tCMS/lib/Trog/DataModule.pm

package Trog::DataModule;

use 5.040;
use re '/aasx';
use warnings FATAL => 'all';

use FindBin::libs;

use List::Util;
use File::Copy;
use Ref::Util();
use File::Basename qw{basename};
use File::Slurper();
use JSON::MaybeXS();
use JSON::Validator::Schema::Troglodyne();

use Time::HiRes();
use Trog::Themes();

use Trog::Log qw{:all};
use Trog::Utils;
use Trog::Auth();

=head1 QUERY FORMAT

The $query_language and $query_help variables are presented to the user as to how to use the search box in the tCMS header.

=head1 POST STRUCTURE

Posts generally need to have the following:

    data: Brief description of content, or the content itself.
    content_type: What this content actually is.  Used to filter into the appropriate pages.
    href: Primary link.  This is the subject of a news post, or a link to the item itself.  Can be local or remote.
    local_href: Backup link.  Automatically created link to a static cache of the content.
    title: Title of the content.  Used as link name for the 'href' attribute.
    user: User was banned for this post
    id: Internal identifier in datastore for the post.
    tags: array ref of appropriate tags.
    created: timestamp of creation of this version of the post
    version: revision # of this post.

=head1 CONSTRUCTOR

=head2 new(Config::Simple $config)

Try not to do expensive things here.

=cut

sub new ( $class, $config ) {
    $config = $config->vars();
    return bless( $config, $class );
}

=head1 ABSTRACT METHODS

Subclasses must implement all of these; the stubs here do nothing but die.
count() belongs to this set as well, and is documented with the rest of the
querying interface below.

=head2 lang() = STRING $language

The name of the query language this data model understands, shown to the user
beside the search bar.

=head2 help() = STRING $help

Documentation for that query language, shown to the user alongside it.

=head2 read($query) = ARRAYREF $posts

Every post the data model holds, as an arrayref of hashrefs.  Filtering is
get()'s job unless your storage engine can do it more cheaply itself, in which
case honour $query here and override get() too.

=head2 write($posts)

Commit an arrayref of posts to storage.  Called by add() once the posts have
been filtered and validated; don't validate again here.

=head2 tags() = ARRAYREF $tags

Every tag known to the datastore, for building tag pickers.

=cut

#It is required that subclasses implement this
sub lang  { ... }
sub help  { ... }
sub read  { ... }
sub write { ... }
sub count { ... }
sub tags  { ... }

=head1 METHODS

=head2 get(%request)

Queries the data model.  Should return the following:

    id   => Filter down to just the post by ID.  May be subsequently filtered by ACL, resulting in a 404 (which is good, as it does not disclose info).

    version => if id is passed, return the provided post version rather than the most recent one

    tags => ARRAYREF of tags, any one of which is required to give a result.  If none are passed, no filtering is performed.

    acls => ARRAYREF of acl tags, any one of which is required to give result. Filter applies after tags.  'admin' ACL being present skips this filter.

    page => Offset multiplier for pagination.

    limit => Offset for pagination.

    like => Search query, as might be passed in the search bar.

    author => filter by post author

If it is more efficient to filter within your data storage engine, you probably should override this method.
As implemented, this takes the data as a given and filters in post.

=cut

sub get ( $self, %request ) {

    my $posts = $self->read( \%request );
    return @$posts if $request{raw};

    my @filtered = $self->filter( \%request, @$posts );
    @filtered = $self->fixup(@filtered);
    @filtered = $self->paginate( \%request, @filtered );
    return @filtered;
}

=head2 fixup(@posts) = LIST

Fill in what a post needs for display and routing, and does not store: the
author's display_name and user_class, and a local_href, callback and method
where the post has none.  Spaces in its href fields become %20.  Called by
get(), and by a subclass whose get() does not call the base one.

=cut

sub fixup ( $, @filtered ) {

    my %user2display;

    # urlencode spaces in filenames
    @filtered = map {
        my $subj = $_;
        foreach my $param (qw{href preview video_href audio_href local_href wallpaper}) {
            next unless exists $subj->{$param};

            #XXX I don't remember what this fixes, but it also breaks things.  URI::Escape usage instead is indicated.
            $subj->{$param} =~ s/[ ]/%20/g;
        }

        $user2display{ $subj->{user} } //= Trog::Auth::username2display( $subj->{user} );
        $subj->{display_name} = $user2display{ $subj->{user} };

        #XXX Add dynamic routing data for posts which don't have them (/posts/$id) and (/users/$user)
        my $is_user_page = List::Util::any { ( $_ // '' ) eq 'about' } @{ $subj->{tags} };
        if ( !exists $subj->{local_href} ) {
            $subj->{local_href} = "/posts/$subj->{id}";

            #XXX this needs to be correctly populated in the form?
            if ($is_user_page) {
                my $display_name = $user2display{ $subj->{user} };
                die "No display name for user!" unless $display_name;
                $subj->{local_href} = "/users/$display_name";
            }
        }
        if ( !exists $subj->{callback} ) {
            $subj->{callback} = "Trog::Routes::HTML::posts";
            $subj->{callback} = "Trog::Routes::HTML::users" if $is_user_page;
        }

        $subj->{method} = 'GET' unless exists( $subj->{method} );

        $subj->{user_class} = Trog::Auth::username2classname( $subj->{user} );
        $subj
    } @filtered;

    return @filtered;
}

my sub filter_param ( $query, $param, @filtered ) {
    @filtered = grep { ( $_->{$param} || '' ) eq $query->{$param} } @filtered;
    @filtered = dedup_versions( $query->{version}, @filtered );
    return @filtered;
}

# Visibility and acls are both stored as tags, so a caller sees a post when it
# holds one of them.  admin sees everything.
my sub filter_acls ( $query, @filtered ) {
    return @filtered if grep { $_ eq 'admin' } @{ $query->{acls} };
    return grep {
        my $tags = $_->{tags};
        grep {
            my $t = $_;
            grep { $t eq $_ } @{ $query->{acls} }
        } @$tags
    } @filtered;
}

=head2 filter($query, @posts) = @filtered

Apply a get() request's filters to a list of posts: tags, exclude_tags, acls,
visibility, id, title, form, search terms and author.

The ACL check is the load bearing one -- a post is only visible if the caller
holds one of its acls, or the post is public or unlisted.  Callers holding the
'admin' acl skip that filter entirely.

=cut

sub filter ( $, $query, @filtered ) {
    $query->{acls}         //= [];
    $query->{tags}         //= [];
    $query->{exclude_tags} //= [];

    # If an ID or title or acl is passed, just get that (and all it's prior versions),
    # if the caller may see it.
    foreach my $key (qw{id title aclname}) {
        next unless $query->{$key};
        return filter_acls( $query, filter_param( $query, $key, @filtered ) );
    }

    @filtered = dedup_versions( undef, @filtered );

    #Filter out posts which are too old
    #Coerce older into numeric
    if ( $query->{older} ) {
        $query->{older} =~ s/[^0-9]//g;
        @filtered = grep { $_->{created} < $query->{older} } @filtered;
    }
    if ( $query->{newer} ) {
        $query->{newer} =~ s/[^0-9]//g;
        @filtered = grep { $_->{created} > $query->{newer} } @filtered;
    }

    # Filter posts not matching the passed tag(s), if any
    @filtered = grep {
        my $tags = $_->{tags};
        grep {
            my $t = $_;
            grep { ( $t // '' ) eq $_ } @{ $query->{tags} }
        } @$tags
    } @filtered if @{ $query->{tags} };

    # Filter posts *matching* the passed exclude_tag(s), if any
    @filtered = grep {
        my $tags = $_->{tags};
        !grep {
            my $t = $_;
            grep { $t eq $_ } @{ $query->{exclude_tags} }
        } @$tags
    } @filtered if @{ $query->{exclude_tags} };

    @filtered = filter_acls( $query, @filtered );

    @filtered = grep { ( $_->{form} || '' ) eq $query->{form} } @filtered if $query->{form};

    @filtered = grep { $_->{title} =~ m/\Q$query->{like}\E/i || $_->{data} =~ m/\Q$query->{like}\E/i } @filtered if $query->{like};

    @filtered = grep { $_->{user} eq $query->{author} } @filtered if $query->{author};

    return @filtered;
}

=head2 paginate($query, @posts) = @page

The slice of @posts named by the request's page and limit.  Both have to be
present to page at all; limit defaults to 25 when computing the offset.

=cut

sub paginate ( $, $query, @filtered ) {
    return @filtered unless $query->{page} && $query->{limit};

    my $offset = int( $query->{limit} // 25 );
    $offset = @filtered < $offset ? @filtered : $offset;

    # splice() warns rather than simply returning nothing when the offset is
    # past the end, and a page past the end is what a stale bookmark or a
    # walked-off paginator asks for.  Empty is the right answer; the warning is
    # not part of it.
    my $start = ( int( $query->{page} ) - 1 ) * $offset;
    return () if $start > scalar(@filtered);

    return splice( @filtered, $start, $offset );
}

=head2 dedup_versions($version, @posts) = LIST

Given a $version, the posts of that version, each with version_max set to the
newest version among @posts.

Given undef, one post per id: its newest version, with version_max set, created
and author taken from its oldest version, and modified from the newest one.

=cut

sub dedup_versions ( $version = -1, @posts ) {

    #ASSUMPTION made here - if we pass version this is direct ID query
    if ( defined $version ) {
        my $version_max = List::Util::max( map { $_->{version} } @posts );

        return map {
            $_->{version_max} //= $version_max;
            $_
        } grep { $_->{version} eq $version } @posts;
    }

    my @uniqids = List::Util::uniq( map { $_->{id} } @posts );
    my %posts_deduped;
    for my $id (@uniqids) {
        my @ofid        = sort { $b->{version} <=> $a->{version} } grep { $_->{id} eq $id } @posts;
        my $version_max = List::Util::max( map { $_->{version} } @ofid );
        $posts_deduped{$id} = $ofid[0];
        $posts_deduped{$id}{version_max} = $version_max;

        # Show orig creation date, and original author.
        # A modified already present is from an earlier pass, which FlatFile makes
        # in read() before get() filters again; validate() never lets one be stored.
        $posts_deduped{$id}{modified} = $ofid[0]{modified} // $ofid[0]{created};
        $posts_deduped{$id}{created}  = $ofid[-1]{created};
        $posts_deduped{$id}{author}   = $ofid[-1]{author};
    }
    my @deduped = @posts_deduped{@uniqids};

    return @deduped;
}

=head2 count() = INT $num

Returns the total number of posts.
Used to determine paginator parameters.

=cut

=head2 add(@posts) = BOOL $failed_or_not

Add the provided posts to the datastore.
If any post already exists with the same id, a new post with a version higher than it will be added.

Passes an array of new posts to add to the data store module's write() function.

These will have their parameters filtered to those described by the post type's
schema (see schema_for()), and then validated against it.  If the post doesn't
hold up, this dies with an ARRAYREF of validation error strings -- a ref rather
than a string so that Carp::Always doesn't staple a stack trace onto something
we intend to show the user.

You probably won't want to override this.

=cut

# The shape of a post, as an OpenAPIv3 object schema.
#
# This covers the fields every post has regardless of which form produced it.
# Anything a *particular* post type ingests lives in that type's JSON sidecar
# next to its template in the forms directory -- see _schema_for().
our %post_schema = (
    type       => 'object',
    properties => {

        ## Parameters which must be in every single post
        title      => { type => 'string' },
        callback   => { type => 'callback' },
        tags       => { type => 'array',   items   => { type => 'string' } },
        version    => { type => 'integer', minimum => 0 },
        visibility => { type => 'string',  enum    => [qw{public unlisted private}] },
        aliases    => { type => 'array',   items   => { type => 'string' } },

        # title links here
        href => { type => 'string' },

        # Link to post locally
        local_href => { type => 'string' },

        # Post body.  Multi-page types (presentations, invoices) send an array.
        data => {
            oneOf => [
                { type => 'string' },
                { type => 'array', items => { type => 'string' } },
            ],
        },

        # How do I edit this post?
        form => { type => 'string' },

        # Post is restricted to visibility to these ACLs if not public/unlisted
        acls => { type => 'array', items => { type => 'string' } },
        id   => { type => 'string' },

        # Author of the post
        user    => { type => 'string' },
        created => { type => 'integer' },

        # Posts are always GET, but it's stored, so it has to be describable.
        method => { type => 'string' },
    },
);

# Uploads arrive as an HTTP::Body hashref on the way in, and as the href string
# they were turned into on the way back out of the datastore.  Sidecars say
# 'upload' and mean this.
our %upload_schema = (
    oneOf => [
        { type => 'string' },
        { type => 'object' },
    ],
);

=head2 schema_for($form)

Return the OpenAPIv3 schema describing what the named post type ingests: the
base post schema above, with the type's JSON sidecar merged over the top.

Sidecars live beside their templates (blog.tx has blog.json) and are written
either by hand or by the Post Type Wizard.  A post with no form, or a form with
no sidecar, just gets the base schema.

The canned editor blocks a type includes are merged in too, since a form which
splices in preview.tx collects a preview image no matter what its own sidecar
says -- see includes_for() and include_schema_for().  A type's own declaration
of a field beats the block's, and the base schema beats both.

Cached, but keyed on the mtimes of the component dirs rather than memoized
outright: adding or editing a sidecar bumps one of them, so every forked worker
picks the change up on its very next add() with no restart needed.

=cut

# Stat the forms dirs, and the components dirs holding them: writing blog.json
# bumps the mtime of forms/ and nothing above it, while the sidecar beside a
# canned include (preview.json) lives in components/ itself.  Both candidates of
# each, since either one gaining a sidecar changes the answer.
#
# Time::HiRes::stat, because core stat truncates mtime to the second: the post
# type wizard writes a sidecar and something reads it back moments later, and at
# one second of resolution those two are frequently the same generation.  The
# reader then answers out of the cache it filled before the write, which is how
# a type saved and used inside the same second validated against the schema it
# had a moment ago.  Falls back to whatever stat gives where the filesystem
# carries no sub-second timestamps.
my sub sidecar_generation {
    return join( ':', map { ( ( Time::HiRes::stat("$_/forms") )[9] // 0, ( Time::HiRes::stat($_) )[9] // 0 ) } Trog::Themes::template_dirs( 'text/html', 1 ) );
}

# A signature default only covers an absent arg, and plenty of posts have an
# explicitly undef form.
my sub type_of ($form) {
    $form = '' if !defined $form || ref $form;
    my ($type) = $form =~ m/^([A-Za-z0-9_-]+)\.tx$/;
    return $type // '';
}

our %relation_schema = ( type => 'string' );

# Sugar: a sidecar says {"type":"upload"} rather than spelling out the
# string-or-hashref dance every upload field would otherwise need.  Uploads are
# a hashref on the way in from the browser and the href string they became on
# the way back out of the datastore.
my sub expand_pseudo_types ($node) {
    return unless Ref::Util::is_hashref($node);

    my %pseudo = (
        upload   => \%upload_schema,
        relation => \%relation_schema,
    );

    my $type = $node->{type} // '';
    if ( $pseudo{$type} ) {
        delete $node->{type};
        %$node = ( %$node, %{ $pseudo{$type} } );
        return;
    }

    __SUB__->($_) foreach values( %{ $node->{properties} // {} } );
    __SUB__->( $node->{items} ) if $node->{items};
    foreach my $key (qw{oneOf anyOf allOf}) {
        next unless Ref::Util::is_arrayref( $node->{$key} );
        __SUB__->($_) foreach @{ $node->{$key} };
    }
    return;
}

my sub read_sidecar ($path) {

    # -e, not -f: the question is whether a sidecar exists at all, and most
    # post types have none.  Letting the read below fail instead would WARN
    # about every type that simply does not have one.
    return undef unless -e $path;

    local $@;
    my $spec = eval { JSON::MaybeXS::decode_json( File::Slurper::read_text($path) ) };
    if ( !$spec ) {
        WARN("Could not parse post type sidecar '$path': $@");
        return undef;
    }
    return undef unless Ref::Util::is_hashref($spec);

    expand_pseudo_types($spec);
    return $spec;
}

=head2 type_meta_for($form)

The parts of a post type's sidecar which aren't schema: x-tcms-relations,
x-tcms-post-type and friends.

schema_for() deliberately keeps only properties and required, since everything
else would be noise to the validator -- but the relation tables and the post
type's own metadata live out there and something has to read them.

=cut

sub type_meta_for ( $form = '' ) {
    state %cache;

    my $generation = sidecar_generation();
    %cache = () unless exists $cache{$generation};

    my $type = type_of($form);
    return $cache{$generation}{$type} if exists $cache{$generation}{$type};

    my $path    = $type ? Trog::Themes::themed_file_in_dir( 'forms', "$type.json", 'text/html', 1 ) : undef;
    my $sidecar = $path ? read_sidecar($path)                                                       : undef;

    my %meta;
    if ($sidecar) {
        %meta = map { $_ => $sidecar->{$_} } grep { $_ ne 'properties' && $_ ne 'required' && $_ ne 'type' } keys(%$sidecar);
    }

    $cache{$generation}{$type} = \%meta;
    return $cache{$generation}{$type};
}

=head2 include_schema_for($include)

The schema fragment describing what a canned include collects, read from the
JSON sidecar beside it: preview.tx has preview.json, exactly as blog.tx has
blog.json.

A canned include is a block of editor markup which several post types splice in,
and the fields it collects belong to every type which does.  Nothing said so
until this existed, which cost twice over: a hand-written type had to repeat the
declaration in its own sidecar, where the Post Type Wizard could no longer tell
it apart from a field somebody had typed in; and a generated type had no
declaration at all, so validate() dropped the preview image its own editor had
just collected.

Returns an empty hashref for an include which has no sidecar -- which is most of
them, since edit_head.tx and friends collect nothing.

=cut

sub include_schema_for ( $include = '' ) {
    state %cache;

    my $generation = sidecar_generation();
    %cache = () unless exists $cache{$generation};

    # Same shape as a post type, so the same check: a name and nothing else,
    # since it is about to be turned into a path.
    my $name = type_of($include);
    return {} unless $name;
    return $cache{$generation}{$name} if exists $cache{$generation}{$name};

    my $path    = Trog::Themes::themed_component("$name.json");
    my $sidecar = $path ? read_sidecar($path) : undef;

    $cache{$generation}{$name} = $sidecar // {};
    return $cache{$generation}{$name};
}

my sub includes_of ($type) {
    my $path     = Trog::Themes::themed_file_in_dir( 'forms', "$type.tx", 'text/html', 1 );
    my $template = $path ? eval { File::Slurper::read_text($path) } : undef;

    if ( !defined $template ) {
        my $meta     = type_meta_for("$type.tx")->{'x-tcms-post-type'};
        my $recorded = Ref::Util::is_hashref($meta) ? $meta->{includes} : undef;
        return [] unless Ref::Util::is_arrayref($recorded);
        return [ grep { defined && !ref } @$recorded ];
    }

    # Kolon's line syntax, which is how every include in the tree is written --
    # including the ones the wizard generates.  Deduped, since a template is
    # free to splice the same block in twice.
    my %seen;
    return [ grep { !$seen{$_}++ } $template =~ m/^\s*:\s*include\s+"([A-Za-z0-9_-]+\.tx)"/gm ];
}

=head2 includes_for($form)

The canned includes a post type splices into its editor, as an arrayref of
template names in the order they appear.

Read out of the template rather than out of the sidecar, because the template is
what actually renders: a hand-written type never recorded the list anywhere, and
a generated one can still be edited by hand afterwards.  The list the wizard
recorded is the fallback for a type whose template cannot be read at all.

Everything the template includes is returned, not just the blocks the wizard
offers -- it is include_schema_for() which decides whether a given one collects
anything.

=cut

sub includes_for ( $form = '' ) {
    state %cache;

    my $generation = sidecar_generation();
    %cache = () unless exists $cache{$generation};

    my $type = type_of($form);
    return [] unless $type;
    return $cache{$generation}{$type} if exists $cache{$generation}{$type};

    $cache{$generation}{$type} = includes_of($type);
    return $cache{$generation}{$type};
}

=head2 private_fields_for($form)

The fields of a post type which only an editor may see, as declared by
x-tcms-private in the sidecar.

A field is public unless it says otherwise.  The point of declaring it here
rather than guarding it in the template is that the value can then be dropped
before rendering, so no amount of getting a template wrong can leak it.

=cut

sub private_fields_for ( $form = '' ) {
    my $properties = schema_for($form)->{properties} // {};
    return [ grep { $properties->{$_}{'x-tcms-private'} } keys(%$properties) ];
}

=head2 indexed_fields_for($form)

The custom fields of a post type which the Post Type Wizard was told to index,
as declared by x-tcms-indexed in the sidecar.

A field is unindexed unless it says otherwise.  What indexing a field actually
means is up to the data model -- see index_fields(), which most of them ignore.

=cut

sub indexed_fields_for ( $form = '' ) {
    my $properties = schema_for($form)->{properties} // {};

    # Sorted, unlike private_fields_for(), which is only ever asked whether a
    # given field is in it.  This list is collected across every post type and
    # handed to a data model to act on, and an answer that comes back in a
    # different order each run makes a reconciliation impossible to reason about.
    return [ sort grep { $properties->{$_}{'x-tcms-indexed'} } keys(%$properties) ];
}

=head2 index_fields(@fields)

Tell the data model which custom fields ought to be queryable, and let it do
whatever that means for it.

The default is to do nothing at all, and that is a perfectly good answer: a
model with no notion of an index has nothing to build, and the wizard should not
have to know which model it is talking to in order to offer the option.

Called with the full set every time rather than with a delta, so a model that
does act on it can bring itself into line in one pass -- including letting go of
what is no longer asked for.

=cut

sub index_fields { return 0 }

=head2 relations_for($form)

The x-tcms-relations table for a post type, as a hashref keyed on the template
variable the resolved posts land in.  Empty when the type declares none.

Each entry is { form => 'other.tx', from => 'field_name' } -- 'from' naming the
field holding a UUID to resolve, or absent to mean every post of that type.

=cut

sub relations_for ( $form = '' ) {
    my $relations = type_meta_for($form)->{'x-tcms-relations'};
    return {} unless Ref::Util::is_hashref($relations);
    return $relations;
}

sub schema_for ( $form = '' ) {
    state %cache;

    my $generation = sidecar_generation();

    # Only ever keep the current generation around.
    %cache = () unless exists $cache{$generation};

    my $type = type_of($form);
    return $cache{$generation}{$type} if exists $cache{$generation}{$type};

    my %merged = ( %post_schema, properties => { %{ $post_schema{properties} } } );

    # Themed like the templates themselves: a theme that overrides blog.tx can
    # ship its own blog.json alongside it, and one that doesn't still gets the
    # stock schema rather than nothing at all.
    my $path    = $type ? Trog::Themes::themed_file_in_dir( 'forms', "$type.json", 'text/html', 1 ) : undef;
    my $sidecar = $path ? read_sidecar($path)                                                       : undef;
    if ($sidecar) {

        # The base schema is merged *last*, so a sidecar can add fields but can
        # never redefine one of ours.  That matters: a sidecar retyping
        # 'callback' as a plain string would defeat the check that the sub it
        # names actually exists, which is a privilege problem rather than a
        # cosmetic one.
        %{ $merged{properties} } = ( %{ $sidecar->{properties} // {} }, %{ $merged{properties} } );

        # A sidecar may insist on its own fields, but it can't relax anything
        # the base schema already demands.
        my @required = List::Util::uniq( @{ $post_schema{required} // [] }, @{ $sidecar->{required} // [] } );

        # An empty required list isn't legal OpenAPI, so don't emit one.
        $merged{required} = \@required if @required;
    }

    # And what the canned blocks this type's editor splices in collect.  A form
    # which includes preview.tx collects a preview image whether or not anybody
    # wrote that down in its sidecar, and validate() drops every field the schema
    # does not describe -- so until this was merged in, a post type generated by
    # the wizard threw its own preview upload away on the way to the datastore.
    #
    # Merged under both of the above: only the properties, since it is the type
    # rather than a block it borrows which gets to say what it insists on, and
    # last of the three in precedence, so a type declaring one of these itself
    # still wins.
    foreach my $include ( @{ includes_for($form) } ) {
        my $properties = include_schema_for($include)->{properties};
        next unless Ref::Util::is_hashref($properties);
        %{ $merged{properties} } = ( %$properties, %{ $merged{properties} } );
    }

    $cache{$generation}{$type} = \%merged;
    return $cache{$generation}{$type};
}

# A relation is stored as the referenced post's UUID, and resolved into the
# post itself at render time -- see the x-tcms-relations table a sidecar
# declares alongside it, and Trog::Routes::HTML::enrich_post.  The pseudo-type
# exists so a sidecar can say what a field *means* rather than just that it
# happens to be a string.

=head2 validate($post)

Filter a post down to the fields its type actually describes, then run what's
left through the schema.

Returns the list of validation errors, which is empty when the post is good.
Note that filtering happens first and silently: the query hash handed to us by
the router is full of things which aren't post fields at all, and never was
meant to round-trip.

=cut

my sub validator {
    state $validator;
    $validator //= JSON::Validator::Schema::Troglodyne->new();
    return $validator;
}

sub validate ($post) {
    my $validator = validator();

    my $schema = schema_for( $post->{form} );

    foreach my $key ( keys(%$post) ) {

        # Drop everything the schema doesn't describe.
        my $property = $schema->{properties}{$key};
        if ( !$property ) {
            delete $post->{$key};
            next;
        }

        # An untouched form input submits the empty string, which is HTML for
        # "I wasn't filled in" rather than a number, a boolean or a file.  We
        # leave it alone for actual string fields, which is what got stored
        # before any of this was typed.
        delete $post->{$key}
          if defined $post->{$key}
          && !ref $post->{$key}
          && $post->{$key} eq ''
          && ( $property->{type} // '' ) ne 'string';
    }

    # OpenAPIv3 coerces as it goes, so "4" lands in the datastore as 4 and
    # checkbox values land as real booleans.
    return $validator->validate( $post, $schema );
}

=head2 validate_built($post)

Check the post that is actually about to be written.

validate() runs on what was submitted.  process() then builds the shape that
gets stored -- it turns a visibility into a tag, folds the acls in, resolves
uploads into hrefs -- and until now nothing looked at the result.  That is how
every post saved without a visibility came to carry a null in its tags: the
schema describes tags as an array of strings and would have rejected it on the
spot, but it was never asked.

Checks rather than filters, unlike validate().  process() adds fields the base
schema has never heard of -- content_type, is_video, attachments -- and dropping
what it does not recognise would throw away the work it just did.  So this looks
only at the fields the schema actually describes, on a copy, and changes nothing.

Returns the list of validation errors, empty when the post is fit to store.

=cut

sub validate_built ($post) {
    my $schema = schema_for( $post->{form} );

    my %described = map { $_ => $post->{$_} } grep { exists $schema->{properties}{$_} } keys(%$post);

    return validator()->validate( \%described, $schema );
}

sub add ( $self, @posts ) {
    my @to_write;

    foreach my $post (@posts) {

        # Filter to what this post type actually describes, then validate it.
        my @errors = validate($post);

        # An arrayref rather than a string on purpose.  These errors get shown
        # to whoever submitted the post, and Carp::Always decorates a string
        # die with a stack trace they have no use for.
        die [ map { "$_" } @errors ] if @errors;

        # ||= rather than //= for the identity and routing scalars.  An
        # untouched hidden input submits the empty string, and validate() keeps
        # an empty string for a field the schema types as a string -- which id,
        # local_href, callback and method all are.  //= then saw a defined value
        # and left it, so a post created through any editor arrived here with
        # id => '' and the flat file model wrote it to 'data/files/', which is
        # the datastore directory rather than a post in it.
        $post->{id} ||= Trog::Utils::uuid();
        $post->{aliases} //= [];
        $post->{aliases} = [ $post->{aliases} ] unless ref $post->{aliases} eq 'ARRAY';

        if ( $post->{aclname} ) {

            # Then this is a series
            $post->{local_href} ||= "/$post->{aclname}";
            push( @{ $post->{aliases} }, "/posts/$post->{id}", "/series/$post->{id}" );
        }

        # Every post needs one: process pushes it into the tags, and an
        # undef there is a tag no query will ever match, which makes the post
        # invisible to everyone but an admin.  A post type built without a
        # visibility selector would otherwise produce exactly that.  Private
        # rather than public, because guessing wrong in the other direction
        # publishes something nobody asked to publish.
        $post->{visibility} ||= 'private';

        $post->{callback} ||= 'Trog::Routes::HTML::posts';

        # If this is a user creation post, add in the /user/ route
        if ( $post->{callback} eq 'Trog::Routes::HTML::users' ) {
            $post->{local_href} ||= "/users/$post->{display_name}";
            $post->{title}      ||= $post->{display_name};
        }

        $post->{local_href} ||= "/posts/$post->{id}";
        $post->{method}     ||= 'GET';
        $post->{created} = time();
        my @existing_posts = $self->get( id => $post->{id}, acls => [qw{admin}] );
        if (@existing_posts) {
            my $existing_post = $existing_posts[0];
            $post->{version} = $existing_post->{version};
            $post->{version}++;
        }
        $post->{version} //= 0;

        #XXX if local_href has /secure paths in it, fix this.  We don't want to save that.
        if ( index( $post->{local_href}, '/secure/' ) == 0 ) {
            $post->{local_href} =~ s|/secure||gmx;
        }

        $post = process($post);

        # And check what process() built, not just what arrived.  A post that
        # fails here is one this code has made wrong on its way to disk, so it
        # is a bug rather than bad input -- but the answer is still to refuse to
        # write it, because the alternative is what we did before, which was to
        # write it and find out years later.
        my @built = validate_built($post);
        if (@built) {
            WARN( "Refusing to store a post this built wrong: " . join( ', ', map { "$_" } @built ) );
            die [ map { "$_" } @built ];
        }

        push @to_write, $post;
    }
    $self->write( \@to_write );

    return 0;
}

my %seen_files;

my sub handle_upload ( $file, $uuid, $private = 0 ) {
    my $fname;
    if ( ref $file ne 'HASH' ) {
        $file =~ s/\\/\//g;
        $fname = basename($file);
        FATAL("Did not expect bogus file before its real counterpart!") unless $seen_files{$fname};
        return $seen_files{$fname};
    }
    my $f = $file->{tempname};
    $fname = basename( $file->{filename} );
    my $newname = "$uuid.$file->{filename}";
    $newname = "private/$newname" if $private;
    File::Copy::move( $f, "www/assets/$newname" );

    # Regrettably, we have no control over the perms of the temp file that starman creates for uploads.
    chmod( 0755, "www/assets/$newname" );
    $seen_files{$fname} = "/assets/$newname";
    return $seen_files{$fname};
}

#XXX this level of post-processing seems gross, but may be unavoidable
# Not actually a subprocess, kek

=head2 process($post) = HASHREF

Build the post that gets stored from the one that was submitted.  It stores
the uploaded files, folds the visibility, the acls and the series into the
tags, removes the fields that only the form uses, and sets the content type
fields from the files the post points at.  Changes $post, and returns it.

=cut

sub process ($post) {

    # Here as well as in add(), because this is the sub that turns a visibility
    # into a tag: without one it pushed an undef into tags, which is a tag no
    # query ever matches, so the post went invisible to everyone but an admin.
    # Nothing complained, and the bunk tag was written to disk and stayed there.
    $post->{visibility} ||= 'private';

    # If the post is private, make sure it's associated assets are too.
    my $is_private_post = $post->{visibility} eq 'private';

    $post->{href}      = handle_upload( $post->{file},           $post->{id}, $is_private_post ) if $post->{file};
    $post->{preview}   = handle_upload( $post->{preview_file},   $post->{id}, $is_private_post ) if $post->{preview_file};
    $post->{wallpaper} = handle_upload( $post->{wallpaper_file}, $post->{id}, $is_private_post ) if $post->{wallpaper_file};
    @{ $post->{attachments} } = map { handle_upload( $_, $post->{id}, $is_private_post ) } @{ $post->{attachments} } if $post->{attachments};
    $post->{preview} = $post->{href} if $post->{app} && $post->{app} eq 'image';
    delete $post->{app};
    delete $post->{file};
    delete $post->{preview_file};
    delete $post->{wallpaper_file};

    delete $post->{scheme};
    delete $post->{route};
    delete $post->{domain};

    # Handle acls/tags
    $post->{tags} //= [];
    $post->{acls} //= [];
    @{ $post->{tags} } = grep {
        my $subj = $_;
        !grep { $_ eq $subj } qw{public private unlisted}
    } @{ $post->{tags} };
    push( @{ $post->{tags} }, grep { defined } @{ $post->{acls} } ) if $is_private_post;
    delete $post->{acls};
    push( @{ $post->{tags} }, $post->{visibility} );

    # Add the 'series' tag if we are in a series, restrict to relevant acl
    if ( $post->{series} ) {
        push( @{ $post->{tags} }, 'series' );
        push( @{ $post->{tags} }, $post->{series} );
    }

    #Filter adding the same acl twice, and anything that isn't a tag at all --
    # an undef here is what every query below compares against, and it warns
    # its way through the lot before matching nothing.
    @{ $post->{tags} }    = List::Util::uniq( grep { $_ } @{ $post->{tags} } );
    @{ $post->{aliases} } = List::Util::uniq( grep { $_ } @{ $post->{aliases} } );

    # Handle multimedia content types
    $post->{content_type}       = Trog::Utils::mime_type("www/$post->{href}")       if $post->{href};
    $post->{video_content_type} = Trog::Utils::mime_type("www/$post->{video_href}") if $post->{video_href};
    $post->{audio_content_type} = Trog::Utils::mime_type("www/$post->{audio_href}") if $post->{audio_href};
    $post->{content_type} ||= 'text/html';

    $post->{is_video}   = 1 if $post->{content_type} =~ m/^video\//;
    $post->{is_audio}   = 1 if $post->{content_type} =~ m/^audio\//;
    $post->{is_image}   = 1 if $post->{content_type} =~ m/^image\//;
    $post->{is_profile} = 1 if grep { $_ eq 'about' } @{ $post->{tags} };

    return $post;
}

# Browsers try and save time in the event that the same file is sent twice, so handle that.

=head2 delete(@posts)

Delete the following posts.
Will remove all versions of said post.

You should override this, it is a stub here.

=cut

sub delete { die 'stub' }

=head2 routes() = HASH

Returns the routes to each post.
You should override this for performance reasons, as it's just a wrapper around get() by defualt.

=cut

sub routes ($self) {
    my %routes = map { $_->{local_href} => { method => $_->{method}, callback => \&{ $_->{callback} } } } ( $self->get( limit => 0, acls => ['admin'] ) );
    return %routes;
}

=head2 aliases() = HASH

Returns the aliases for each post, indexed by aliases.
You should override this for performance reasons, as it's just a wrapper around get() by defualt.

=cut

sub aliases ($self) {
    my @posts = $self->get( limit => 0, acls => ['admin'] );
    my %aliases;
    foreach my $post (@posts) {
        $aliases{$_} = $post->{local_href} for @{ $post->{aliases} // [] };
    }
    return %aliases;
}

1;