TODO: add buffering plans

TODO

@@ -1,6 +1,9 @@
 Things TODO
 ===========
 
+* Fix the numerous malloc+copy operations for sending data, see below for
+  details
+
 * make sure the windowing code adapts better to slow situations so that it
   doesn't then use as much memory as today. Possibly by an app-controllable
   "Window mode"?
@@ -11,9 +14,11 @@ Things TODO
 
 * Add more info to the man pages.
 
-* Decrease the number of mallocs. Everywhere.
+* Decrease the number of mallocs. Everywhere. Will get easier once the
+  buffering improvements have been done.
 
-* Use SO_NOSIGPIPE for Mac OS/BSD systems where MSG_NOSIGNAL doesn't exist/work
+* Use SO_NOSIGPIPE for Mac OS/BSD systems where MSG_NOSIGNAL doesn't
+  exist/work
 
 * Extend the test suite to actually test lots of aspects of libssh2
 
@@ -21,6 +26,9 @@ Things TODO
 
 * Expose error messages sent by the server
 
+* Make SFTP transfers ask for and send several packages at once so that it
+  doesn't have to send-waitforack-send-waitforack as much.
+
 At next SONAME bump
 ===================
 
@@ -59,3 +67,48 @@ At next SONAME bump
 
 * remove the existing libssh2_knownhost_check() functin and rename
   libssh2_knownhost_checkp() to become the new libssh2_knownhost_check instead
+
+Buffering Improvements
+======================
+
+compression
+
+  - needs to ne able to take the input to compress from two pointers (with two
+  sizes) and generate a single compressed output chunk
+
+  - should also not allocate the compressed output buffer but use a single one
+  within the session struct (as transport_write() will copy the data off from
+  that buffer immediately anyway), or even better the transport_write()
+  function could allocate a buffer to fix an uncompressed buffer as then the
+  compression function can write its output directly into the allocated send
+  buffer
+
+  - should probably be split off into a separate compression and decompression
+  function instead of a single unified as they are now, to make these changes
+  easier
+
+transport_write
+
+  - should accept a "packet struct" input with separate pointers to headers
+  and to payload (each of which may be allocated or not)
+
+  - Separating the headers from the payload will greatly enhance our ability
+  to use a fixed buffer within the session handle for the header part and
+  provide the payload part as a mere pointer.
+
+  - This function is used >30 times in the code.
+
+  - If this function gets called with a total packet size that is larger than
+  32K, it should create more than one SSH packet so that it keeps the largest
+  one below 32K
+
+channel_write
+
+  - should not copy/allocate anything for the data, only create a header chunk
+  and pass on the payload data to transport_write "pointed to"
+
+sftp_write
+
+  - should not copy/allocate anything for the data, only create a header chunk
+  and pass on the payload data to channel_write "pointed to"
+