Hauke Duden provided an example program that made the multi
interface crash.  His example simply used the multi interface and
did first one FTP transfer and after completion it used a second
easy handle and did another FTP transfer on the same FTP server.

This triggered a bug in the "delayed easy handle kill" system
that curl uses: when an FTP connection is left alive it must keep
an easy handle around internally - only for the purpose of having
an easy handle when it later disconnects it. The code assumed
that when the easy handle was removed and an internal reference
was made, that version could be killed later on when a new easy
handle came using the same connection. This was wrong as Hauke's
example showed that the removed handle wasn't killed for real
until later. This caused a double close attempt => segfault.

which could have caused a double free when reusing curl handle.

  The problem mentioned on Dec 10 2009
  (http://curl.haxx.se/bug/view.cgi?id=2905220) was only partially fixed.
  Partially because an easy handle can be associated with many connections in
  the cache (e.g. if there is a redirect during the lifetime of the easy
  handle).  The previous patch only cleaned up the first one. The new fix now
  removes the easy handle from all connections, not just the first one.

  the easy interface was used.

Constantine Sapuntzakis detected and fixed a double free in builds done
with threaded resolver enabled (Windows default configuration) that would
get triggered when a curl handle is closed while doing DNS resolution.

  ran into some issues with the GSSAPI tests in configure.ac. The tests first
  try to determine the include dirs and libs and set CPPFLAGS and LIBS
  accordingly. It then checks for the headers and finally sets LIBS a second
  time, causing the libs to be included twice. The first setting of LIBS seems
  redundant and should be left out, since the first part is otherwise just
  about finding headers.

  My second issue is that 'krb5-config --libs gssapi' on Darwin is less than
  useless and returns junk that, while it happens to work with gcc, causes
  clang to choke. For example, --libs returns $CFLAGS along with the libs,
  which is really retarded. Simply setting 'LIBS="$LIBS -lgssapi_krb5
  -lresolv"' on Darwin is sufficient.

  makes sure that when using sub-second timeouts, there's no final bad 1000ms
  wait. Previously, a sub-second timeout would often make the elapsed time end
  up the time rounded up to the nearest second (e.g. 1s for 200ms timeout)

  CURLOPT_CERTINFO feature leaked memory due to a missing OpenSSL function
  call. He provided the patch to fix it too.

  http://curl.haxx.se/bug/view.cgi?id=2956698

  poll() function it doesn't quite work the way we want it so we must disable
  it, and he also provided a patch for it.

  http://curl.haxx.se/bug/view.cgi?id=2961796

  the global timeout if set. Also, as was reported in the bug report #2956437
  by Ryan Chan, the time stamp to use as basis for the per command timeout was
  not set properly in the DONE phase for FTP (and not for SMTP) so I fixed
  that just now. This was a regression compared to 7.19.7 due to the
  conversion of FTP code over to the generic pingpong concepts.

  http://curl.haxx.se/bug/view.cgi?id=2956437

  OpenSSL based SSL handshaking even though the multi interface was used and
  there was no good reason for it.

  http://curl.haxx.se/bug/view.cgi?id=2958179

  chunked-encoding trailer.

  http://curl.haxx.se/bug/view.cgi?id=2958474

Fixed bug report #2958074 indicating
(http://curl.haxx.se/bug/view.cgi?id=2958074) that curl on Windows with
option --trace-time did not use local time when timestamping trace lines.
This could also happen on other systems depending on time souurce.

  properly in angle brackets. Recipients provided with CURLOPT_MAIL_RCPT now
  get angle bracket wrapping automatically by libcurl unless the recipient
  starts with an angle bracket as then the app is assumed to deal with that
  properly on its own.

  full DATA has been sent, and I modified the test SMTP server to also send
  that response. As usual, the DONE operation that is made after a completed
  transfer is still not doable in a non-blocking way so this waiting for 250
  is unfortunately made blockingly.

  (http://curl.haxx.se/bug/view.cgi?id=2951319)

  in the same RCPT TO line, when they should be sent in separate single
  commands. I updated test case 802 to verify this.

- I also fixed a bad use of my_setopt_str() of CURLOPT_MAIL_RCPT in the curl
  tool which made it try to output it as string for the --libcurl feature
  which could lead to crashes.

mention last changes

  to automatically uncompress it with the CURLOPT_ENCODING option, libcurl
  could wrongly provide the callback with more data than what the maximum
  documented amount. An application could thus get tricked into badness if the
  maximum limit was trusted to be enforced by libcurl itself (as it is
  documented).

  This is further detailed and explained in the libcurl security advisory
  20100209 at

    http://curl.haxx.se/docs/adv_20100209.html

- Symbol CURL_FORMAT_OFF_T now obsoleted, will be removed in a future release,
  symbol will not be available when building with CURL_NO_OLDIES defined. Use
  of CURL_FORMAT_CURL_OFF_T is preferred since 7.19.0

simply check for CURLM_CALL_MULTI_PERFORM internally. This has the added
benefit that this goes in line with my long-term wishes to get rid of the
CURLM_CALL_MULTI_PERFORM all together from the public API.

mention asynchronous DNS lookups enhancements

  proxy that cannot be resolved when using c-ares. This matches the behaviour
  when not using c-ares.

libcurl options for controlling what to get and how to receive posssibly
interleaved RTP data. Initial commit.

  HTTP Cookie: header _needs_ to be sorted on the path length in the cases
  where two cookies using the same name are set more than once using
  (overlapping) paths. Realizing this, identically named cookies must be
  sorted correctly. But detecting only identically named cookies and take care
  of them individually is harder than just to blindly and unconditionally sort
  all cookies based on their path lengths. All major browsers also already do
  this, so this makes our behavior one step closer to them in the cookie area.

  Test case 8 was the only one that broke due to this change and I updated it
  accordingly.

  again when downloading files over FTP using ASCII and it turns out that the
  final size of the file is not the same as the initial size the server
  reported. This is very common since servers don't take the newline
  conversions into account.

  transfers: curl_multi_fdset() would return -1 and not set and file
  descriptors several times during a transfer of a single file. It turned out
  to be due to two different flaws now fixed. Gil's excellent recipe helped me
  nail this.

  ossl_connect_step3() increments an SSL session handle reference counter on
  each call. When sessions are re-used this reference counter may be
  incremented many times, but it will be decremented only once when done (by
  Curl_ossl_session_free()); and the internal OpenSSL data will not be freed
  if this reference count remains positive. When a session is re-used the
  reference counter should be corrected by explicitly calling
  SSL_SESSION_free() after each consecutive SSL_get1_session() to avoid
  introducing a memory leak.

  (http://curl.haxx.se/bug/view.cgi?id=2926284)

  name resolves when c-ares is used for resolving.