safely report IOErrors even in the face of unexpectedly encoded file names

--HG--
extra : rebase_source : e19f5f1b11ba54126f147b2a67110b2d9b66754d

2 files changed, 14 insertions, 2 deletions

diff --git a/src/lxml/apihelpers.pxi b/src/lxml/apihelpers.pxi
index 3ddcf6fb..50026cd8 100644
--- a/src/lxml/apihelpers.pxi
+++ b/src/lxml/apihelpers.pxi
@@ -1384,7 +1384,11 @@ cdef object _encodeFilename(object filename):
 cdef object _decodeFilename(const_xmlChar* c_path):
     u"""Make the filename a unicode string if we are in Py3.
     """
-    c_len = tree.xmlStrlen(c_path)
+    return _decodeFilenameWithLength(c_path, tree.xmlStrlen(c_path))
+
+cdef object _decodeFilenameWithLength(const_xmlChar* c_path, size_t c_len):
+    u"""Make the filename a unicode string if we are in Py3.
+    """
     if _isFilePath(c_path):
         try:
             return python.PyUnicode_Decode(
diff --git a/src/lxml/parser.pxi b/src/lxml/parser.pxi
index b606badc..c036284f 100644
--- a/src/lxml/parser.pxi
+++ b/src/lxml/parser.pxi
@@ -586,9 +586,17 @@ cdef int _raiseParseError(xmlparser.xmlParserCtxt* ctxt, filename,
                           _ErrorLog error_log) except 0:
     if filename is not None and \
            ctxt.lastError.domain == xmlerror.XML_FROM_IO:
+        if isinstance(filename, bytes):
+            filename = _decodeFilenameWithLength(
+                <bytes>filename, len(<bytes>filename))
         if ctxt.lastError.message is not NULL:
+            try:
+                message = (ctxt.lastError.message).decode('utf-8')
+            except UnicodeDecodeError:
+                # the filename may be in there => play safe
+                message = (ctxt.lastError.message).decode('iso8859-1')
             message = u"Error reading file '%s': %s" % (
-                filename, (ctxt.lastError.message).strip())
+                filename, message.strip())
         else:
             message = u"Error reading '%s'" % filename
         raise IOError, message