Add tests of existing broken parsePatch behaviour, to fix before next…

… release (#525)

test/patch/parse.js

@@ -425,5 +425,103 @@ Index: test2
         }
       ]);
     });
+
+    it('should tolerate patches with extra trailing newlines after hunks', () => {
+      // Regression test for https://github.com/kpdecker/jsdiff/issues/524
+      // Not only are these considered valid by GNU patch, but jsdiff's own formatPatch method
+      // emits patches like this, which jsdiff used to then be unable to parse!
+      const patchStr = `--- foo 2024-06-14 22:16:31.444276792 +0100
++++ bar 2024-06-14 22:17:14.910611219 +0100
+@@ -1,7 +1,7 @@
+ first
+ second
+ third
+-fourth
+-fifth
++vierte
++fünfte
+ sixth
+ seventh
+
+`;
+      expect(parsePatch(patchStr)).to.eql([{
+        oldFileName: 'foo',
+        oldHeader: '2024-06-14 22:16:31.444276792 +0100',
+        newFileName: 'bar',
+        newHeader: '2024-06-14 22:17:14.910611219 +0100',
+        hunks: [
+          {
+            oldStart: 1,
+            oldLines: 7,
+            newStart: 1,
+            newLines: 7,
+            lines: [
+              ' first',
+              ' second',
+              ' third',
+              '-fourth',
+              '-fifth',
+              '+vierte',
+              '+fünfte',
+              ' sixth',
+              ' seventh'
+            ]
+          }
+        ]
+      }]);
+    });
+
+    it("shouldn't be caught out by removal/addition of lines starting with -- or ++", () => {
+      // The patch below is a valid patch generated by diffing this file, foo:
+
+      // first
+      // second
+      // third
+      // -- bla
+      // fifth
+      // sixth
+
+      // against this file, bar:
+
+      // first
+      // second
+      // third
+      // ++ bla
+      // fifth
+      // sixth
+      // seventh
+
+      // with the command `diff -u0 foo bar`. (All lines in `foo` and `bar` have no leading
+      // whitespace and a trailing LF.)
+
+      // This is effectively an adversarial example meant to catch out a parser that tries to
+      // detect the end of a file in a multi-file diff by looking for lines starting with '---',
+      // '+++', and then '@@'. jsdiff used to do this. However, as this example illustrates, it is
+      // unsound, since the '---' and '+++' lines might actually just represent the deletion and
+      // insertion of lines starting with '--' and '++'. The only way to disambiguate these
+      // interpretations is to heed the line counts in the @@ hunk headers; you *cannot* reliably
+      // determine where a hunk or file ends in a unified diff patch without heeding those line
+      // counts.
+
+      const patchStr = `--- foo 2024-06-14 21:57:04.341065736 +0100
++++ bar 2024-06-14 22:00:57.988080321 +0100
+@@ -4 +4 @@
+--- bla
++++ bla
+@@ -6,0 +7 @@
++seventh
+`;
+
+      expect(parsePatch(patchStr)).to.eql([{
+        oldFileName: 'foo 2024-06-14 21:57:04.341065736 +0100',
+        oldHeader: '',
+        newFileName: 'bar 2024-06-14 22:00:57.988080321 +0100',
+        newHeader: '',
+        hunks: [
+          { oldStart: 4, oldLines: 1, newStart: 4, newLines: 1, lines: ['--- bla', '+++ bla'] },
+          { oldStart: 7, oldLines: 0, newStart: 7, newLines: 1, lines: ['+seventh'] }
+        ]
+      }]);
+    });
   });
 });