Log Message:
Update to 1.21, from Quentin Garnier in PR 21918.
Changes:
  1.21 - 28 April 2003

      * Tests 148 & 150 from t/02zlib.t were failing on redhat 9.

      * Added a few words about the problems with Mac OS X to the README file.

  1.20 - 4 April 2003

      * Fixed bug in gzopen where $gzerrno wasn't being set correctly.
        The symptom was $gzerrno was set to Z_MEM_ERROR although the file
	was opened ok. If gzopen failed, $gzerrno was being set correctly.
	This problem wasn't spotted before because the typical test
	to determine whether gzopen passed or failed was to check its
	return value.

  1.19 - 31 October 2002

      * fixed a problem with t/02zlib.t that was failing with redhat 8.

  1.18 - 24 October 2002

      * fixed a Win32 problem in t/02zlib.t by changing sysread to read.

      * zlib 1.0.5 & older doesn't have gzsetparams & gzeof. Added a new
        variable to config.in to flag an old version of zlib. Split
        out the tests for gzsetparams into t/05gzsetp.t

  1.17 - 22 October 2002

      * Moved the test to check the versions of libz & zlib.h into a separate
        file and added troubleshooting notes to README.

      * In gzopen, only attempt to call "tell" for normal files.

      * Broke changes out of README into Changes file.

      * Replaced internal use of Z_PARTIAL_FLUSH symbol with Z_SYNC_FLUSH.
        zlib.h says  /* will be removed, use Z_SYNC_FLUSH instead */

      * Added support for zlib functions inflateSync and deflateParams.

      * Added support for zlib functions gzeof and gzsetparams.

      * Added support for access to adler, total_in & total_out

      * The compress function can now take an optional parameter that
        allows the compression level to be specified. This mirrors the
        compress2 function available in zlib.

      * memGzip doesn't work properly with perl 5.8.0 when it is given
        UTF-8 data. Bug spotted by Andreas J. Koenig.

      * Added note about Linux zlib-devel RPM to README.

      * Fixed recursive build problem on win32 machines.

      * Fixed problem with the test harness on Mac OS X.
        Thanks to Carl Johan Berglund for reporting the problem and
        helping track it down.