test (7fe4cba7) · Commits · jan.koester / blogi

editor/src/htmlimport.cpp

+40 −5

Changes for editor/src/htmlimport.cpp: 40 added lines, 5 removed lines.
Original line number	Diff line number	Diff line
		@@ -260,7 +260,8 @@ void blogi::htmlimport::mergeStylesheetRules(
		const std::string &cssClass,
		const std::string &id,
		std::map<std::string,std::string> &props,
		std::string &mediaRules)
		std::string &mediaRules,
		std::set<std::string> &seenMediaBlocks)
		{
		std::vector<std::string> classes;
		if (!cssClass.empty()) {
		@@ -313,19 +314,38 @@ void blogi::htmlimport::mergeStylesheetRules(
		singleSel = singleSel.substr(start, end - start + 1);

		std::string matchSel = singleSel;
		bool hadCombinator = false;
		{
		size_t lastSep = matchSel.find_last_of(" >+~");
		if (lastSep != std::string::npos) {
		size_t mstart = matchSel.find_first_not_of(" >+~", lastSep);
		if (mstart != std::string::npos) {
		matchSel = matchSel.substr(mstart);
		hadCombinator = true;
		}
		}
		}

		bool matches = false;

		if (!matchSel.empty() && matchSel[0] == '#') {
		// A bare tag name left over after stripping a combinator's
		// ancestor part (e.g. "[data-atom=header]>div" reduces to "div")
		// can't be verified here -- this function only ever sees the
		// current element's own tag/class/id, never its ancestor chain
		// -- so treating it as a match would apply the rule to literally
		// every element of that tag in the whole document. Confirmed
		// against a real scraped page where exactly this pattern matched
		// ~24,000 times and blew up the imported output to 36MB. A
		// class/id qualifier on the rightmost compound (e.g. "div.foo")
		// is still specific enough to check below.
		bool bareTagAfterCombinator = hadCombinator &&
		!matchSel.empty() && matchSel[0] != '#' && matchSel[0] != '.' &&
		matchSel.find('.') == std::string::npos &&
		matchSel.find('#') == std::string::npos;

		if (bareTagAfterCombinator) {
		// matches stays false
		} else if (!matchSel.empty() && matchSel[0] == '#') {
		if (!id.empty() && matchSel == "#" + id) matches = true;
		} else if (!matchSel.empty() && matchSel[0] == '.') {
		std::string clsSel = matchSel.substr(1);
		@@ -368,7 +388,16 @@ void blogi::htmlimport::mergeStylesheetRules(
		block += prop.getName() + ": " + prop.getValue() + "; ";
		}
		block += "} } ";
		// The same @media block, once present anywhere in the
		// final page's HTML, applies document-wide regardless of
		// which widget's <style> tag carries it -- so including
		// it more than once across the whole import is pure
		// bloat, not a correctness requirement. Real scraped
		// pages routinely have many widgets share one rule (e.g.
		// by class), each independently re-matching it here.
		if (seenMediaBlocks.insert(block).second) {
		mediaRules += block;
		}
		} else {
		for (const auto &prop : rule.getDeclaration().getProperties()) {
		if (props.find(prop.getName()) == props.end()) {
		@@ -573,6 +602,7 @@ void blogi::htmlimport::htmlElementToWidgetXml(
		tinyxml2::XMLDocument &doc,
		tinyxml2::XMLElement *parentXml,
		const libhtmlpp::CSSStyleSheet &sheet,
		std::set<std::string> &seenMediaBlocks,
		libhtmlpp::Element *stopBefore)
		{
		if (!htmlEl) return;
		@@ -650,7 +680,7 @@ void blogi::htmlimport::htmlElementToWidgetXml(
		std::string elemId = el->getAtributte("id");

		std::string mediaRules;
		mergeStylesheetRules(sheet, tag, cssClass, elemId, cssProps, mediaRules);
		mergeStylesheetRules(sheet, tag, cssClass, elemId, cssProps, mediaRules, seenMediaBlocks);

		// ---- Map HTML tags to widget types ----

		@@ -1824,16 +1854,21 @@ bool blogi::htmlimport::convertHtmlToWidgetXml(const std::string &html,
		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 mergeStylesheetRules's dedup comment.
		std::set<std::string> seenMediaBlocks;

		if (root) {
		if (selectedPaths) {
		for (const auto &path : *selectedPaths) {
		libhtmlpp::Element *el = resolveImportPath(root, path);
		if (el) {
		htmlElementToWidgetXml(el, xmlDoc, xmlRoot, stylesheet, el->nextElement());
		htmlElementToWidgetXml(el, xmlDoc, xmlRoot, stylesheet, seenMediaBlocks, el->nextElement());
		}
		}
		} else {
		htmlElementToWidgetXml(root, xmlDoc, xmlRoot, stylesheet);
		htmlElementToWidgetXml(root, xmlDoc, xmlRoot, stylesheet, seenMediaBlocks);
		}
		}

editor/src/htmlimport.h

+13 −1

Changes for editor/src/htmlimport.h: 13 added lines, 1 removed line.
Original line number	Diff line number	Diff line
		@@ -74,13 +74,24 @@ namespace htmlimport {
		std::map<std::string,std::string> declarationToMap(
		const libhtmlpp::CSSDeclaration &decl);

		/** @p seenMediaBlocks accumulates the exact text of every @media block
		* already appended to some element's @p mediaRules across the whole
		* import (see convertHtmlToWidgetXml/htmlElementToWidgetXml, which
		* share one instance across every element processed) -- a block whose
		* text is already present is skipped rather than appended again. This
		* is safe because a <style> block's effect is document-wide regardless
		* of which widget's XML happens to carry it, and it matters because a
		* rule matched by a real page can independently match thousands of
		* elements (e.g. by shared class), each otherwise re-appending the
		* identical block text. */
		void mergeStylesheetRules(
		const libhtmlpp::CSSStyleSheet &sheet,
		const std::string &tag,
		const std::string &cssClass,
		const std::string &id,
		std::map<std::string,std::string> &props,
		std::string &mediaRules);
		std::string &mediaRules,
		std::set<std::string> &seenMediaBlocks);

		std::string extractCustomCss(
		std::map<std::string,std::string> &props,
		@@ -116,6 +127,7 @@ namespace htmlimport {
		tinyxml2::XMLDocument &doc,
		tinyxml2::XMLElement *parentXml,
		const libhtmlpp::CSSStyleSheet &sheet,
		std::set<std::string> &seenMediaBlocks,
		libhtmlpp::Element *stopBefore = nullptr);

		/** One node of the HTML-import block picker's tree: an element reachable