Preserve indent inside unclosed inline HTML attributes

lib/blocks.js

@@ -70,6 +70,83 @@ var peek = function(ln, pos) {
     }
 };
 
+// True when `s` ends inside an unclosed HTML tag, comment, PI,
+// declaration, or CDATA section. Used so paragraph-indent stripping
+// does not eat significant whitespace inside a raw HTML attribute.
+var isInsideUnclosedHtmlTag = function(s) {
+    var i = 0;
+    var n = s.length;
+    while (i < n) {
+        var lt = s.indexOf("<", i);
+        if (lt < 0) {
+            return false;
+        }
+        if (s.slice(lt, lt + 4) === "<!--") {
+            var endComment = s.indexOf("-->", lt + 4);
+            if (endComment < 0) {
+                return true;
+            }
+            i = endComment + 3;
+            continue;
+        }
+        if (s.slice(lt, lt + 2) === "<?") {
+            var endPi = s.indexOf("?>", lt + 2);
+            if (endPi < 0) {
+                return true;
+            }
+            i = endPi + 2;
+            continue;
+        }
+        if (s.slice(lt, lt + 9) === "<![CDATA[") {
+            var endCdata = s.indexOf("]]>", lt + 9);
+            if (endCdata < 0) {
+                return true;
+            }
+            i = endCdata + 3;
+            continue;
+        }
+        if (s.slice(lt, lt + 2) === "<!" && /[A-Za-z]/.test(s.charAt(lt + 2))) {
+            var endDecl = s.indexOf(">", lt + 2);
+            if (endDecl < 0) {
+                return true;
+            }
+            i = endDecl + 1;
+            continue;
+        }
+        var namePos = lt + 1;
+        if (s.charAt(namePos) === "/") {
+            namePos += 1;
+        }
+        if (!/[A-Za-z]/.test(s.charAt(namePos))) {
+            i = lt + 1;
+            continue;
+        }
+        var j = namePos + 1;
+        while (j < n && /[A-Za-z0-9-]/.test(s.charAt(j))) {
+            j += 1;
+        }
+        var quote = null;
+        while (j < n) {
+            var c = s.charAt(j);
+            if (quote !== null) {
+                if (c === quote) {
+                    quote = null;
+                }
+            } else if (c === '"' || c === "'") {
+                quote = c;
+            } else if (c === ">") {
+                break;
+            }
+            j += 1;
+        }
+        if (j >= n) {
+            return true;
+        }
+        i = j + 1;
+    }
+    return false;
+};
+
 // DOC PARSER
 
 // These are methods of a Parser object, defined below.
@@ -83,13 +160,25 @@ var endsWithBlankLine = function(block) {
 // Add a line to the block at the tip.  We assume the tip
 // can accept lines -- that check should be done before calling this.
 var addLine = function() {
-    if (this.partiallyConsumedTab) {
-        this.offset += 1; // skip over tab
+    var offset = this.offset;
+    var partiallyConsumedTab = this.partiallyConsumedTab;
+    // Keep indent that was skipped before sending a paragraph to the
+    // inline parser when that indent sits inside unclosed raw HTML.
+    // Otherwise `<a class="foo\n    bar">` loses the attribute spaces.
+    if (
+        this.tip.type === "paragraph" &&
+        isInsideUnclosedHtmlTag(this.tip._string_content || "")
+    ) {
+        offset = this.lineContentOffset;
+        partiallyConsumedTab = false;
+    }
+    if (partiallyConsumedTab) {
+        offset += 1; // skip over tab
         // add space characters:
         var charsToTab = 4 - (this.column % 4);
         this.tip._string_content += " ".repeat(charsToTab);
     }
-    this.tip._string_content += this.currentLine.slice(this.offset) + "\n";
+    this.tip._string_content += this.currentLine.slice(offset) + "\n";
 };
 
 // Add block of type tag as a child of the tip.  If the tip can't
@@ -812,6 +901,8 @@ var incorporateLine = function(ln) {
 
     this.allClosed = container === this.oldtip;
     this.lastMatchedContainer = container;
+    // Remaining line text, including indent skipped by advanceNextNonspace.
+    this.lineContentOffset = this.offset;
 
     var matchedLeaf =
         container.type !== "paragraph" && blocks[container.type].acceptsLines;
@@ -988,6 +1079,7 @@ function Parser(options) {
         lineNumber: 0,
         offset: 0,
         column: 0,
+        lineContentOffset: 0,
         nextNonspace: 0,
         nextNonspaceColumn: 0,
         indent: 0,

test/regression.txt

@@ -625,6 +625,21 @@ A dotless i reference does not resolve against an ASCII I definition
 .
 <p>[ı]</p>
 ````````````````````````````````
+Inline HTML attribute whitespace is not stripped as paragraph indent (#303).
+
+```````````````````````````````` example
+<div class="abc
+    d  e f">text</dev>
+
+<a class="abc
+    d  e f">text</a>
+.
+<div class="abc
+    d  e f">text</dev>
+<p><a class="abc
+    d  e f">text</a></p>
+````````````````````````````````
+
 Entities inside autolinks (#263).
 
 ```````````````````````````````` example