More dirty hacks to deal with Parsoid's behavior re newlines in <p>s

This strips certain newlines added by Parsoid so they don't end up in
the linear model, and puts them back in on the way out so Parsoid
doesn't freak out and produce invalid wikitext

Change-Id: I256aaded4229c915868dc868ec6eaa1a73e00be1

modules/ve2/dm/ve.dm.Converter.js

@@ -303,6 +303,42 @@ ve.dm.Converter.prototype.getDataFromDom = function( domElement, annotations, da
 				if ( text === '' ) {
 					break;
 				}
+				// HACK: strip implied leading and trailing newlines in <p> tags
+				// Workaround for a Parsoid bug
+				/*
+				 * Leading newlines:
+				 * If the previous sibling is a paragraph, do not strip leading newlines
+				 * If there is no previous sibling, do not strip leading newlines
+				 * Otherwise, strip 1 leading newline
+				 *
+				 * Trailing newlines:
+				 * If the next sibling is a paragraph, strip 2 trailing newlines
+				 * If there is no next sibling, do not strip trailing newlines
+				 * Otherwise, strip 1 trailing newline
+				 */
+				var contentNode = childDomElement.parentNode;
+				if ( contentNode.nodeName.toLowerCase() === 'p' ) {
+					//debugger;
+					if (
+						contentNode.previousSibling &&
+						contentNode.previousSibling.nodeName.toLowerCase() !== 'p' &&
+						text.charAt( 0 ) === '\n'
+					) {
+						text = text.substr( 1 );
+					}
+					if ( contentNode.nextSibling ) {
+						// Strip one trailing newline
+						if ( text.charAt( text.length - 1 ) === "\n" ) {
+							text = text.substr( 0, text.length - 1 );
+						}
+						if ( contentNode.nextSibling.nodeName.toLowerCase() === 'p' ) {
+							// Strip another one
+							if ( text.charAt( text.length - 1 ) === "\n" ) {
+								text = text.substr( 0, text.length - 1 );
+							}
+						}
+					}
+				}
 
 				// Start auto-wrapping of bare content
 				if ( !wrapping && !alreadyWrapped && !branchIsContent ) {
@@ -339,6 +375,40 @@ ve.dm.Converter.prototype.getDataFromDom = function( domElement, annotations, da
  * @returns {HTMLElement} Wrapper div containing the resulting HTML
  */
 ve.dm.Converter.prototype.getDomFromData = function( data ) {
+	function fixupText( text, node ) {
+		// HACK reintroduce newlines needed to make Parsoid not freak out
+		// This reverses the newline stripping done in getDataFromDom()
+		/*
+			* Leading newlines:
+			* If the previous sibling is a paragraph, do not add any leading newlines
+			* If there is no previous sibling, do not add any leading newlines
+			* Otherwise, add 1 leading newline
+			*
+			* Trailing newlines:
+			* If the next sibling is a paragraph, add 2 trailing newlines
+			* If there is no next sibling, do not add any trailing newlines
+			* Otherwise, add 1 trailing newline
+			*/
+		if ( node.nodeName.toLowerCase() === 'p' ) {
+			//debugger;
+			if (
+				node.previousSibling &&
+				node.previousSibling.nodeName.toLowerCase() !== 'p'
+			) {
+				text = "\n" + text;
+			}
+			if ( node.nextSibling ) {
+				// Add one trailing newline
+				text += "\n";
+				if ( node.nextSibling.nodeName.toLowerCase() === 'p' ) {
+					// Add another one
+					text += "\n";
+				}
+			}
+		}
+		return text;
+	}
+
 	var container = document.createElement( 'div' ),
 		domElement = container,
 		text;
@@ -503,6 +573,18 @@ ve.dm.Converter.prototype.getDomFromData = function( data ) {
 			}
 		});
 	});
+
+	// HACK more postprocessing, this time to add newlines to paragraphs so Parsoid doesn't freak out
+	$( container )
+		// Get all text nodes
+		.find( '*' )
+		.contents()
+		.filter( function() {
+			return this.nodeType == 3;
+		} )
+		.each( function() {
+			this.data = fixupText( this.data, this.parentNode );
+		} );
 	return container;
 };