aboutsummaryrefslogtreecommitdiff
path: root/nxcomp/ListFontsReply.cpp
diff options
context:
space:
mode:
authorMike Gabriel <mike.gabriel@das-netzwerkteam.de>2011-11-19 15:59:16 +0100
committerMike Gabriel <mike.gabriel@das-netzwerkteam.de>2011-11-19 15:59:16 +0100
commita48361b11a5abb5a345dac5ec83a8f56c4d50b74 (patch)
treea16fb870a072450bb45ee09ac4df9887bfa7f98e /nxcomp/ListFontsReply.cpp
parent9997e13bb583de4012914006c7507839a4e11227 (diff)
parent232dfc41d41390bfffa75ec2ed065c109fa03a0e (diff)
downloadnx-libs-a48361b11a5abb5a345dac5ec83a8f56c4d50b74.tar.gz
nx-libs-a48361b11a5abb5a345dac5ec83a8f56c4d50b74.tar.bz2
nx-libs-a48361b11a5abb5a345dac5ec83a8f56c4d50b74.zip
Merge branch 'nxcomp'
Diffstat (limited to 'nxcomp/ListFontsReply.cpp')
-rw-r--r--nxcomp/ListFontsReply.cpp204
1 files changed, 204 insertions, 0 deletions
diff --git a/nxcomp/ListFontsReply.cpp b/nxcomp/ListFontsReply.cpp
new file mode 100644
index 000000000..5bace82f2
--- /dev/null
+++ b/nxcomp/ListFontsReply.cpp
@@ -0,0 +1,204 @@
+/**************************************************************************/
+/* */
+/* Copyright (c) 2001, 2010 NoMachine, http://www.nomachine.com/. */
+/* */
+/* NXCOMP, NX protocol compression and NX extensions to this software */
+/* are copyright of NoMachine. Redistribution and use of the present */
+/* software is allowed according to terms specified in the file LICENSE */
+/* which comes in the source distribution. */
+/* */
+/* Check http://www.nomachine.com/licensing.html for applicability. */
+/* */
+/* NX and NoMachine are trademarks of Medialogic S.p.A. */
+/* */
+/* All rights reserved. */
+/* */
+/**************************************************************************/
+
+#include "ListFontsReply.h"
+
+#include "ServerCache.h"
+
+#include "EncodeBuffer.h"
+#include "DecodeBuffer.h"
+
+//
+// Set the verbosity level.
+//
+
+#define PANIC
+#define WARNING
+#undef DUMP
+#undef TEST
+#undef DEBUG
+
+ListFontsReplyStore::ListFontsReplyStore(StaticCompressor *compressor)
+
+ : MessageStore(compressor)
+{
+ enableCache = LISTFONTSREPLY_ENABLE_CACHE;
+ enableData = LISTFONTSREPLY_ENABLE_DATA;
+ enableSplit = LISTFONTSREPLY_ENABLE_SPLIT;
+ enableCompress = LISTFONTSREPLY_ENABLE_COMPRESS;
+
+ if (control -> isProtoStep7() == 1)
+ {
+ enableCompress = LISTFONTSREPLY_ENABLE_COMPRESS_IF_PROTO_STEP_7;
+ }
+
+ dataLimit = LISTFONTSREPLY_DATA_LIMIT;
+ dataOffset = LISTFONTSREPLY_DATA_OFFSET;
+
+ cacheSlots = LISTFONTSREPLY_CACHE_SLOTS;
+ cacheThreshold = LISTFONTSREPLY_CACHE_THRESHOLD;
+ cacheLowerThreshold = LISTFONTSREPLY_CACHE_LOWER_THRESHOLD;
+
+ messages_ -> resize(cacheSlots);
+
+ for (T_messages::iterator i = messages_ -> begin();
+ i < messages_ -> end(); i++)
+ {
+ *i = NULL;
+ }
+
+ temporary_ = NULL;
+}
+
+ListFontsReplyStore::~ListFontsReplyStore()
+{
+ for (T_messages::iterator i = messages_ -> begin();
+ i < messages_ -> end(); i++)
+ {
+ destroy(*i);
+ }
+
+ destroy(temporary_);
+}
+
+//
+// Here are the methods to handle messages' content.
+//
+
+int ListFontsReplyStore::parseIdentity(Message *message, const unsigned char *buffer,
+ unsigned int size, int bigEndian) const
+{
+ ListFontsReplyMessage *listFontsReply = (ListFontsReplyMessage *) message;
+
+ //
+ // Here is the fingerprint.
+ //
+
+ listFontsReply -> number_of_names = GetUINT(buffer + 8, bigEndian);
+
+ //
+ // Clean up padding bytes.
+ //
+
+ if ((int) size > dataOffset)
+ {
+ unsigned int current;
+ unsigned int length;
+ unsigned int nstringInNames;
+
+ unsigned char *end = NULL;
+ unsigned char *pad = NULL;
+
+ #ifdef DUMP
+
+ *logofs << "\n" << logofs_flush;
+
+ *logofs << "Number of STRING8 " << listFontsReply -> number_of_names << ".\n" << logofs_flush;
+
+ *logofs << "Size " << size << ".\n" << logofs_flush;
+
+ DumpHexData(buffer, size);
+
+ *logofs << "\n" << logofs_flush;
+
+ #endif
+
+ length = LISTFONTSREPLY_DATA_OFFSET;
+
+ for (nstringInNames = 0;
+ nstringInNames < listFontsReply -> number_of_names &&
+ listFontsReply -> number_of_names > 0;
+ nstringInNames++)
+ {
+ //
+ // Start with offset LISTFONTSREPLY_DATA_OFFSET 32.
+ //
+
+ current = buffer[length];
+
+ length += current + 1;
+
+ #ifdef DUMP
+ *logofs << "\nString number : " << nstringInNames << " Current length : "
+ << current << "\n" << logofs_flush;
+ #endif
+ }
+
+ #ifdef DUMP
+ *logofs << "\nFinal length " << length << "\n" << logofs_flush;
+ #endif
+
+ end = ((unsigned char *) buffer) + size;
+
+ for (pad = ((unsigned char *) buffer) + length; pad < end; pad++)
+ {
+ *pad = 0;
+
+ #ifdef DUMP
+ *logofs << "\nPadding ." << "\n" << logofs_flush;
+ #endif
+ }
+ }
+
+ #ifdef DEBUG
+ *logofs << name() << ": Parsed identity for message at " << message << ".\n" << logofs_flush;
+ #endif
+
+ return 1;
+}
+
+int ListFontsReplyStore::unparseIdentity(const Message *message, unsigned char *buffer,
+ unsigned int size, int bigEndian) const
+{
+ ListFontsReplyMessage *listFontsReply = (ListFontsReplyMessage *) message;
+
+ //
+ // Fill all the message's fields.
+ //
+
+ PutUINT(listFontsReply -> number_of_names, buffer + 8, bigEndian);
+
+ #ifdef DEBUG
+ *logofs << name() << ": Unparsed identity for message at "
+ << message << ".\n" << logofs_flush;
+ #endif
+
+ return 1;
+}
+
+void ListFontsReplyStore::dumpIdentity(const Message *message) const
+{
+ #ifdef DUMP
+
+ ListFontsReplyMessage *listFontsReply = (ListFontsReplyMessage *) message;
+
+ *logofs << name() << ": Identity number_of_names "
+ << listFontsReply -> number_of_names << ", size "
+ << listFontsReply -> size_ << ".\n";
+
+ #endif
+}
+
+void ListFontsReplyStore::identityChecksum(const Message *message, const unsigned char *buffer,
+ unsigned int size, int bigEndian) const
+{
+ //
+ // Field number_of_names.
+ //
+
+ md5_append(md5_state_, buffer + 8, 2);
+}