Statum/tools/statum-mkpstm/statum-mkpstm.vala

using Xml;
using Html;
using Astralis;
using Invercargill;
using Invercargill.DataStructures;

namespace Statum.Tools {

    /**
     * `statum-mkpstm` — the build-time Statum pre-rendering pipeline.
     *
     * Turns an app HTML page (plus its templates/fragments) into a Vala
     * {@link Statum.StatumPage} subclass: a static, precompressed document whose
     * route comes from its `<pstm-uri>`. Composition (templates, fragments,
     * `pstm-res` rewriting) is applied, then the result is precompressed with
     * identity/gzip/zstd/br.
     */
    public class Mkpstm : Object {

        private static string? output_file = null;
        private static string? namespace_name = null;
        private static string? class_name_override = null;
        private static string? route_override = null;
        private static List<string> template_dirs = new List<string>();

        private Dictionary<string, Xml.Doc*> templates_by_name = new Dictionary<string, Xml.Doc*>();
        private Dictionary<string, Xml.Doc*> fragments_by_name = new Dictionary<string, Xml.Doc*>();

        /**
         * The Statum client ships as four modules with a strict load order
         * (core first: every later module registers into the seams core
         * installs). A legacy single-file `pstm-res-src="statum.js"` reference
         * expands at build time into exactly these four tags.
         */
        private const string[] CLIENT_MODULE_ORDER = {
            "statum-core.js", "statum-document.js", "statum-markdown.js", "statum-forms.js"
        };

        private const OptionEntry[] options = {
            { "output", 'o', 0, OptionArg.FILENAME, ref output_file, "Output Vala file name", "FILE" },
            { "ns", '\0', 0, OptionArg.STRING, ref namespace_name, "Namespace for the generated class", "NAMESPACE" },
            { "name", 'n', 0, OptionArg.STRING, ref class_name_override, "Generated class name (default: <InputName>Page)", "NAME" },
            { "route", 'r', 0, OptionArg.STRING, ref route_override, "Page route (default: from <pstm-uri>)", "ROUTE" },
            { "template-dir", 't', 0, OptionArg.FILENAME_ARRAY, ref template_dirs, "Directory to search for templates/fragments (repeatable)", "DIR" },
            { null }
        };

        public static int main(string[] args) {
            try {
                var opt = new OptionContext("PAGE_FILE - Generate a Statum StatumPage");
                opt.add_main_entries(options, null);
                opt.parse(ref args);
            } catch (OptionError e) {
                stderr.printf("Error: %s\n", e.message);
                return 1;
            }

            string? page_file = null;
            if (args.length > 1) {
                page_file = args[1];
            }
            if (page_file == null) {
                stderr.printf("Error: No page file specified.\n");
                return 1;
            }

            try {
                var tool = new Mkpstm();
                var generated = tool.run((!)page_file);
                var out_path = tool.output_file ?? (make_pascal_case((!)page_file) + "Page.vala");
                write_file(out_path, generated);
                stdout.printf("Generated: %s\n", out_path);
                return 0;
            } catch (GLib.Error e) {
                stderr.printf("Error: %s\n", e.message);
                return 1;
            }
        }

        public string run(string page_file) throws GLib.Error {
            if (template_dirs.length() == 0) {
                var parent = File.new_for_path(page_file).get_parent();
                template_dirs.append(parent != null ? ((!)parent).get_path() : ".");
            }

            load_templates_and_fragments(page_file);

            var doc = parse_html_file(page_file);
            validate_document(doc, page_file);
            validate_head_directives(doc, page_file);

            // Capture the page's routes before composition (the <pstm-uri>
            // directive is dropped while merging templates).
            var routes = read_routes(doc);

            doc = compose_templates(doc);
            inline_fragments(doc);
            expand_legacy_client(doc, page_file);
            rewrite_resources(doc);

            var variant = build_variant(doc);
            var class_name = class_name_override ?? (make_pascal_case(page_file) + "Page");
            var route = route_override ?? (routes.length > 0 ? routes[0] : "");

            return emit(class_name, route, variant);
        }

        // ------------------------------------------------------------------
        // Template/fragment loading
        // ------------------------------------------------------------------

        private void load_templates_and_fragments(string page_file) throws GLib.Error {
            foreach (var dir in template_dirs) {
                var dir_file = File.new_for_path(dir);
                if (!dir_file.query_exists()) {
                    continue;
                }
                var enumerator = dir_file.enumerate_children("standard::name", 0);
                FileInfo info;
                while ((info = enumerator.next_file()) != null) {
                    var name = info.get_name();
                    if (!name.has_suffix(".html") || name.has_prefix(".")) {
                        continue;
                    }
                    var path = Path.build_filename(dir, name);
                    if (path == page_file) {
                        continue;
                    }
                    var tdoc = parse_html_file(path);
                    var source = Path.build_filename(dir, name);
                    validate_document(tdoc, source);
                    validate_head_directives(tdoc, source);
                    var tname = read_pstm_name(tdoc);
                    if (tname == null) {
                        continue;
                    }
                    if (find_first_element(tdoc->get_root_element(), "pstm-content") != null) {
                        templates_by_name.set((!)tname, tdoc);
                    } else {
                        fragments_by_name.set((!)tname, tdoc);
                    }
                }
            }
        }

        // ------------------------------------------------------------------
        // Composition
        // ------------------------------------------------------------------

        private Xml.Doc* compose_templates(Xml.Doc* doc) throws GLib.Error {
            Xml.Doc* current_doc = doc;
            string merged_title = read_directive_in_doc(current_doc, "pstm-title") ?? "";

            // Guard against cyclic template chains (A templates B templates A).
            var visited = new HashSet<string>();

            while (true) {
                var parent_name = read_directive_in_doc(current_doc, "pstm-template");
                if (parent_name == null || parent_name.length == 0) {
                    break;
                }
                if (visited.contains((!)parent_name)) {
                    break;
                }
                visited.add((!)parent_name);

                Xml.Doc* parent_doc;
                if (!templates_by_name.try_get((!)parent_name, out parent_doc)) {
                    throw new IOError.FAILED(@"Unknown template \"$((!)parent_name)\"");
                }

                // Merge this document's <head> into the parent's <head> (so
                // pages, templates and nested components can all declare head
                // tags), then splice its <body> into the parent's content outlet.
                merge_head(current_doc, parent_doc);
                splice_body(current_doc, parent_doc);

                var parent_title = read_directive_in_doc(parent_doc, "pstm-title");
                if (parent_title != null && parent_title.length > 0) {
                    merged_title = merged_title + (!)parent_title;
                }

                current_doc = parent_doc;
            }

            apply_title(current_doc, merged_title);
            return current_doc;
        }

        /**
         * Copies the children of `child_doc`'s <head> into `parent_doc`'s <head>,
         * skipping the consumed structural directives (pstm-template/name/title/
         * uri) which are page-level metadata and must not be carried up.
         */
        private void merge_head(Xml.Doc* child_doc, Xml.Doc* parent_doc) {
            var child_head = find_head(child_doc);
            var parent_head = find_head(parent_doc);
            if (child_head == null || parent_head == null) {
                return;
            }
            for (var child = child_head->children; child != null; child = child->next) {
                if (!is_element(child)) {
                    continue;
                }
                var name = (string) child->name;
                if (name == "pstm-template" || name == "pstm-name" || name == "pstm-title" || name == "pstm-uri") {
                    continue;
                }
                var copied = child->doc_copy(parent_doc, 1);
                if (copied != null) {
                    parent_head->add_child(copied);
                }
            }
        }

        /**
         * Splices `child_doc`'s <body> into `parent_doc`'s <pstm-content> outlet
         * (or appends to the parent's <body> when there is no outlet). When the
         * child <body> carries `pstm-materialise-as`, it is wrapped in an element
         * of that tag (carrying the body's non-pstm attributes); otherwise its
         * children are spliced directly.
         */
        private void splice_body(Xml.Doc* child_doc, Xml.Doc* parent_doc) {
            var child_body = find_body(child_doc);
            var parent_body = find_body(parent_doc);
            if (child_body == null || parent_body == null) {
                return;
            }
            var outlet = find_first_element(parent_body, "pstm-content");
            var materialise_as = child_body->get_prop("pstm-materialise-as");

            if (materialise_as != null) {
                var wrapper = parent_body->new_child(null, (!)materialise_as, null);
                wrapper->unlink();
                copy_non_pstm_attrs(child_body, wrapper);
                for (var child = child_body->children; child != null; child = child->next) {
                    if (is_structural_directive(child)) {
                        continue;
                    }
                    var copied = child->doc_copy(parent_doc, 1);
                    if (copied != null) {
                        wrapper->add_child(copied);
                    }
                }
                if (outlet != null) {
                    outlet->add_prev_sibling(wrapper);
                } else {
                    parent_body->add_child(wrapper);
                }
            } else {
                for (var child = child_body->children; child != null; child = child->next) {
                    if (is_structural_directive(child)) {
                        continue;
                    }
                    var copied = child->doc_copy(parent_doc, 1);
                    if (copied == null) {
                        continue;
                    }
                    if (outlet != null) {
                        outlet->add_prev_sibling(copied);
                    } else {
                        parent_body->add_child(copied);
                    }
                }
            }

            if (outlet != null) {
                outlet->unlink();
            }
        }

        /** Copies an element's non-`pstm-*` attributes onto another element. */
        private static void copy_non_pstm_attrs(Xml.Node* src, Xml.Node* dst) {
            for (var attr = src->properties; attr != null; attr = attr->next) {
                var attr_name = (string) attr->name;
                if (attr_name.has_prefix("pstm-")) {
                    continue;
                }
                var value = src->get_prop(attr_name);
                if (value != null) {
                    dst->set_prop(attr_name, (!)value);
                }
            }
        }

        private static string? read_directive_in_doc(Xml.Doc* doc, string name) {
            var root = doc->get_root_element();
            if (root == null) {
                return null;
            }
            return read_directive_text(root, name);
        }

        private static bool is_structural_directive(Xml.Node* node) {
            if (!is_element(node)) {
                return false;
            }
            var name = (string) node->name;
            return name == "pstm-template" || name == "pstm-name" || name == "pstm-title" || name == "pstm-uri" || name == "pstm-content";
        }

        /** Fragment-body metadata elements skipped when inlining (e.g. pstm-input). */
        private static bool is_fragment_metadata(Xml.Node* node) {
            if (!is_element(node)) {
                return false;
            }
            var name = (string) node->name;
            return name == "pstm-input" || name.has_prefix("pstm-");
        }

        private void inline_fragments(Xml.Doc* doc) {
            bool changed = true;
            while (changed) {
                changed = false;
                var includes = find_all_elements(doc, "pstm-fragment");
                foreach (var include in includes) {
                    if (inline_one_fragment(doc, include)) {
                        changed = true;
                    }
                }
            }
        }

        private bool inline_one_fragment(Xml.Doc* doc, Xml.Node* include) {
            var name = include->get_prop("name");
            if (name == null) {
                return false;
            }
            Xml.Doc* fragment_doc;
            if (!fragments_by_name.try_get((!)name, out fragment_doc)) {
                return false;
            }

            var fragment_body = find_body(fragment_doc);
            if (fragment_body == null) {
                return false;
            }

            // Merge the fragment's <head> into the document's <head>.
            merge_head(fragment_doc, doc);

            var inputs = collect_input_names(fragment_body);
            var include_materialise = include->get_prop("pstm-materialise-as");
            var fragment_materialise = fragment_body->get_prop("pstm-materialise-as");
            var materialise_as = include_materialise ?? fragment_materialise;
            bool wrap = (materialise_as != null) || inputs.length > 0;
            string tag = materialise_as ?? "div";

            if (wrap) {
                var parent = include->parent;
                if (parent == null) {
                    return false;
                }
                var wrapper = parent->new_child(null, tag, null);
                wrapper->unlink();
                include->add_prev_sibling(wrapper);
                // Carry the fragment <body>'s attributes when it supplied the tag.
                if (include_materialise == null && fragment_materialise != null) {
                    copy_non_pstm_attrs(fragment_body, wrapper);
                }
                for (var child = fragment_body->children; child != null; child = child->next) {
                    if (is_fragment_metadata(child)) {
                        continue;
                    }
                    var copied = child->doc_copy(doc, 1);
                    if (copied != null) {
                        wrapper->add_child(copied);
                    }
                }
                foreach (var input_name in inputs) {
                    var expr = include->get_prop(input_name);
                    if (expr != null) {
                        wrapper->set_prop(@"stm-def-$input_name-as", (!)expr);
                    }
                }
            } else {
                for (var child = fragment_body->children; child != null; child = child->next) {
                    if (is_fragment_metadata(child)) {
                        continue;
                    }
                    var copied = child->doc_copy(doc, 1);
                    if (copied != null) {
                        include->add_prev_sibling(copied);
                    }
                }
            }

            include->unlink();
            return true;
        }

        private string[] collect_input_names(Xml.Node* fragment_body) {
            var names = new string[0];
            for (var n = fragment_body->children; n != null; n = n->next) {
                if (is_element(n) && (string) n->name == "pstm-input") {
                    var input_name = n->get_prop("name");
                    if (input_name != null) {
                        names += (!)input_name;
                    }
                }
            }
            return names;
        }

        private void rewrite_resources(Xml.Doc* doc) {
            var root = doc->get_root_element();
            if (root != null) {
                rewrite_resources_in(root);
            }
        }

        /**
         * Expands a legacy single-file client reference into the four ordered
         * module tags.
         *
         * `<script pstm-res-src="statum.js">` (the pre-split client, no longer
         * an embedded resource) becomes one `<script src="/_statum/resource/
         * statum-<mod>.js">` tag per module, in canonical load order. Later
         * duplicate `statum.js` references on the same page are dropped, and
         * explicit `pstm-res-src="statum-<mod>.js"` references the expansion
         * already covers are removed (subsumed) so every module appears exactly
         * once. Pages referencing only module files (or only plain `src="…"`
         * scripts, which are already final) are left entirely alone.
         *
         * Runs on the composed document before {@link rewrite_resources}: the
         * inserted tags use plain `src` attributes, so the generic rewrite and
         * `strip_pstm` both leave them alone.
         */
        private void expand_legacy_client(Xml.Doc* doc, string page_file) {
            var root = doc->get_root_element();
            if (root == null) {
                return;
            }

            Xml.Node* legacy = null;
            var refs = new ClientRefs();
            collect_client_refs(root, refs);
            legacy = refs.legacy;
            if (legacy == null) {
                return;
            }

            var parent = legacy->parent;
            if (parent == null) {
                return;
            }

            foreach (var module in CLIENT_MODULE_ORDER) {
                var tag = ((!)parent)->new_child(null, "script", null);
                tag->unlink();
                // Script tags carrying pstm-res-src carry no other attributes
                // today; copy any extras onto all four for good measure.
                copy_non_pstm_attrs(legacy, tag);
                tag->set_prop("src", @"/_statum/resource/$module");
                legacy->add_prev_sibling(tag);
            }

            string[] subsumed = new string[0];
            foreach (var node in refs.modules) {
                var name = node->get_prop("pstm-res-src");
                if (name != null) {
                    subsumed += (!)name;
                }
                node->unlink();
            }
            legacy->unlink();

            // Exactly one build warning per page, naming the page and the
            // expansion.
            var page = page_name(page_file);
            stderr.printf("warning: %s: expanding legacy pstm-res-src=\"statum.js\" into %s",
                page, string.joinv(", ", CLIENT_MODULE_ORDER));
            if (subsumed.length > 0) {
                stderr.printf("; subsumed explicit module tag(s): %s", string.joinv(", ", subsumed));
            }
            if (refs.duplicates > 0) {
                stderr.printf("; dropped %d duplicate statum.js reference(s)", refs.duplicates);
            }
            stderr.printf("\n");
        }

        /**
         * Collects the client script references of a composed document: the
         * first `<script pstm-res-src="statum.js">` (if any), every explicit
         * module `pstm-res-src="statum-<mod>.js"` reference, and the count of
         * further `statum.js` duplicates.
         */
        private void collect_client_refs(Xml.Node* node, ClientRefs refs) {
            if (!is_element(node)) {
                return;
            }
            if ((string) node->name == "script") {
                var res = node->get_prop("pstm-res-src");
                if (res != null) {
                    if ((!)res == "statum.js") {
                        if (refs.legacy == null) {
                            refs.legacy = node;
                        } else {
                            refs.duplicates++;
                        }
                    } else if (is_client_module((!)res)) {
                        refs.modules += node;
                    }
                }
            }
            for (var child = node->children; child != null; child = child->next) {
                collect_client_refs(child, refs);
            }
        }

        /** The page's short name, as warnings name it (e.g. `home`). */
        private static string page_name(string page_file) {
            var page = Path.get_basename(page_file);
            var dot = page.last_index_of(".");
            if (dot > 0) {
                page = page.substring(0, dot);
            }
            return page;
        }

        private static bool is_client_module(string name) {
            foreach (var module in CLIENT_MODULE_ORDER) {
                if (name == module) {
                    return true;
                }
            }
            return false;
        }

        private void rewrite_resources_in(Xml.Node* node) {
            if (!is_element(node)) {
                return;
            }

            string[] to_add_attr = new string[0];
            string[] to_add_val = new string[0];
            string[] to_remove = new string[0];
            for (var attr = node->properties; attr != null; attr = attr->next) {
                var attr_name = (string) attr->name;
                if (attr_name.has_prefix("pstm-res-")) {
                    var target_attr = attr_name.substring("pstm-res-".length);
                    var value = node->get_prop(attr_name);
                    if (value != null) {
                        to_add_attr += target_attr;
                        to_add_val += @"/_statum/resource/$((!)value)";
                    }
                    to_remove += attr_name;
                }
            }
            for (int i = 0; i < to_add_attr.length; i++) {
                node->set_prop(to_add_attr[i], to_add_val[i]);
            }
            foreach (var attr_name in to_remove) {
                node->unset_prop(attr_name);
            }

            for (var child = node->children; child != null; child = child->next) {
                rewrite_resources_in(child);
            }
        }

        // ------------------------------------------------------------------
        // Variant rendering
        // ------------------------------------------------------------------

        /**
         * Renders the composed document to its single HTML form (stripping every
         * `pstm-*` directive) and precompresses it. Pages no longer have
         * variants: a page is a static, precompressed document, so there is
         * exactly one.
         */
        private VariantData build_variant(Xml.Doc* doc) throws GLib.Error {
            strip_pstm(doc);
            var html = serialize(doc);
            return compress_variant(html);
        }

        private VariantData compress_variant(string html) throws GLib.Error {
            var data = new VariantData();
            data.identity = html.data;
            var buffer = new ByteBuffer.from_byte_array(html.data);

            var gzip = new GzipCompressor(9).compress_buffer(buffer, null).to_array();
            var zstd = new ZstdCompressor(19).compress_buffer(buffer, null).to_array();
            var br = new BrotliCompressor(11).compress_buffer(buffer, null).to_array();

            data.has_gzip = gzip.length < html.length;
            data.has_zstd = zstd.length < html.length;
            data.has_br = br.length < html.length;
            data.gzip = data.has_gzip ? gzip : null;
            data.zstd = data.has_zstd ? zstd : null;
            data.br = data.has_br ? br : null;
            return data;
        }

        /** Remove every `pstm-*` attribute and `pstm-*` element from the tree. */
        private void strip_pstm(Xml.Doc* doc) {
            var root = doc->get_root_element();
            if (root != null) {
                strip_pstm_from(root);
            }
        }

        private void strip_pstm_from(Xml.Node* node) {
            string[] to_remove = new string[0];
            for (var attr = node->properties; attr != null; attr = attr->next) {
                var attr_name = (string) attr->name;
                if (attr_name.has_prefix("pstm-")) {
                    to_remove += attr_name;
                }
            }
            foreach (var attr_name in to_remove) {
                node->unset_prop(attr_name);
            }

            var child = node->children;
            while (child != null) {
                var next = child->next;
                if (is_element(child) && ((string) child->name).has_prefix("pstm-")) {
                    child->unlink();
                } else if (is_element(child)) {
                    strip_pstm_from(child);
                }
                child = next;
            }
        }

        // ------------------------------------------------------------------
        // Code generation
        // ------------------------------------------------------------------

        private string emit(string class_name, string route, VariantData variant) throws GLib.Error {
            var page_hash = compute_hash(variant.identity);

            var b = new StringBuilder();
            string i1 = namespace_name != null ? "    " : "";
            string i2 = namespace_name != null ? "        " : "    ";
            string i3 = namespace_name != null ? "            " : "        ";

            b.append("using Statum;\n");
            b.append("using Invercargill;\n");
            b.append("using Invercargill.DataStructures;\n");
            b.append("using Astralis;\n\n");

            if (namespace_name != null) {
                b.append_printf("namespace %s {\n\n", namespace_name);
            }

            b.append_printf("%s// Generated by statum-mkpstm\n", i1);
            b.append_printf("%spublic class %s : StatumPage {\n\n", i1, class_name);

            b.append_printf("%spublic override string route { get { return \"%s\"; } }\n", i2, escape(route));
            b.append_printf("%spublic override string content_type { get { return \"text/html; charset=utf-8\"; } }\n\n", i2);

            // get_best_encoding: smallest available supported encoding.
            b.append_printf("%spublic override string get_best_encoding(Set<string> supported) {\n", i2);
            if (variant.has_br) {
                b.append_printf("%sif (supported.has(\"br\")) return \"br\";\n", i3);
            }
            if (variant.has_zstd) {
                b.append_printf("%sif (supported.has(\"zstd\")) return \"zstd\";\n", i3);
            }
            if (variant.has_gzip) {
                b.append_printf("%sif (supported.has(\"gzip\")) return \"gzip\";\n", i3);
            }
            b.append_printf("%sreturn \"identity\";\n", i3);
            b.append_printf("%s}\n\n", i2);

            // get_encoding
            b.append_printf("%spublic override unowned uint8[] get_encoding(string encoding) {\n", i2);
            b.append_printf("%sswitch (encoding) {\n", i3);
            b.append_printf("%scase \"identity\": return IDENTITY;\n", i3);
            if (variant.has_gzip && variant.gzip != null) {
                b.append_printf("%scase \"gzip\": return GZIP;\n", i3);
            }
            if (variant.has_zstd && variant.zstd != null) {
                b.append_printf("%scase \"zstd\": return ZSTD;\n", i3);
            }
            if (variant.has_br && variant.br != null) {
                b.append_printf("%scase \"br\": return BR;\n", i3);
            }
            b.append_printf("%sdefault: return IDENTITY;\n", i3);
            b.append_printf("%s}\n", i3);
            b.append_printf("%s}\n\n", i2);

            // get_etag_for (ETag = "<page-hash>-<encoding>")
            b.append_printf("%spublic override string get_etag_for(string encoding) {\n", i2);
            b.append_printf("%sreturn \"\\\"%s-%%s\\\"\".printf(encoding);\n", i3, page_hash);
            b.append_printf("%s}\n\n", i2);

            // Byte arrays.
            emit_one_array(b, i2, i3, "IDENTITY", variant.identity);
            if (variant.has_gzip && variant.gzip != null) {
                emit_one_array(b, i2, i3, "GZIP", (!)variant.gzip);
            }
            if (variant.has_zstd && variant.zstd != null) {
                emit_one_array(b, i2, i3, "ZSTD", (!)variant.zstd);
            }
            if (variant.has_br && variant.br != null) {
                emit_one_array(b, i2, i3, "BR", (!)variant.br);
            }

            b.append_printf("%s}\n", i1);
            if (namespace_name != null) {
                b.append("}\n");
            }

            return b.str;
        }

        private void emit_one_array(StringBuilder b, string i2, string i3, string name, uint8[] data) {
            b.append_printf("%sprivate const uint8[] %s = {\n", i2, name);
            for (int i = 0; i < data.length; i++) {
                if (i == 0) {
                    b.append(i3);
                } else if (i % 16 == 0) {
                    b.append(",\n").append(i3);
                } else {
                    b.append(", ");
                }
                b.append_printf("0x%02x", data[i]);
            }
            b.append_printf("\n%s};\n\n", i2);
        }


        // ------------------------------------------------------------------
        // DOM helpers
        // ------------------------------------------------------------------

        private Xml.Doc* parse_html_file(string path) throws GLib.Error {
            var file = File.new_for_path(path);
            if (!file.query_exists()) {
                throw new IOError.NOT_FOUND(@"Page file '$path' does not exist");
            }

            uint8[] contents;
            string etag_out;
            file.load_contents(null, out contents, out etag_out);

            // Parse with libxml2's HTML parser (mirrors Astralis' MarkupDocument)
            // so the tree has HTML semantics and serializes as HTML, not XML.
            int opts = (int)(Html.ParserOption.RECOVER |
                Html.ParserOption.NOERROR |
                Html.ParserOption.NOWARNING |
                Html.ParserOption.NOBLANKS |
                Html.ParserOption.NONET);
            char[] buffer = ((string) contents).to_utf8();
            var doc = Html.Doc.read_memory(buffer, buffer.length, path, "utf-8", opts);
            if (doc == null) {
                throw new IOError.FAILED(@"Failed to parse HTML file '$path'");
            }
            return doc;
        }

        private static Xml.Node* find_body(Xml.Doc* doc) {
            var root = doc->get_root_element();
            if (root == null) {
                return null;
            }
            return find_first_element(root, "body");
        }

        private static Xml.Node* find_head(Xml.Doc* doc) {
            var root = doc->get_root_element();
            if (root == null) {
                return null;
            }
            return find_first_element(root, "head");
        }

        /**
         * Validates that `doc` has both a <head> and a <body> element, which every
         * Statum HTML input (pages, templates and fragments) must declare.
         */
        private void validate_document(Xml.Doc* doc, string source) throws GLib.Error {
            if (find_head(doc) == null) {
                throw new IOError.FAILED(@"\"$source\" has no <head> element (a <head> and <body> are required)");
            }
            if (find_body(doc) == null) {
                throw new IOError.FAILED(@"\"$source\" has no <body> element (a <head> and <body> are required)");
            }
        }

        /**
         * Validates that the structural directives <pstm-title>, <pstm-template>
         * and <pstm-name> appear only inside <head>.
         */
        private void validate_head_directives(Xml.Doc* doc, string source) throws GLib.Error {
            var root = doc->get_root_element();
            if (root == null) {
                return;
            }
            string[] names = { "pstm-title", "pstm-template", "pstm-name", "pstm-uri" };
            foreach (var name in names) {
                var nodes = new Series<Xml.Node*>();
                find_all_from(root, name, nodes);
                foreach (var node in nodes) {
                    if (!is_descendant_of_head(node)) {
                        throw new IOError.FAILED(@"\"$source\": <$name> must appear inside <head>");
                    }
                }
            }
        }

        /** Walks the ancestor chain to determine whether `node` is within <head>. */
        private static bool is_descendant_of_head(Xml.Node* node) {
            for (var p = node->parent; p != null; p = p->parent) {
                if (is_element(p) && ((string) p->name).down() == "head") {
                    return true;
                }
            }
            return false;
        }

        private static Xml.Node* find_first_element(Xml.Node* root, string name) {
            if (!is_element(root)) {
                return null;
            }
            if (((string) root->name).down() == name.down()) {
                return root;
            }
            for (var child = root->children; child != null; child = child->next) {
                var found = find_first_element(child, name);
                if (found != null) {
                    return found;
                }
            }
            return null;
        }

        private static Xml.Node*[] find_all_elements(Xml.Doc* doc, string name) {
            var collected = new Series<Xml.Node*>();
            var root = doc->get_root_element();
            if (root != null) {
                find_all_from(root, name, collected);
            }
            return collected.to_array();
        }

        private static void find_all_from(Xml.Node* node, string name, Series<Xml.Node*> result) {
            if (is_element(node) && ((string) node->name).down() == name.down()) {
                result.add(node);
            }
            if (is_element(node)) {
                for (var child = node->children; child != null; child = child->next) {
                    find_all_from(child, name, result);
                }
            }
        }

        private static string? read_directive_text(Xml.Node* body, string directive) {
            var node = find_first_element(body, directive);
            if (node == null) {
                return null;
            }
            var text = element_text(node);
            return text.strip().length > 0 ? text.strip() : null;
        }

        private static string? read_pstm_name(Xml.Doc* doc) {
            var node = find_first_element(doc->get_root_element(), "pstm-name");
            if (node == null) {
                return null;
            }
            return element_text(node).strip();
        }

        private static string[] read_routes(Xml.Doc* doc) {
            string[] routes = new string[0];
            var nodes = find_all_elements(doc, "pstm-uri");
            foreach (var node in nodes) {
                var text = element_text(node).strip();
                if (text.length > 0) {
                    routes += text;
                }
            }
            return routes;
        }

        private static string element_text(Xml.Node* node) {
            var b = new StringBuilder();
            for (var child = node->children; child != null; child = child->next) {
                if (child->type == ElementType.TEXT_NODE && child->content != null) {
                    b.append((string) child->content);
                } else if (is_element(child)) {
                    b.append(element_text(child));
                }
            }
            return b.str;
        }

        private static void apply_title(Xml.Doc* doc, string? title) {
            if (title == null || title.length == 0) {
                return;
            }
            var head = find_first_element(doc->get_root_element(), "head");
            if (head == null) {
                return;
            }
            var title_node = find_first_element(head, "title");
            if (title_node == null) {
                title_node = head->new_child(null, "title", null);
            }
            for (var child = title_node->children; child != null; child = child->next) {
                child->unlink();
            }
            title_node->add_content(title);
        }

        private static string serialize(Xml.Doc* doc) {
            // Serialize with libxml2's HTML serializer (htmlDocDumpMemory) so the
            // output is valid HTML: no <?xml?> prolog, no self-closing
            // `<script/>`, and correct void/non-void element handling. The doc was
            // produced by the HTML parser, so it carries HTML structure.
            string mem;
            int len;
            ((Html.Doc*) doc)->dump_memory(out mem, out len);
            return len > 0 ? mem.substring(0, len) : mem;
        }

        private static bool is_element(Xml.Node* node) {
            return node != null && node->type == ElementType.ELEMENT_NODE;
        }

        private static string compute_hash(uint8[] data) {
            var checksum = new Checksum(ChecksumType.SHA512);
            checksum.update(data, data.length);
            return checksum.get_string();
        }

        private static string escape(string s) {
            return s.replace("\\", "\\\\").replace("\"", "\\\"");
        }

        private static void write_file(string path, string contents) throws GLib.Error {
            var file = File.new_for_path(path);
            var stream = new DataOutputStream(file.replace(null, false, FileCreateFlags.NONE));
            stream.put_string(contents);
            stream.close();
        }

        private static string make_pascal_case(string input) {
            var basename = Path.get_basename(input);
            var dot = basename.last_index_of(".");
            if (dot > 0) {
                basename = basename.substring(0, dot);
            }
            var result = new StringBuilder();
            var capitalize_next = true;
            foreach (var c in basename.data) {
                if (c == '-' || c == '_' || c == ' ' || c == '.') {
                    capitalize_next = true;
                } else {
                    if (capitalize_next) {
                        result.append_printf("%c", c >= 'a' && c <= 'z' ? c - 32 : c);
                        capitalize_next = false;
                    } else {
                        result.append_printf("%c", c);
                    }
                }
            }
            return result.str;
        }

        // ------------------------------------------------------------------
        // Model
        // ------------------------------------------------------------------

        /** Scratch state for {@link collect_client_refs}. */
        private class ClientRefs : Object {
            public Xml.Node* legacy = null;
            public Xml.Node*[] modules = new Xml.Node*[0];
            public int duplicates = 0;
        }

        private class VariantData : Object {
            public uint8[] identity;
            public uint8[]? gzip;
            public uint8[]? zstd;
            public uint8[]? br;
            public bool has_gzip;
            public bool has_zstd;
            public bool has_br;
        }
    }
}