Fix receiving page, if it have not obj num.

core/fpdfapi/parser/cpdf_document.cpp

Index: core/fpdfapi/parser/cpdf_document.cpp
diff --git a/core/fpdfapi/parser/cpdf_document.cpp b/core/fpdfapi/parser/cpdf_document.cpp
index 1c73ce1015b168f633a2a4b5362a1229ee98d39a..b2a3d82e216165c3b3b2108bd485590205f3e60e 100644
--- a/core/fpdfapi/parser/cpdf_document.cpp
+++ b/core/fpdfapi/parser/cpdf_document.cpp
@@ -492,9 +492,20 @@ CPDF_Dictionary* CPDF_Document::GetPage(int iPage) {
   if (!pPages)
     return nullptr;
 
+  int nPagesToGo = iPage - m_iNextPageToTraverse + 1;
+  if (nPagesToGo <= 0) {

[npm, 2016/11/09 20:32:56]

This could also happen in cases where we just skip that page number.
See places in TraversePDFPages where m_pTreeTraversal[level].second++ but
m_PageList is not updated (no idea how common this is).

[snake, 2016/11/09 20:46:47]

I see this before. I will check this case. 

+    // This can be happened, when page have not objNum.

[Lei Zhang, 2016/11/09 20:19:41]

Rephrase: "This can happen when the page does not have an object number. On
repeated calls to this function for the same page index, this condition causes
TraversePDFPages() to incorrectly return nullptr."

[snake, 2016/11/09 20:30:41]

Done.

+    // And on call GetPage with page index, which was traversed before,
+    // nPagesToGo will be less or zero and TraversePDFPages returns nullptr;
+    // Example "testing\corpus\fx\other\jetman_std.pdf"

[Lei Zhang, 2016/11/09 20:19:41]

I would prefer to use "/" for path separator, since that's what we use for
include paths, and I'm biased against Windows. ;-)

[snake, 2016/11/09 20:30:41]

Done.

+    // We should restart traversing in this case.
+    // TODO(art-snake): optimize this.

[npm, 2016/11/09 20:32:56]

To optimize, we could store the dict pointer when objnum is 0.

[snake, 2016/11/09 20:46:47]

I should think about this. May be more better generate objNum for such pages.
I'm not sure now.

[npm, 2016/11/09 21:55:03]

AFAIK objNum is 0 for CPDF_Object when its direct reference.Maybe Tom has
some idea about the best way to obtain/store something in this case.

[snake, 2016/11/09 22:39:47]

Anyway, i will do this in other CL.

+    m_pTreeTraversal.clear();

[npm, 2016/11/09 20:32:56]

nit: use ResetTraversal() instead.

[snake, 2016/11/09 20:46:47]

Done.

+    m_iNextPageToTraverse = 0;
+    nPagesToGo = iPage - m_iNextPageToTraverse + 1;

[Lei Zhang, 2016/11/09 20:19:41]

nit: You already know |m_iNextPageToTraverse| is 0, so just omit it?

[snake, 2016/11/09 20:30:41]

Done.

+  }
   if (m_pTreeTraversal.empty())
     m_pTreeTraversal.push_back(std::make_pair(pPages, 0));
-  int nPagesToGo = iPage - m_iNextPageToTraverse + 1;
   CPDF_Dictionary* pPage = TraversePDFPages(iPage, &nPagesToGo, 0);
   m_iNextPageToTraverse = iPage + 1;
   return pPage;