Navigation to relative fragments does not work correctly for OOP pdf.

pdf/out_of_process_instance.cc

Index: pdf/out_of_process_instance.cc
diff --git a/pdf/out_of_process_instance.cc b/pdf/out_of_process_instance.cc
index 59b04394419410bc1eda3518839f2323169bbe94..fb1a67dc750492f77395d07cfd2a808ca0bb00cf 100644
--- a/pdf/out_of_process_instance.cc
+++ b/pdf/out_of_process_instance.cc
@@ -842,7 +842,23 @@ void OutOfProcessInstance::NavigateTo(const std::string& url,
     // If |url_copy| starts with '#', then it's for the same URL with a
     // different URL fragment.
     if (url_copy[0] == '#') {
-      url_copy = url_ + url_copy;
+      // if '#' is already present in |url_| then remove old tag and add new
+      // |url_copy| tag.
+      std::size_t found_idx = url_.find('#');
+      if (found_idx != std::string::npos)
+        url_copy = url_.substr(0, found_idx) + url_copy;
+      else
+        url_copy = url_ + url_copy;
+
+      if (!open_in_new_tab) {
+        found_idx = url_copy.find('#');

[Lei Zhang, 2015/01/07 23:08:43]

There's some weird problems with OOP PDF. Without this if-block, open a new tab,
go to http://examples.itextpdf.com/results/part1/chapter02/movie_links_1.pdf,
scroll to the last page, and click the #US link -> results in no navigation.
Clicking it a second time works.

The other issue is when it does work, we are not simply scrolling to the page
with the anchor. Instead, we are reloading the document. We shouldn't have to
reload. This may be out of the scope of this CL, but I want to mention it.

In-Process PDF doesn't have these problems, but it does suffer from the #US#US
issue.

+        if (found_idx != std::string::npos) {
+          int page_number =
+              engine_->GetNamedDestinationPage(url_copy.substr(found_idx + 1));
+          if (page_number >= 0)
+            ScrollToPage(page_number);
+        }
+      }
     }
     // If there's no scheme, add http.
     if (url_copy.find("://") == std::string::npos &&