # Not in embed.fnc, is missing if not
                                        # a macro or typedef
                $missing{$element_name} = $file;
            }

            die "flag '$1' is not legal (for function $element_name (from $file))"
                        if $flags =~ / ( [^AabCDdEeFfhiMmNnTOoPpRrSsUuWXxy] ) /x;

            die "'u' flag must also have 'm' or 'y' flags' for $element_name"
                                            if $flags =~ /u/ && $flags !~ /[my]/;
            warn ("'$element_name' not \\w+ in '$proto_in_file' in $file")
                        if $flags !~ /N/ && $element_name !~ / ^ [_[:alpha:]] \w* $ /x;

            if (exists $seen{$element_name} && $flags !~ /h/) {
                die ("'$element_name' in $file was already documented in $seen{$element_name}");
            }
            else {
                $seen{$element_name} = $file;
            }
        }

        # Here we have processed the initial line in the heading text or API
        # element, and have saved the important information from it into the
        # corresponding variables.  Now accumulate the text that applies to it
        # up to a terminating line, which is one of:
        # 1) =cut
        # 2) =head (in a C file only =head1)
        # 3) an end comment line in a C file: m:^\s*\*/:
        # 4) =for apidoc... (except apidoc_item lines)
        $text = "";
        my $head_ender_num = ($file_is_C) ? 1 : "";
        while (defined($in = $get_next_line->())) {

            last if $in =~ /^=cut/x;
            last if $in =~ /^=head$head_ender_num/;

            if ($file_is_C && $in =~ m: ^ \s* \* / $ :x) {

                # End of comment line in C files is a fall-back terminator,
                # but warn only if there actually is some accumulated text
                warn "=cut missing? $file:$line_num:$in" if $text =~ /\S/;
                last;
            }

            if ($in !~ / ^ =for [ ]+ apidoc /x) {
                $text .= $in;
                next;
            }

            # Here, the line is an apidoc line.  All but apidoc_item terminate
            # the text being accumulated.
            last if $in =~ / ^ =for [ ]+ apidoc_section /x;

            my ($item_name, $item_flags, $item_ret_type, $is_item,
                    $item_proto, @item_args) = check_api_doc_line($file, $in);
            last unless $is_item;

            # Here, is an apidoc_item_line; They can only come within apidoc
            # paragraphs.
            die "Unexpected api_doc_item line '$item_proto'"
                                                        unless $element_name;

            # We accept blank lines between these, but nothing else;
            die "apidoc_item lines must immediately follow apidoc lines for "
              . " '$element_name' in $file"
                                                            if $text =~ /\S/;
            # Override this line with any info in embed.fnc
            my ($embed_flags, $embed_ret_type, @embed_args)
                                                = embed_override($item_name);
            if ($embed_ret_type) {
                warn "embed.fnc entry overrides redundant information in"
                    . " '$item_proto' in $file"
                                if $item_flags || $item_ret_type || @item_args;

                $item_flags = $embed_flags;
                $item_ret_type = $embed_ret_type;
                @item_args = @embed_args;
            }

            # Use the base entry flags if none for this item; otherwise add in
            # any non-display base entry flags.
            if ($item_flags) {
                $item_flags .= $flags =~ s/[$display_flags]//rg;
            }
            else {
                $item_flags = $flags;
            }
            $item_ret_type = $ret_type unless $item_ret_type;
            @item_args = @args unless @item_args;
            push @items, { name     => $item_name,
                           ret_type => $item_ret_type,
                           flags    => $item_flags,
                           args     => [ @item_args ],
                         };

            # This line shows that this element is documented.
            delete $funcflags{$item_name};
        }

        # Here, are done accumulating the text for this item.  Trim it
        $text =~ s/ ^ \s* //x;
        $text =~ s/ \s* $ //x;
        $text .= "\n" if $text ne "";

        # And treat all-spaces as nothing at all
        undef $text unless $text =~ /\S/;

        if ($element_name) {

            # Here, we have accumulated into $text, the pod for $element_name
            my $where = $flags =~ /A/ ? 'api' : 'guts';

            $section = "Functions in file $file" unless defined $section;
            die "No =for apidoc_section nor =head1 in $file for '$element_name'\n"
                                                    unless defined $section;
            if (exists $docs{$where}{$section}{$element_name}) {
                warn "$0: duplicate API entry for '$element_name' in"
                    . " $where/$section\n";
                next;
            }

            # Override the text with just a link if the flags call for that
            my $is_link_only = ($flags =~ /h/);
            if ($is_link_only) {
                if ($file_is_C) {
                    die "Can't currently handle link with items to it:\n$in" if @items;
                    redo;    # Don't put anything if C source
                }

                # Here, is an 'h' flag in pod.  We add a reference to the pod (and
                # nothing else) to perlapi/intern.  (It would be better to add a
                # reference to the correct =item,=header, but something that makes
                # it harder is that it that might be a duplicate, like '=item *';
                # so that is a future enhancement XXX.  Another complication is
                # there might be more than one deserving candidates.)
                my $podname = $file =~ s!.*/!!r;    # Rmv directory name(s)
                $podname =~ s/\.pod//;
                $text = "Described in L<$podname>.\n";

                # Don't output a usage example for linked to documentation if
                # it is trivial (has no arguments) and we aren't to add a
                # semicolon
                $flags .= 'U' if $flags =~ /n/ && $flags !~ /[Us]/;

                # Keep track of all the pod files that we refer to.
                push $described_elsewhere{$podname}->@*, $podname;
            }

            $docs{$where}{$section}{$element_name}{flags} = $flags;
            $docs{$where}{$section}{$element_name}{pod} = $text;
            $docs{$where}{$section}{$element_name}{file} = $file;
            $docs{$where}{$section}{$element_name}{ret_type} = $ret_type;
            push $docs{$where}{$section}{$element_name}{args}->@*, @args;
            push $docs{$where}{$section}{$element_name}{items}->@*, @items;
        }
        elsif ($text) {
            $valid_sections{$section}{header} = "" unless
                                    defined $valid_sections{$section}{header};
            $valid_sections{$section}{header} .= "\n$text";
        }

        # We already have the first line of what's to come in $in
        redo;

    } # End of loop through input
}

my %configs;
my @has_defs;
my @has_r_defs;     # Reentrant symbols
my @include_defs;
sub parse_config_h {
    use re '/aa';   # Everthing is ASCII in this file

    # Process config.h
    my $config_h = 'config.h';
    $config_h = 'win32/config.h' unless -e $config_h;
    die "Can't find $config_h" unless -e $config_h;
    open my $fh, '<', $config_h or die "Can't open $config_h: $!";
    while (<$fh>) {

        # Look for lines like /* FOO_BAR:
        # By convention all config.h descriptions begin like that
        if (m[ ^ /\* [ ] ( [[:alpha:]] \w+ ) : \s* $ ]ax) {
            my $name = $1;

            # Here we are starting the description for $name in config.h.  We
            # accumulate the entire description for it into @description.
            # Flowing text from one input line to another is appended into the
            # same array element to make a single flowing line element, but
            # verbatim lines are kept as separate elements in @description.
            # This will facilitate later doing pattern matching without regard
            # to line boundaries on non-verbatim text.

            die "Multiple config.h entries for '$name'"
                                        if defined $configs{$name}{description};

            # Get first line of description
            $_ = <$fh>;

            # Each line in the description begins with blanks followed by '/*'
            # and some spaces.
            die "Unexpected config.h initial line for $name: '$_'"
                                            unless s/ ^ ( \s* \* \s* ) //x;
            my $initial_text = $1;

            # Initialize the description with this first line (after having
            # stripped the prefix text)
            my @description = $_;

            # The first line is used as a template for how much indentation
            # each normal succeeding line has.  Lines indented further
            # will be considered as intended to be verbatim.  But, empty lines
            # likely won't have trailing blanks, so just strip the whole thing
            # for them.
            my $strip_initial_qr = qr!   \s* \* \s* $
                                    | \Q$initial_text\E
                                    !x;
            $configs{$name}{verbatim} = 0;

            # Read in the remainder of the description
            while (<$fh>) {
                last if s| ^ \s* \* / ||x;  # A '*/' ends it

                die "Unexpected config.h description line for $name: '$_'"
                                                unless s/$strip_initial_qr//;

                # Fix up the few flawed lines in config.h wherein a new
                # sentence begins with a tab (and maybe a space after that).
                # Although none of them currently do, let it recognize
                # something like
                #
                #   "... text").  The next sentence ...
                #
                s/ ( \w "? \)? \. ) \t \s* ( [[:alpha:]] ) /$1  $2/xg;

                # If this line has extra indentation or looks to have columns,
                # it should be treated as verbatim.  Columns are indicated by
                # use of interior: tabs, 3 spaces in a row, or even 2 spaces
                # not preceded by punctuation.
                if ($_ !~ m/  ^ \s
                              | \S (?:                    \t
                                    |                     \s{3}
                                    |  (*nlb:[[:punct:]]) \s{2}
                                   )
                           /x)
                {
                    # But here, is not a verbatim line.  Add an empty line if
                    # this is the first non-verbatim after a run of verbatims
                    if ($description[-1] =~ /^\s/) {
                        push @description, "\n", $_;
                    }
                    else {  # Otherwise, append this flowing line to the
                            # current flowing line
                        $description[-1] .= $_;
                    }
                }
                else {
                    $configs{$name}{verbatim} = 1;

                    # The first verbatim line in a run of them is separated by an
                    # empty line from the flowing lines above it
                    push @description, "\n" if $description[-1] =~ /^\S/;

                    $_ = Text::Tabs::expand($_);

                    # Only a single space so less likely to wrap
                    s/ ^ \s* / /x;

                    push @description, $_;
                }
            }

            push $configs{$name}{description}->@*, @description

        }   # Not a description; see if it is a macro definition.
        elsif (m! ^
                  (?: / \* )?                   # Optional commented-out
                                                # indication
                      \# \s* define \s+ ( \w+ ) # $1 is the name
                  (   \s* )                     # $2 indicates if args or not
                  (   .*? )                     # $3 is any definition
                  (?: / \s* \* \* / )?          # Optional trailing /**/ or / **/
                  $
                !x)
        {
            my $name = $1;

            # There can be multiple definitions for a name.  We want to know
            # if any of them has arguments, and if any has a body.
            $configs{$name}{has_args} //= $2 eq "";
            $configs{$name}{has_args} ||= $2 eq "";
            $configs{$name}{has_defn} //= $3 ne "";
            $configs{$name}{has_defn} ||= $3 ne "";
        }
    }

    # We now have stored the description and information about every #define
    # in the file.  The description is in a form convenient to operate on to
    # convert to pod.  Do that now.
    foreach my $name (keys %configs) {
        next unless defined $configs{$name}{description};

        # All adjacent non-verbatim lines of the description are appended
        # together in a single element in the array.  This allows the patterns
        # to work across input line boundaries.

        my $pod = "";
        while (defined ($_ = shift $configs{$name}{description}->@*)) {
            chomp;

            if (/ ^ \S /x) {  # Don't edit verbatim lines

                # Enclose known file/path names not already so enclosed
                # with <...>.  (Some entries in config.h are already
                # '<path/to/file>')
                my $file_name_qr = qr! [ \w / ]+ \.
                                    (?: c | h | xs | p [lm] | pmc | PL
                                        | sh | SH | exe ) \b
                                    !xx;
                my $path_name_qr = qr! (?: / \w+ )+ !x;
                for my $re ($file_name_qr, $path_name_qr) {
                    s! (*nlb:[ < \w / ]) ( $re ) !<$1>!gxx;
                }

                # Enclose <... file/path names with F<...> (but no double
                # angle brackets)
                for my $re ($file_name_qr, $path_name_qr) {
                    s! < ( $re ) > !F<$1>!gxx;
                }

                # Explain metaconfig units
                s/ ( \w+ \. U \b ) /$1 (part of metaconfig)/gx;

                # Convert "See foo" to "See C<L</foo>>" if foo is described in
                # this file.  Also create a link to the known file INSTALL.
                # And, to be more general, handle "See also foo and bar", and
                # "See also foo, bar, and baz"
                while (m/ \b [Ss]ee \s+
                         (?: also \s+ )?    ( \w+ )
                         (?: ,  \s+         ( \w+ ) )?
                         (?: ,? \s+ and \s+ ( \w+ ) )? /xg) {
                    my @links = $1;
                    push @links, $2 if defined $2;
                    push @links, $3 if defined $3;
                    foreach my $link (@links) {
                        if ($link eq 'INSTALL') {
                            s/ \b INSTALL \b /C<L<INSTALL>>/xg;
                        }
                        elsif (grep { $link =~ / \b $_ \b /x } keys %configs) {
                            s| \b $link \b |C<L</$link>>|xg;
                            $configs{$link}{linked} = 1;
                            $configs{$name}{linked} = 1;
                        }
                    }
                }

                # Enclose what we think are symbols with C<...>.
                no warnings 'experimental::vlb';
                s/ (*nlb:<)
                   (
                        # Any word followed immediately with parens or
                        # brackets
                        \b \w+ (?: \( [^)]* \)    # parameter list
                                 | \[ [^]]* \]    # or array reference
                               )
                    | (*plb: ^ | \s ) -D \w+    # Also -Dsymbols.
                    | \b (?: struct | union ) \s \w+

                        # Words that contain underscores (which are
                        # definitely not text) or three uppercase letters in
                        # a row.  Length two ones, like IV, aren't enclosed,
                        # because they often don't look as nice.
                    | \b \w* (?: _ | [[:upper:]]{3,} ) \w* \b
                   )
                    (*nla:>)
                 /C<$1>/xg;

                # These include foo when the name is HAS_foo.  This is a
                # heuristic which works in most cases.
                if ($name =~ / ^ HAS_ (.*) /x) {
                    my $symbol = lc $1;

                    # Don't include path components, nor things already in
                    # <>, or with trailing '(', '['
                    s! \b (*nlb:[/<]) $symbol (*nla:[[/>(]) \b !C<$symbol>!xg;
                }
            }

            $pod .=  "$_\n";
        }
        delete $configs{$name}{description};

        $configs{$name}{pod} = $pod;
    }

    # Now have converted the description to pod.  We also now have enough
    # information that we can do cross checking to find definitions without
    # corresponding pod, and see if they are mentioned in some description;
    # otherwise they aren't documented.
  NAME:
    foreach my $name (keys %configs) {

        # A definition without pod
        if (! defined $configs{$name}{pod}) {

            # Leading/trailing underscore means internal to config.h, e.g.,
            # _GNU_SOURCE
            next if $name =~ / ^ _ /x;
            next if $name =~ / _ $ /x;

            # MiXeD case names are internal to config.h; the first 4
            # characters are sufficient to determine this
            next if $name =~ / ^ [[:upper:]] [[:lower:]]
                                 [[:upper:]] [[:lower:]]
                            /x;

            # Here, not internal to config.h.  Look to see if this symbol is
            # mentioned in the pod of some other.  If so, assume it is
            # documented.
            foreach my $check_name (keys %configs) {
                my $this_element = $configs{$check_name};
                my $this_pod = $this_element->{pod};
                if (defined $this_pod) {
                    next NAME if $this_pod =~ / \b $name \b /x;
                }
            }

            warn "$name has no documentation\n";
            $missing_macros{$name} = 'config.h';

            next;
        }

        my $has_defn = $configs{$name}{has_defn};
        my $has_args = $configs{$name}{has_args};

        # Check if any section already has an entry for this element.
        # If so, it better be a placeholder, in which case we replace it
        # with this entry.
        foreach my $section (keys $docs{'api'}->%*) {
            if (exists $docs{'api'}{$section}{$name}) {
                my $was = $docs{'api'}{$section}{$name}->{pod};
                $was = "" unless $was;
                chomp $was;
                if ($was ne "" && $was !~ m/$link_text/) {
                    die "Multiple descriptions for $name\n"
                        . "$section contained '$was'";
                }
                $docs{'api'}{$section}{$name}->{pod} = $configs{$name}{pod};
                $configs{$name}{section} = $section;
                last;
            }
        }

        my $handled = 0;    # Haven't handled this yet

        if (defined $configs{$name}{'section'}) {
            # This has been taken care of elsewhere.
            $handled = 1;
        }
        else {
            my $flags = "";
            if ($has_defn && ! $has_args) {
                $configs{$name}{args} = 1;
            }

            # Symbols of the form I_FOO are for #include files.  They have
            # special usage information
            if ($name =~ / ^ I_ ( .* ) /x) {
                my $file = lc $1 . '.h';
                $configs{$name}{usage} = <<~"EOT";
                    #ifdef $name
                        #include <$file>
                    #endif
                    EOT
            }

            # Compute what section this variable should go into.  This
            # heuristic was determined by manually inspecting the current
            # things in config.h, and should be adjusted as necessary as
            # deficiencies are found.
            #
            # This is the default section for macros with a definiton but
            # no arguments, meaning it is replaced unconditionally
            #
            my $sb = qr/ _ | \b /x; # segment boundary
            my $dash_or_spaces = qr/ - | \s+ /x;
            my $pod = $configs{$name}{pod};
            if ($name =~ / ^ USE_ /x) {
                $configs{$name}{'section'} = $site_scn;
            }
            elsif ($name =~ / SLEEP | (*nlb:SYS_) TIME | TZ | $sb TM $sb /x)
            {
                $configs{$name}{'section'} = $time_scn;
            }
            elsif ($name =~ /  DOUBLE | FLOAT | LONGDBL | LDBL | ^ NV
                            | $sb CASTFLAGS $sb
                            | QUADMATH
                            | $sb (?: IS )? NAN
                            | $sb (?: IS )? FINITE
                            /x)
            {
                $configs{$name}{'section'} =
                                    $floating_scn;
            }
            elsif ($name =~ / (?: POS | OFF | DIR ) 64 /x) {
                $configs{$name}{'section'} = $filesystem_scn;
            }
            elsif (   $name =~ / $sb (?: BUILTIN | CPP ) $sb | ^ CPP /x
                   || $configs{$name}{pod} =~ m/ \b align /x)
            {
                $configs{$name}{'section'} = $compiler_scn;
            }
            elsif ($name =~ / ^ [IU] [ \d V ]
                            | ^ INT | SHORT | LONG | QUAD | 64 | 32 /xx)
            {
                $configs{$name}{'section'} = $integer_scn;
            }
            elsif ($name =~ / $sb t $sb /x) {
                $configs{$name}{'section'} = $typedefs_scn;
                $flags .= 'y';
            }
            elsif (   $name =~ / ^ PERL_ ( PRI | SCN ) | $sb FORMAT $sb /x
                    && $configs{$name}{pod} =~ m/ \b format \b /ix)
            {
                $configs{$name}{'section'} = $formats_scn;
            }
            elsif ($name =~ / BACKTRACE /x) {
                $configs{$name}{'section'} = $dump_scn;
            }
            elsif ($name =~ / ALLOC $sb /x) {
                $configs{$name}{'section'} = $memory_scn;
            }
            elsif (   $name =~ /   STDIO | FCNTL | EOF | FFLUSH
                                | $sb FILE $sb
                                | $sb DIR $sb
                                | $sb LSEEK
                                | $sb INO $sb
                                | $sb OPEN
                                | $sb CLOSE
                                | ^ DIR
                                | ^ INO $sb
                                | DIR $
                                | FILENAMES
                                /x
                    || $configs{$name}{pod} =~ m!  I/O | stdio
                                                | file \s+ descriptor
                                                | file \s* system
                                                | statfs
                                                !x)
            {
                $configs{$name}{'section'} = $filesystem_scn;
            }
            elsif ($name =~ / ^ SIG | SIGINFO | signal /ix) {
                $configs{$name}{'section'} = $signals_scn;
            }
            elsif ($name =~ / $sb ( PROTO (?: TYPE)? S? ) $sb /x) {
                $configs{$name}{'section'} = $prototypes_scn;
            }
            elsif (   $name =~ / ^ LOC_ /x
                    || $configs{$name}{pod} =~ /full path/i)
            {
                $configs{$name}{'section'} = $paths_scn;
            }
            elsif ($name =~ / $sb LC_ | LOCALE | langinfo /xi) {
                $configs{$name}{'section'} = $locales_scn;
            }
            elsif ($configs{$name}{pod} =~ /  GCC | C99 | C\+\+ /xi) {
                $configs{$name}{'section'} = $compiler_scn;
            }
            elsif ($name =~ / PASSW (OR)? D | ^ PW | ( PW | GR ) ENT /x)
            {
                $configs{$name}{'section'} = $password_scn;
            }
            elsif ($name =~ /  SOCKET | $sb SOCK /x) {
                $configs{$name}{'section'} = $sockets_scn;
            }
            elsif (   $name =~ / THREAD | MULTIPLICITY /x
                    || $configs{$name}{pod} =~ m/ \b pthread /ix)
            {
                $configs{$name}{'section'} = $concurrency_scn;
            }
            elsif ($name =~ /  PERL | ^ PRIV | SITE | ARCH | BIN
                                | VENDOR | ^ USE
                            /x)
            {
                $configs{$name}{'section'} = $site_scn;
            }
            elsif (   $pod =~ / \b floating $dash_or_spaces point \b /ix
                    || $pod =~ / \b (double | single) $dash_or_spaces precision \b /ix
                    || $pod =~ / \b doubles \b /ix
                    || $pod =~ / \b (?: a | the | long ) \s+ (?: double | NV ) \b /ix)
            {
                $configs{$name}{'section'} =
                                    $floating_scn;
            }
            else {
                # Above are the specific sections.  The rest go into a
                # grab-bag of general configuration values.  However, we put
                # two classes of them into lists of their names, without their
                # descriptions, when we think that the description doesn't add
                # any real value.  One list contains the #include variables:
                # the description is basically boiler plate for each of these.
                # The other list contains the very many things that are of the
                # form HAS_foo, and \bfoo\b is contained in its description,
                # and there is no verbatim text in the pod or links to/from it
                # (which would add value).  That means that it is likely the
                # intent of the variable can be gleaned from just its name,
                # and unlikely the description adds signficant value, so just
                # listing them suffices.  Giving their descriptions would
                # expand this pod significantly with little added value.
                if (   ! $has_defn
                    && ! $configs{$name}{verbatim}
                    && ! $configs{$name}{linked})
                {
                    if ($name =~ / ^ I_ ( .* ) /x) {
                        push @include_defs, $name;
                        next;
                    }
                    elsif ($name =~ / ^ HAS_ ( .* ) /x) {
                        my $canonical_name = $1;
                        $canonical_name =~ s/_//g;

                        my $canonical_pod = $configs{$name}{pod};
                        $canonical_pod =~ s/_//g;

                        if ($canonical_pod =~ / \b $canonical_name \b /xi) {
                            if ($name =~ / $sb R $sb /x) {
                                push @has_r_defs, $name;
                            }
                            else {
                                push @has_defs, $name;
                            }
                            next;
                        }
                    }
                }

                $configs{$name}{'section'} = $genconfig_scn;
            }

            my $section = $configs{$name}{'section'};
            die "Internal error: '$section' not in \%valid_sections"
                            unless grep { $_ eq $section } keys %valid_sections;
            $flags .= 'AdmnT';
            $flags .= 'U' unless defined $configs{$name}{usage};

            $docs{'api'}{$section}{$name}{flags} = $flags;
            $docs{'api'}{$section}{$name}{pod} = $configs{$name}{pod};
            $docs{'api'}{$section}{$name}{ret_type} = "";
            $docs{'api'}{$section}{$name}{file} = 'config.h';
            $docs{'api'}{$section}{$name}{usage}
                = $configs{$name}{usage} if defined $configs{$name}{usage};
            push $docs{'api'}{$section}{$name}{args}->@*, ();
            push $docs{'api'}{$section}{$name}{items}->@*, ();
        }
    }
}

sub docout ($$$) { # output the docs for one function
    my($fh, $element_name, $docref) = @_;

    # Trim trailing space
    $element_name =~ s/\s*$//;

    my $flags = $docref->{flags};
    my $pod = $docref->{pod} // "";
    my $file = $docref->{file};

    my @items = $docref->{items}->@*;

    # Make the main element the first of the items.  This allows uniform
    # treatment below
    unshift @items, {   name => $element_name,
                        flags => $flags,
                        ret_type => $docref->{ret_type},
                        args => [ $docref->{args}->@* ],
                    };

    warn("Empty pod for $element_name (from $file)") unless $pod =~ /\S/;

    print $fh "\n=over $description_indent\n";
    print $fh "\n=item C<$_->{name}>\n" for @items;

    # If we're printing only a link to an element, this isn't the major entry,
    # so no X<> here.
    if ($flags !~ /h/) {
        print $fh "X<$_->{name}>" for @items;
        print $fh "\n";
    }

    for my $item (@items) {
        if ($item->{flags} =~ /D/) {
            print $fh <<~"EOT";

                C<B<DEPRECATED!>>  It is planned to remove C<$item->{name}> from a
                future release of Perl.  Do not use it for new code; remove it from
                existing code.
                EOT
        }
        elsif ($item->{flags} =~ /x/) {
            print $fh <<~"EOT";

                NOTE: C<$item->{name}> is B<experimental> and may change or be
                removed without notice.
                EOT
        }
    }

    chomp $pod;     # Make sure prints pod with a single trailing \n
    print $fh "\n", $pod, "\n";

    for my $item (@items) {
        my $item_flags = $item->{flags};
        my $item_name = $item->{name};

        print $fh "\nNOTE: the C<perl_$item_name()> form is B<deprecated>.\n"
                                                    if $item_flags =~ /O/;
        # Is Perl_, but no #define foo # Perl_foo
        if (($item_flags =~ /p/ && $item_flags =~ /o/ && $item_flags !~ /M/)

             # Can't handle threaded varargs
         || ($item_flags =~ /f/ && $item_flags !~ /T/))
        {
            $item->{name} = "Perl_$item_name";
            print $fh <<~"EOT";

                NOTE: C<$item_name> must be explicitly called as
                C<$item->{name}>
                EOT
            print $fh "with an C<aTHX_> parameter" if $item_flags !~ /T/;
            print $fh ".\n";
        }
    }

    if ($flags =~ /[Uy]/) { # no usage; typedefs are considered simple enough
                            # to never warrant a usage line
        warn("U and s flags are incompatible")
                                            if $flags =~ /U/ && $flags =~ /s/;
        # nothing
    } else {

        print $fh "\n=over $usage_indent\n";

        if (defined $docref->{usage}) {     # An override of the usage section
            print $fh "\n", ($docref->{usage} =~ s/^/ /mrg), "\n";
        }
        else {

            # Add the thread context formal parameter on expanded-out names
            for my $item (@items) {
                unshift $item->{args}->@*, (($item->{args}->@*)
                                            ? "pTHX_"
                                            : "pTHX")
                                                   if $item->{flags} !~ /T/
                                                   && $item->{name} =~ /^Perl_/;
            }

            # Look through all the items in this entry.  If all have the same
            # return type and arguments, only the main entry is displayed.
            # Also, find the longest return type and longest name so that if
            # multiple ones are shown, they can be vertically aligned nicely
            my $need_individual_usage = 0;
            my $longest_name_length = length $items[0]->{name};
            my $base_ret_type = $items[0]->{ret_type};
            my $longest_ret = length $base_ret_type;
            my @base_args = $items[0]->{args}->@*;
            for (my $i = 1; $i < @items; $i++) {
                no warnings 'experimental::smartmatch';
                my $item = $items[$i];
                $need_individual_usage = 1
                                    if    $item->{ret_type} ne $base_ret_type
                                    || ! ($item->{args}->@* ~~ @base_args);
                my $ret_length = length $item->{ret_type};
                $longest_ret = $ret_length if $ret_length > $longest_ret;
                my $name_length = length $item->{name};
                $longest_name_length = $name_length
                                        if $name_length > $longest_name_length;
            }

            # If we're only showing one entry, only its length matters.
            $longest_name_length = length($items[0]->{name})
                                                unless $need_individual_usage;
            print $fh "\n";

            my $indent = 1;     # 1 is sufficient for verbatim; =over is used
                                # for more
            my $ret_name_sep_length = 2; # spaces between return type and name
            my $name_indent = $indent + $longest_ret;
            $name_indent += $ret_name_sep_length if $longest_ret;

            # 80 column terminal - 1 for pager adding a column; -7 for nroff
            # indent;
            my $max_length = 80 - 1 - 7 - $description_indent - $usage_indent;

            for my $item (@items) {
                my $ret_type = $item->{ret_type};
                my @args = $item->{args}->@*;
                my $name = $item->{name};
                my $item_flags = $item->{flags};

                # The return type
                print $fh (" " x $indent), $ret_type;

                print $fh " " x (  $ret_name_sep_length
                                 + $longest_ret - length $ret_type);
                print $fh $name;

                if ($item_flags =~ /n/) { # no args
                    warn("$file: $element_name: n flag without m")
                                                    unless $item_flags =~ /m/;
                    warn("$file: $name: n flag but apparently has args")
                                                                    if @args;
                }
                else {
                    # +1 for the '('
                    my $arg_indent = $name_indent + $longest_name_length + 1;

                    # Align the argument lists of the items
                    print $fh " " x ($longest_name_length - length($name));
                    print $fh "(";

                    # Display as many of the arguments on the same line as
                    # will fit.
                    my $total_length = $arg_indent;
                    my $first_line = 1;
                    for (my $i = 0; $i < @args; $i++) {
                        my $arg = $args[$i];
                        my $arg_length = length($arg);

                        # All but the first arg are preceded by a blank
                        my $use_blank = $i > 0;

                        # +1 here and below because either the argument has a
                        # trailing comma or trailing ')'
                        $total_length += $arg_length + $use_blank + 1;

                        # We want none of the arguments to be positioned so
                        # they extend too far to the right.  Ideally, they
                        # should all start in the same column as the arguments
                        # on the first line of the function display do.  But, if
                        # necessary, outdent them so that they all start in
                        # another column, with the longest ending at the right
                        # margin, like so:
                        #                   void  function_name(pTHX_ short1,
                        #                                    short2,
                        #                                    very_long_argument,
                        #                                    short3)
                        if ($total_length > $max_length) {

                            # If this is the first continuation line,
                            # calculate the longest argument; this will be the
                            # one we may have to outdent for.
                            if ($first_line) {
                                $first_line = 0;

                                # We will need at least as much as the current
                                # argument
                                my $longest_arg_length = $arg_length
                                                       + $use_blank + 1;

                                # Look through the rest of the args to see if
                                # any are longer than this one.
                                for (my $j = $i + 1; $j < @args; $j++) {

                                    # Include the trailing ',' or ')' in the
                                    # length.  No need to concern ourselves
                                    # with a leading blank, as the argument
                                    # would be positioned first on the next
                                    # line
                                    my $peek_arg_length = length ($args[$j])
                                                        + 1;
                                    $longest_arg_length = $peek_arg_length
                                      if $peek_arg_length > $longest_arg_length;
                                }

                                # Calculate the new indent if necessary.
                                $arg_indent = $max_length - $longest_arg_length
                                        if $arg_indent + $longest_arg_length
                                                                > $max_length;
                            }

                            print $fh "\n", (" " x $arg_indent);
                            $total_length = $arg_indent + $arg_length + 1;
                            $use_blank = 0;
                        }

                        # Display this argument
                        print $fh " " if $use_blank;
                        print $fh $arg;
                        print $fh "," if $i < @args - 1 && $args[$i] ne 'pTHX_';

                    } # End of loop through args

                    print $fh ")";
                }

                print $fh ";" if $item_flags =~ /s/; # semicolon: "dTHR;"
                print $fh "\n";

                # Only the first entry is normally displayed
                last unless $need_individual_usage;
            }
        }

        print $fh "\n=back\n";
    }

    print $fh "\n=back\n";
    print $fh "\n=for hackers\nFound in file $file\n";
}

sub construct_missings_section {
    my ($pod_name, $missings_ref) = @_;
    my $text = "";

    return $text unless $missings_ref->@*;

    $text .= <<~EOT;

        =head1 Undocumented functions

        EOT
    if ($pod_name eq 'perlapi') {
        $text .= <<~'EOT';
            The following functions have been flagged as part of the public
            API, but are currently undocumented.  Use them at your own risk,
            as the interfaces are subject to change.  Functions that are not
            listed in this document are not intended for public use, and
            should NOT be used under any circumstances.

            If you feel you need to use one of these functions, first send
            email to L<perl5-porters@perl.org|mailto:perl5-porters@perl.org>.
            It may be that there is a good reason for the function not being
            documented, and it should be removed from this list; or it may
            just be that no one has gotten around to documenting it.  In the
            latter case, you will be asked to submit a patch to document the
            function.  Once your patch is accepted, it will indicate that the
            interface is stable (unless it is explicitly marked otherwise) and
            usable by you.
            EOT
    }
    else {
        $text .= <<~'EOT';
            The following functions are currently undocumented.  If you use
            one of them, you may wish to consider creating and submitting
            documentation for it.
            EOT
    }

    $text .= "\n";

    # Sort the elements.
    my @missings = sort dictionary_order $missings_ref->@*;

    # Output the X<> references to the names, packed since they don't get
    # displayed, but not too many per line so that when someone is editing the
    # file, it doesn't run on
    my $line_length = 0;
    for my $missing (sort dictionary_order @missings) {
        my $entry = "X<$missing>";
        my $entry_length = length $entry;

        # Don't loop forever if we have a verrry long name, and don't go too
        # far to the right.
        if ($line_length > 0 && $line_length + $entry_length > $max_width) {
            $text .= "\n";
            $line_length = 0;
        }

        $text .= $entry;
        $line_length += $entry_length;
    }

    use integer;

    # Look through all the elements in the list and see how many columns we
    # could place them in the output what will fit in the available width.
    my $min_spacer = 2;     # Need this much space between columns
    my $columns;
    my $rows;
    my @col_widths;

  COLUMN:
    # We start with more columns, and work down until we find a number that
    # can accommodate all the data.  This algorithm doesn't require the
    # resulting columns to all have the same width.  This can allow for
    # as tight of packing as the data will possibly allow.
    for ($columns = 7; $columns > 1; $columns--) {

        # For this many columns, we will need this many rows (final row might
        # not be completely filled)
        $rows = (@missings + $columns - 1) / $columns;

        my $row_width = 0;
        my $i = 0;  # Which missing element

        # For each column ...
        for my $col (0 .. $columns - 1) {

            # Calculate how wide the column needs to be, which is based on the
            # widest element in it
            $col_widths[$col] = 0;

            # Look through all the rows to find the widest element
            for my $row (0 .. $rows - 1) {

                # Skip if this row doesn't have an entry for this column
                last if $i >= @missings;

                # This entry occupies this many bytes.
                my $this_width = length $missings[$i];

                # All but the final column need a spacer between it and the
                # next column over.
                $this_width += $min_spacer if $col < $columns - 1;


                # This column will need to have enough width to accommodate
                # this element
                if ($this_width > $col_widths[$col]) {

                    # We can't have this many columns if the total width
                    # exceeds the available; bail now and try fewer columns
                    next COLUMN if $row_width + $this_width > $max_width;

                    $col_widths[$col] = $this_width;
                }

                $i++;   # The next row will contain the next item
            }

            $row_width += $col_widths[$col];
            next COLUMN if $row_width > $max_width;
        }

        # If we get this far, this many columns works
        last;
    }

    # Here, have calculated the number of rows ($rows) and columns ($columns)
    # required to list the elements.  @col_widths contains the width of each
    # column.

    $text .= "\n\n=over $description_indent\n\n";

    # Assemble the output
    for my $row (0 .. $rows - 1) {
        for my $col (0 .. $columns - 1) {
            $text .= " " if $col == 0;  # Indent one to mark as verbatim

            my $index = $row + $rows * $col;  # Convert 2 dimensions to 1

            # Skip if this row doesn't have an entry for this column
            next if $index >= @missings;

            my $element = $missings[$index];
            $text .= $element;

            # Add alignment spaces for all but final column
            $text .= " " x ($col_widths[$col] - length $element)
                                                        if $col < $columns - 1;
        }

        $text .= "\n";  # End of row
    }

    $text .= "\n=back\n";

    return $text;
}

sub dictionary_order {
    # Do a case-insensitive dictionary sort, with only alphabetics
    # significant, falling back to using everything for determinancy
    return (uc($a =~ s/[[:^alpha:]]//r) cmp uc($b =~ s/[[:^alpha:]]//r))
           || uc($a) cmp uc($b)
           || $a cmp $b;
}

sub output {
    my ($podname, $header, $dochash, $missings_ref, $footer) = @_;
    #
    # strip leading '|' from each line which had been used to hide
    # pod from pod checkers.
    s/^\|//gm for $header, $footer;

    my $fh = open_new("pod/$podname.pod", undef,
                      {by => "$0 extracting documentation",
                       from => 'the C source files'}, 1);

    print $fh $header, "\n";

    for my $section_name (sort dictionary_order keys %valid_sections) {
        my $section_info = $dochash->{$section_name};

        # We allow empty sections in perlintern.
        if (! $section_info && $podname eq 'perlapi') {
            warn "Empty section '$section_name'; skipped";
            next;
        }

        print $fh "\n=head1 $section_name\n";

        if ($podname eq 'perlapi') {
            print $fh "\n", $valid_sections{$section_name}{header}, "\n"
                                if defined $valid_sections{$section_name}{header};

            # Output any heading-level documentation and delete so won't get in
            # the way later
            if (exists $section_info->{""}) {
                print $fh "\n", $section_info->{""}, "\n";
                delete $section_info->{""};
            }
        }


        if ($section_info) {
            for my $function_name (sort dictionary_order keys %$section_info) {
                docout($fh, $function_name, $section_info->{$function_name});
            }
        }
        else {
            print $fh "\nThere are only public API items currently in $section_name\n";
        }

        print $fh "\n", $valid_sections{$section_name}{footer}, "\n"
                            if $podname eq 'perlapi'
                            && defined $valid_sections{$section_name}{footer};
    }

    print $fh construct_missings_section($podname, $missings_ref);

    print $fh "\n$footer\n=cut\n";

    read_only_bottom_close_and_rename($fh);
}

foreach (@{(setup_embed())[0]}) {
    next if @$_ < 2;
    my ($flags, $ret_type, $func, @args) = @$_;
    s/\b(?:NN|NULLOK)\b\s+//g for @args;

    $funcflags{$func} = {
                         flags => $flags,
                         ret_type => $ret_type,
                         args => \@args,
                        };
}

# glob() picks up docs from extra .c or .h files that may be in unclean
# development trees.
open my $fh, '<', 'MANIFEST'
    or die "Can't open MANIFEST: $!";
while (my $line = <$fh>) {
    next unless my ($file) = $line =~ /^(\S+\.(?:[ch]|pod))\t/;

    # Don't pick up pods from these.  (We may pick up generated stuff from
    # /lib though)
    next if $file =~ m! ^ ( cpan | dist | ext ) / !x;

    open F, '<', $file or die "Cannot open $file for docs: $!\n";
    autodoc(\*F,$file);
    close F or die "Error closing $file: $!\n";
}
close $fh or die "Error whilst reading MANIFEST: $!";

parse_config_h();

for (sort keys %funcflags) {
    next unless $funcflags{$_}{flags} =~ /d/;
    next if $funcflags{$_}{flags} =~ /h/;
    warn "no docs for $_\n";
}

foreach (sort keys %missing) {
    warn "Function '$_', documented in $missing{$_}, not listed in embed.fnc";
}

# List of funcs in the public API that aren't also marked as core-only,
# experimental nor deprecated.
my @missing_api = grep $funcflags{$_}{flags} =~ /A/
                    && $funcflags{$_}{flags} !~ /[xD]/
                    && !$docs{api}{$_}, keys %funcflags;
push @missing_api, keys %missing_macros;

my $other_places = join ", ", map { "L<$_>" } sort dictionary_order qw( perlclib perlxs),
                                                               keys %described_elsewhere;

# The S< > makes things less densely packed, hence more readable
my $has_defs_text .= join ",S< > ", map { "C<$_>" } sort dictionary_order @has_defs;
my $has_r_defs_text .= join ",S< > ", map { "C<$_>" } sort dictionary_order @has_r_defs;
$valid_sections{$genconfig_scn}{footer} =~ s/__HAS_LIST__/$has_defs_text/;
$valid_sections{$genconfig_scn}{footer} =~ s/__HAS_R_LIST__/$has_r_defs_text/;

my $include_defs_text .= join ",S< > ", map { "C<$_>" } sort dictionary_order @include_defs;
$valid_sections{$genconfig_scn}{footer} =~ s/__INCLUDE_LIST__/$include_defs_text/;

my $section_list = join "\n\n", map { "=item L</$_>" } sort dictionary_order keys %valid_sections;

output('perlapi', <<"_EOB_", $docs{api}, \@missing_api, <<"_EOE_");
|=encoding UTF-8
|
|=head1 NAME
|
|perlapi - autogenerated documentation for the perl public API
|
|=head1 DESCRIPTION
|X<Perl API> X<API> X<api>
|
|This file contains most of the documentation of the perl public API, as
|generated by F<embed.pl>.  Specifically, it is a listing of functions,
|macros, flags, and variables that may be used by extension writers.  Besides
|L<perlintern> and F<config.h>, some items are listed here as being actually
|documented in another pod.
|
|L<At the end|/Undocumented functions> is a list of functions which have yet
|to be documented.  Patches welcome!  The interfaces of these are subject to
|change without notice.
|
|Some of the functions documented here are consolidated so that a single entry
|serves for multiple functions which all do basically the same thing, but have
|some slight differences.  For example, one form might process magic, while
|another doesn't.  The name of each variation is listed at the top of the
|single entry.  But if all have the same signature (arguments and return type)
|except for their names, only the usage for the base form is shown.  If any
|one of the forms has a different signature (such as returning C<const> or
|not) every function's signature is explicitly displayed.
|
|Anything not listed here or in the other mentioned pods is not part of the
|public API, and should not be used by extension writers at all.  For these
|reasons, blindly using functions listed in F<proto.h> is to be avoided when
|writing extensions.
|
|In Perl, unlike C, a string of characters may generally contain embedded
|C<NUL> characters.  Sometimes in the documentation a Perl string is referred
|to as a "buffer" to distinguish it from a C string, but sometimes they are
|both just referred to as strings.
|
|Note that all Perl API global variables must be referenced with the C<PL_>
|prefix.  Again, those not listed here are not to be used by extension writers,
|and can be changed or removed without notice; same with macros.
|Some macros are provided for compatibility with the older,
|unadorned names, but this support may be disabled in a future release.
|
|Perl was originally written to handle US-ASCII only (that is characters
|whose ordinal numbers are in the range 0 - 127).
|And documentation and comments may still use the term ASCII, when
|sometimes in fact the entire range from 0 - 255 is meant.
|
|The non-ASCII characters below 256 can have various meanings, depending on
|various things.  (See, most notably, L<perllocale>.)  But usually the whole
|range can be referred to as ISO-8859-1.  Often, the term "Latin-1" (or
|"Latin1") is used as an equivalent for ISO-8859-1.  But some people treat
|"Latin1" as referring just to the characters in the range 128 through 255, or
|sometimes from 160 through 255.
|This documentation uses "Latin1" and "Latin-1" to refer to all 256 characters.
|
|Note that Perl can be compiled and run under either ASCII or EBCDIC (See
|L<perlebcdic>).  Most of the documentation (and even comments in the code)
|ignore the EBCDIC possibility.
|For almost all purposes the differences are transparent.
|As an example, under EBCDIC,
|instead of UTF-8, UTF-EBCDIC is used to encode Unicode strings, and so
|whenever this documentation refers to C<utf8>
|(and variants of that name, including in function names),
|it also (essentially transparently) means C<UTF-EBCDIC>.
|But the ordinals of characters differ between ASCII, EBCDIC, and
|the UTF- encodings, and a string encoded in UTF-EBCDIC may occupy a different
|number of bytes than in UTF-8.
|
|The organization of this document is tentative and subject to change.
|Suggestions and patches welcome
|L<perl5-porters\@perl.org|mailto:perl5-porters\@perl.org>.
|
|The sections in this document currently are
|
|=over

|$section_list
|
|=back
|
|The listing below is alphabetical, case insensitive.
_EOB_
|=head1 AUTHORS
|
|Until May 1997, this document was maintained by Jeff Okamoto
|<okamoto\@corp.hp.com>.  It is now maintained as part of Perl itself.
|
|With lots of help and suggestions from Dean Roehrich, Malcolm Beattie,
|Andreas Koenig, Paul Hudson, Ilya Zakharevich, Paul Marquess, Neil
|Bowers, Matthew Green, Tim Bunce, Spider Boardman, Ulrich Pfeifer,
|Stephen McCamant, and Gurusamy Sarathy.
|
|API Listing originally by Dean Roehrich <roehrich\@cray.com>.
|
|Updated to be autogenerated from comments in the source by Benjamin Stuhl.
|
|=head1 SEE ALSO
|
|F<config.h>, L<perlintern>, $other_places
_EOE_

# List of non-static internal functions
my @missing_guts =
 grep $funcflags{$_}{flags} !~ /[AS]/ && !$docs{guts}{$_}, keys %funcflags;

output('perlintern', <<'_EOB_', $docs{guts}, \@missing_guts, <<"_EOE_");
|=head1 NAME
|
|perlintern - autogenerated documentation of purely B<internal>
|Perl functions
|
|=head1 DESCRIPTION
|X<internal Perl functions> X<interpreter functions>
|
|This file is the autogenerated documentation of functions in the
|Perl interpreter that are documented using Perl's internal documentation
|format but are not marked as part of the Perl API.  In other words,
|B<they are not for use in extensions>!

|It has the same sections as L<perlapi>, though some may be empty.
|
_EOB_
|
|=head1 AUTHORS
|
|The autodocumentation system was originally added to the Perl core by
|Benjamin Stuhl.  Documentation is by whoever was kind enough to
|document their functions.
|
|=head1 SEE ALSO
|
|F<config.h>, L<perlapi>, $other_places
_EOE_