You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
XPath queries without an explicit namespace list (SelectNodes, SelectSingleNode, XPathEval) saw only the namespace of the context node itself. Every prefix declared on
the context node or on any of its ancestors is now registered, the nearest declaration
winning, so //p:item works from an unprefixed root that carries xmlns:p.
xmlNodeHelper.GetElementsByTagName never returned when called on an element with
children: the walk restarted from the first child on every step instead of advancing
from the current node. The DOM layer (IXMLElement.GetElementsByTagName) uses its own
list and was not affected.
Parsing single-byte encodings (windows-1251, ISO 8859, KOI8) through the static build
was about 20 % slower than the DLL: the iconv shim probed every input byte for a DBCS
lead byte. Single-byte code pages are recognised once per descriptor (GetCPInfo) and
converted in 1024-character chunks; the same document now parses about 9 % faster than
with the DLL.