Implement document fragment parsing.

Introduce a notable behavior change in default parsing mode: documents without a
document element node are now considered invalid. This is technically a breaking change,
however the amount of documents it affects is very small, all parsed data still persists,
and lack of this check results in very confusing behavior in a number of cases.

In order to be able to parse documents without an element node, a fragment parsing flag is
introduced.

Parsing a buffer in fragment mode treats the buffer as a fragment of a valid XML.
As a consequence, top-level PCDATA is added to the tree; additionally, there are no
restrictions on the number of nodes -- so documents without a document element are considered
valid.

Due to the way parsing works internally, load_buffer_inplace occasionally can not preserve
the document contents if it's parsed in a fragment mode. While unfortunate, this problem is
fundamental; since the use case is relatively obscure, hopefully documenting this shortcoming
will be enough.

git-svn-id: https://pugixml.googlecode.com/svn/trunk@980 99668b35-9821-0410-8761-19e4c4f06640

1 files changed, 1 insertions, 1 deletions

diff --git a/tests/test_memory.cpp b/tests/test_memory.cpp
index a37b91e..32d395b 100644
--- a/tests/test_memory.cpp
+++ b/tests/test_memory.cpp
@@ -119,7 +119,7 @@ TEST(memory_large_allocations)
 		CHECK(allocate_count == deallocate_count + 1); // only one live page left (it waits for new allocations)
 
 		char buffer;
-		CHECK(doc.load_buffer_inplace(&buffer, 0, parse_default, get_native_encoding()));
+		CHECK(doc.load_buffer_inplace(&buffer, 0, parse_fragment, get_native_encoding()));
 
 		CHECK(allocate_count == deallocate_count); // no live pages left
 	}