js support import (2ac8ea1b) · Commits · jan.koester / blogi

editor/AI_HTML_TO_BLOGI_XML.md

+1 −1

Changes for editor/AI_HTML_TO_BLOGI_XML.md: 1 added line, 1 removed line.
Original line number	Diff line number	Diff line
		@@ -70,7 +70,7 @@ Regeln:
		1. Bei Unsicherheit lieber `CustomHtml` statt falsches Widget.
		2. Keine leeren Widgets ohne Inhalt erzeugen.
		3. ID/Klassen aus HTML nur uebernehmen, wenn sie fuer Funktion oder Styling relevant sind.
		4. Script-Tags nicht als aktive Scripts uebernehmen; stattdessen Struktur sichern.
		4. Script-Tags: externe `<script src="http(s)://...">` und Inline-Scripts (kein `src`, JS-Typ oder kein `type`) werden als aktives `GlobalJs` uebernommen und beim Rendern der Seite ausgefuehrt; `<script type="application/ld+json">`/`"text/template">`-artige Nicht-JS-Payloads werden ausgeschlossen. Da das importierte HTML-Element-`id`-Attribut nie in die Ausgabe uebernommen wird (jedes Widget bekommt stattdessen eine generierte `we-<uuid>`-Klasse), werden `getElementById("x")`/`querySelector("#x")`/`querySelectorAll("#x")`-Aufrufe im uebernommenen JS automatisch auf `.we-<uuid>` umgeschrieben, sofern `x` einer beim Import gesehenen id entspricht -- unbekannte ids bleiben unveraendert. Klassenbasierte Lookups (`getElementsByClassName`, `querySelector(".x")`) werden nicht umgeschrieben, da eine Quell-Klasse mehrere Elemente treffen kann, von denen jedes seine eigene uuid bekommt.

		## Prompt-Vorlage fuer eine KI

editor/src/htmlimport.cpp

+131 −15

Changes for editor/src/htmlimport.cpp: 131 added lines, 15 removed lines.
Original line number	Diff line number	Diff line
		@@ -31,6 +31,7 @@
		#include <algorithm>
		#include <cctype>
		#include <cstring>
		#include <regex>
		#include <sstream>
		#include <stack>

		@@ -665,6 +666,83 @@ std::map<std::string,std::string> resolveCustomProperties(
		return env;
		}

		// Sets @p w's instance_id, generating one first if @p newInstanceId (whatever
		// extractCustomCss came back with) is empty -- extractCustomCss only mints a
		// uuid when there's leftover CustomCss text to key it off of, but an element
		// with an original `id` needs a stable `we-<uuid>` class regardless, since
		// that id is otherwise dropped entirely and imported inline script would
		// have no other way to address the element (see rewriteJsIdReferences).
		// Records the id -> uuid mapping into @p idToInstanceId so that rewrite can
		// find it.
		void finalizeInstanceId(tinyxml2::XMLElement *w, const std::string &elemId,
		std::string &newInstanceId,
		std::map<std::string,std::string> &idToInstanceId)
		{
		if (newInstanceId.empty() && !elemId.empty()) {
		uuid::uuid id;
		id.generate();
		newInstanceId = id.c_str();
		}
		if (!newInstanceId.empty()) {
		w->SetAttribute("instance_id", newInstanceId.c_str());
		if (!elemId.empty()) idToInstanceId[elemId] = newInstanceId;
		}
		}

		// Runs every match of @p pattern (capturing the quote character as group 1,
		// via backreference, and the looked-up id as group 2) through
		// @p idToInstanceId, replacing the whole match with @p makeReplacement's
		// result when found and leaving it untouched otherwise.
		std::string rewriteMatchedCalls(const std::string &text, const std::regex &pattern,
		const std::map<std::string,std::string> &idToInstanceId,
		const std::function<std::string(const std::string&)> &makeReplacement)
		{
		std::string out;
		out.reserve(text.size());
		size_t lastEnd = 0;
		for (auto it = std::sregex_iterator(text.begin(), text.end(), pattern),
		end = std::sregex_iterator();
		it != end; ++it) {
		const std::smatch &m = *it;
		size_t pos = static_cast<size_t>(m.position(0));
		out.append(text, lastEnd, pos - lastEnd);
		auto found = idToInstanceId.find(m[2].str());
		out += (found == idToInstanceId.end()) ? m[0].str() : makeReplacement(found->second);
		lastEnd = pos + static_cast<size_t>(m.length(0));
		}
		out.append(text, lastEnd, text.size() - lastEnd);
		return out;
		}

		// Rewrites `.getElementById("x")` / `.querySelector("#x")` /
		// `.querySelectorAll("#x")` calls in imported script text so they still
		// resolve after import -- see convertHtmlToWidgetXml's doc comment
		// (htmlimport.h) for the full rationale. An id with no entry in
		// @p idToInstanceId (never seen during import, or the element that had it
		// didn't become its own widget) is left untouched -- best-effort, same
		// philosophy as this file's URL rewriting elsewhere.
		std::string rewriteJsIdReferences(const std::string &js,
		const std::map<std::string,std::string> &idToInstanceId)
		{
		if (js.empty() \|\| idToInstanceId.empty()) return js;

		// Group 1 is the opening quote, re-required as the closing quote via the
		// \1 backreference (so a `"..."`-quoted call can't be closed by a stray
		// `'`); group 2 is the id itself. rewriteMatchedCalls always reads
		// group 2.
		static const std::regex byId(R"(\.getElementById\(\s(['"])([^'"])\1\s*\))");
		static const std::regex qsAll(R"(\.querySelectorAll\(\s(['"])#([^'"])\1\s*\))");
		static const std::regex qs(R"(\.querySelector\(\s(['"])#([^'"])\1\s*\))");

		std::string out = rewriteMatchedCalls(js, byId, idToInstanceId,
		[](const std::string &uuid) { return ".querySelector(\".we-" + uuid + "\")"; });
		out = rewriteMatchedCalls(out, qsAll, idToInstanceId,
		[](const std::string &uuid) { return ".querySelectorAll(\".we-" + uuid + "\")"; });
		out = rewriteMatchedCalls(out, qs, idToInstanceId,
		[](const std::string &uuid) { return ".querySelector(\".we-" + uuid + "\")"; });
		return out;
		}

		} // namespace

		std::string blogi::htmlimport::extractCustomCss(
		@@ -849,6 +927,7 @@ void blogi::htmlimport::htmlElementToWidgetXml(
		tinyxml2::XMLElement *parentXml,
		const libhtmlpp::CSSStyleSheet &sheet,
		std::set<std::string> &seenMediaBlocks,
		std::map<std::string,std::string> &idToInstanceId,
		libhtmlpp::Element *stopBefore,
		const std::map<std::string,std::string> &rootCustomProperties)
		{
		@@ -990,7 +1069,7 @@ void blogi::htmlimport::htmlElementToWidgetXml(
		applyFlexLayoutAttributes(w, cssProps, knownCss);
		std::string newInstanceId;
		std::string custom = extractCustomCss(cssProps, knownCss, mediaRules, newInstanceId);
		if (!newInstanceId.empty()) w->SetAttribute("instance_id", newInstanceId.c_str());
		finalizeInstanceId(w, elemId, newInstanceId, idToInstanceId);
		if (!custom.empty()) {
		tinyxml2::XMLElement *ccEl = doc.NewElement("CustomCss");
		ccEl->SetText(custom.c_str());
		@@ -1035,7 +1114,7 @@ void blogi::htmlimport::htmlElementToWidgetXml(
		applyFlexLayoutAttributes(w, cssProps, knownCss);
		std::string newInstanceId;
		std::string custom = extractCustomCss(cssProps, knownCss, mediaRules, newInstanceId);
		if (!newInstanceId.empty()) w->SetAttribute("instance_id", newInstanceId.c_str());
		finalizeInstanceId(w, elemId, newInstanceId, idToInstanceId);
		if (!custom.empty()) {
		tinyxml2::XMLElement *ccEl = doc.NewElement("CustomCss");
		ccEl->SetText(custom.c_str());
		@@ -1090,7 +1169,7 @@ void blogi::htmlimport::htmlElementToWidgetXml(
		}
		std::string newInstanceId;
		std::string custom = extractCustomCss(cssProps, knownCss, mediaRules, newInstanceId);
		if (!newInstanceId.empty()) w->SetAttribute("instance_id", newInstanceId.c_str());
		finalizeInstanceId(w, elemId, newInstanceId, idToInstanceId);
		if (!custom.empty()) {
		tinyxml2::XMLElement *ccEl = doc.NewElement("CustomCss");
		ccEl->SetText(custom.c_str());
		@@ -1110,7 +1189,7 @@ void blogi::htmlimport::htmlElementToWidgetXml(
		applyFlexLayoutAttributes(w, cssProps, knownCss);
		std::string newInstanceId;
		std::string custom = extractCustomCss(cssProps, knownCss, mediaRules, newInstanceId);
		if (!newInstanceId.empty()) w->SetAttribute("instance_id", newInstanceId.c_str());
		finalizeInstanceId(w, elemId, newInstanceId, idToInstanceId);
		if (!custom.empty()) {
		tinyxml2::XMLElement *ccEl = doc.NewElement("CustomCss");
		ccEl->SetText(custom.c_str());
		@@ -1145,7 +1224,7 @@ void blogi::htmlimport::htmlElementToWidgetXml(
		"display","visibility"};
		std::string newInstanceId;
		std::string custom = extractCustomCss(cssProps, knownCss, mediaRules, newInstanceId);
		if (!newInstanceId.empty()) w->SetAttribute("instance_id", newInstanceId.c_str());
		finalizeInstanceId(w, elemId, newInstanceId, idToInstanceId);
		if (!custom.empty()) {
		tinyxml2::XMLElement *ccEl = doc.NewElement("CustomCss");
		ccEl->SetText(custom.c_str());
		@@ -1167,7 +1246,7 @@ void blogi::htmlimport::htmlElementToWidgetXml(
		std::vector<std::string> knownCss = {"width","height","display"};
		std::string newInstanceId;
		std::string custom = extractCustomCss(cssProps, knownCss, mediaRules, newInstanceId);
		if (!newInstanceId.empty()) w->SetAttribute("instance_id", newInstanceId.c_str());
		finalizeInstanceId(w, elemId, newInstanceId, idToInstanceId);
		if (!custom.empty()) {
		tinyxml2::XMLElement *ccEl = doc.NewElement("CustomCss");
		ccEl->SetText(custom.c_str());
		@@ -1312,7 +1391,7 @@ void blogi::htmlimport::htmlElementToWidgetXml(
		"display","visibility"};
		std::string newInstanceId;
		std::string custom = extractCustomCss(cssProps, knownCss, mediaRules, newInstanceId);
		if (!newInstanceId.empty()) w->SetAttribute("instance_id", newInstanceId.c_str());
		finalizeInstanceId(w, elemId, newInstanceId, idToInstanceId);
		std::string structureCss = collectTableStructureCss(sheet, usedClasses);
		if (!structureCss.empty()) {
		if (!custom.empty()) custom += " ";
		@@ -1364,7 +1443,7 @@ void blogi::htmlimport::htmlElementToWidgetXml(
		applyFlexLayoutAttributes(w, cssProps, knownCss);
		std::string newInstanceId;
		std::string custom = extractCustomCss(cssProps, knownCss, mediaRules, newInstanceId);
		if (!newInstanceId.empty()) w->SetAttribute("instance_id", newInstanceId.c_str());
		finalizeInstanceId(w, elemId, newInstanceId, idToInstanceId);
		if (!custom.empty()) {
		tinyxml2::XMLElement *ccEl = doc.NewElement("CustomCss");
		ccEl->SetText(custom.c_str());
		@@ -1423,7 +1502,7 @@ void blogi::htmlimport::htmlElementToWidgetXml(
		"visibility","text-decoration","cursor","text-align"};
		std::string newInstanceId;
		std::string custom = extractCustomCss(cssProps, knownCss, mediaRules, newInstanceId);
		if (!newInstanceId.empty()) w->SetAttribute("instance_id", newInstanceId.c_str());
		finalizeInstanceId(w, elemId, newInstanceId, idToInstanceId);
		if (!custom.empty()) {
		tinyxml2::XMLElement *ccEl = doc.NewElement("CustomCss");
		ccEl->SetText(custom.c_str());
		@@ -1484,7 +1563,7 @@ void blogi::htmlimport::htmlElementToWidgetXml(
		"padding","margin","line-height"};
		std::string newInstanceId;
		std::string custom = extractCustomCss(cssProps, knownCss, mediaRules, newInstanceId);
		if (!newInstanceId.empty()) w->SetAttribute("instance_id", newInstanceId.c_str());
		finalizeInstanceId(w, elemId, newInstanceId, idToInstanceId);
		if (!custom.empty()) {
		tinyxml2::XMLElement *ccEl = doc.NewElement("CustomCss");
		ccEl->SetText(custom.c_str());
		@@ -2190,6 +2269,36 @@ bool blogi::htmlimport::convertHtmlToWidgetXml(const std::string &html,
		if (outStats) outStats->jsFailed++;
		}
		}
		} else if (src.empty()) {
		// Inline script -- collect its own source text, unless
		// its type marks it as non-JS payload (JSON-LD, a
		// page-builder's <script type="text/template"> clone
		// source, etc.): executing those as JS would just throw,
		// and dumping raw JSON/HTML into globalJs serves nobody.
		std::string type = sel->getAtributte("type");
		std::transform(type.begin(), type.end(), type.begin(),
		[](unsigned char c) { return std::tolower(c); });
		bool isJs = type.empty() \|\| type == "text/javascript" \|\|
		type == "application/javascript" \|\|
		type == "module" \|\| type == "text/ecmascript" \|\|
		type == "application/ecmascript";
		if (isJs) {
		std::string text = sel->getScript();
		if (!text.empty()) {
		// Dropped whole rather than truncated when it
		// would blow the budget -- a truncated script is
		// far more likely to break every script
		// concatenated after it (unterminated string/
		// brace) than one left out cleanly.
		if (globalJs.size() + text.size() > kMaxTotalJsBytes) {
		if (outStats) outStats->jsFailed++;
		} else {
		if (!globalJs.empty()) globalJs += "\n;\n";
		globalJs += text;
		if (outStats) outStats->jsFetched++;
		}
		}
		}
		}
		cur = cur->nextElement();
		continue;
		@@ -2304,14 +2413,15 @@ bool blogi::htmlimport::convertHtmlToWidgetXml(const std::string &html,
		if (!globalCss.empty()) {
		xmlRoot->SetAttribute("GlobalCss", globalCss.c_str());
		}
		if (!globalJs.empty()) {
		xmlRoot->SetAttribute("GlobalJs", globalJs.c_str());
		}

		// Shared across every top-level node imported below (not per-node) so
		// an @media block already emitted via one node's widgets isn't repeated
		// by another's -- see collectApproximateMatches's dedup comment.
		std::set<std::string> seenMediaBlocks;
		// Populated by htmlElementToWidgetXml as elements with an original `id`
		// get converted; used below to rewrite id-based DOM lookups in globalJs
		// once every element's instance_id is known (see rewriteJsIdReferences).
		std::map<std::string,std::string> idToInstanceId;

		if (root) {
		if (selectedPaths) {
		@@ -2319,12 +2429,18 @@ bool blogi::htmlimport::convertHtmlToWidgetXml(const std::string &html,
		libhtmlpp::Element *el = resolveImportPath(root, path);
		if (el) {
		htmlElementToWidgetXml(el, xmlDoc, xmlRoot, stylesheet, seenMediaBlocks,
		el->nextElement(), rootEnv);
		idToInstanceId, el->nextElement(), rootEnv);
		}
		}
		} else {
		htmlElementToWidgetXml(root, xmlDoc, xmlRoot, stylesheet, seenMediaBlocks, nullptr, rootEnv);
		htmlElementToWidgetXml(root, xmlDoc, xmlRoot, stylesheet, seenMediaBlocks,
		idToInstanceId, nullptr, rootEnv);
		}
		}

		globalJs = rewriteJsIdReferences(globalJs, idToInstanceId);
		if (!globalJs.empty()) {
		xmlRoot->SetAttribute("GlobalJs", globalJs.c_str());
		}

		tinyxml2::XMLPrinter printer;

editor/src/htmlimport.h

+33 −6

Changes for editor/src/htmlimport.h: 33 added lines, 6 removed lines.
Original line number	Diff line number	Diff line
		@@ -135,13 +135,22 @@ namespace htmlimport {
		* cascade/inheritance) and threaded down to its children -- see
		* resolveCustomProperties in htmlimport.cpp. libhtmlpp::Element has no
		* parent pointer, so this is threaded top-down through the traversal
		* rather than looked up via ancestor walk. */
		* rather than looked up via ancestor walk.
		*
		* @p idToInstanceId records, for every converted element that had an
		* original `id` attribute, the `we-<uuid>` instance id blogi assigned it
		* (see finalizeInstanceId in htmlimport.cpp) -- the only stable way to
		* address that element in the rendered output, since the original id
		* itself is never re-emitted. convertHtmlToWidgetXml uses the
		* accumulated map afterwards to rewrite id-based DOM lookups in
		* collected inline script text (see rewriteJsIdReferences). */
		void htmlElementToWidgetXml(
		libhtmlpp::Element *htmlEl,
		tinyxml2::XMLDocument &doc,
		tinyxml2::XMLElement *parentXml,
		const libhtmlpp::CSSStyleSheet &sheet,
		std::set<std::string> &seenMediaBlocks,
		std::map<std::string,std::string> &idToInstanceId,
		libhtmlpp::Element *stopBefore = nullptr,
		const std::map<std::string,std::string> &rootCustomProperties = {});

		@@ -168,11 +177,29 @@ namespace htmlimport {

		/** Convert raw HTML string to a widget XML string (<WebEditor>...).
		* Returns true on success, writes the XML to @p outXml, optional
		* global CSS to @p outGlobalCss, and the concatenated contents of any
		* externally hosted <script src="http(s)://..."> tags (each fetched via
		* fetchExternalResource, same limits/safety as external stylesheets) to
		* @p outGlobalJs -- inline <script> blocks (no src) are left untouched,
		* matching existing behavior. If @p selectedPaths is non-null, only
		* global CSS to @p outGlobalCss, and page-level JS to @p outGlobalJs:
		* the concatenated contents of externally hosted <script src="http(s)://
		* ..."> tags (each fetched via fetchExternalResource, same limits/safety
		* as external stylesheets) plus every inline <script> block (no src)
		* whose type is a recognized JS type or unset -- <script
		* type="application/ld+json">/"text/template">-style non-JS payloads are
		* never included, since executing them would just throw. Both are
		* subject to the same combined size budget (kMaxTotalJsBytes); a script
		* that alone would exceed the remaining budget is dropped whole rather
		* than truncated (a truncated script is far more likely to break every
		* script concatenated after it than one dropped cleanly). Since every
		* imported element's original `id` attribute is discarded in favor of a
		* generated `we-<uuid>` instance id/class (see finalizeInstanceId), the
		* collected script text has `.getElementById("x")` / `.querySelector(
		* "#x")` / `.querySelectorAll("#x")` calls rewritten to address that
		* same element's `.we-<uuid>` class instead, wherever `x` matches an
		* id blogi actually saw (see rewriteJsIdReferences) -- an id with no
		* match (never seen, or the element didn't become its own widget) is
		* left untouched, best-effort like this file's URL rewriting elsewhere.
		* Class-based lookups (getElementsByClassName, `querySelector(".x")`)
		* are deliberately not rewritten: a source-page class can match many
		* elements, each of which becomes its own distinct uuid on import, so
		* there's no sound one-to-one substitution. If @p selectedPaths is non-null, only
		* the nodes (see ImportNode::path, at any depth) named in the set are
		* converted/imported, each as its own top-level element -- otherwise
		* the whole document is imported (today's behavior). Selecting both an

editor/src/webedit_api.h

+7 −4

Changes for editor/src/webedit_api.h: 7 added lines, 4 removed lines.
Original line number	Diff line number	Diff line
		@@ -125,10 +125,13 @@ namespace webedit {
		// <style> tag by renderTree().
		std::string globalCss;
		// Concatenated contents of externally hosted <script src="...">
		// tags collected from an imported HTML page -- round-tripped via
		// the <WebEditor> root's GlobalJs attribute and appended as a
		// <script> tag by renderTree(). Inline <script> blocks are not
		// included (see convertHtmlToWidgetXml's doc comment).
		// tags plus inline <script> blocks (JS-typed or untyped; JSON-LD/
		// template-typed payloads excluded) collected from an imported
		// HTML page, with id-based DOM lookups already rewritten to the
		// imported elements' generated we-<uuid> classes -- round-tripped
		// via the <WebEditor> root's GlobalJs attribute and appended as a
		// <script> tag by renderTree() (see convertHtmlToWidgetXml's doc
		// comment).
		std::string globalJs;
		// Name of the publish target (e.g. "staticpage") last used for a
		// successful publish -- round-tripped via the <WebEditor> root's