* Netdb: Speed up collection of all the floodfill peers by using
           PeerManager.getPeersByCapability()
  * PeerManager: Capability cleanups

    - Add reseed hosts thx matt .at. drollette.com
    - Remove https://a.netdb.i2p2.de/, bad cert for months
    - Remove https://c.netdb.i2p2.de/, no data for months

    - Cancel our JVM shutdown hook when shutting down
    - Run a spinner task so shutdown always completes
    - exit() instead of halt() so other JVM shutdown hooks run
    - Prevent duplicate wrapper notifier hooks
    - Notify the wrapper twice, once for stopping and once for stopped

    - Fix delayed lookup of reason from file
    - Tag strings for translation
    - Sort IPs on configpeer.jsp correctly
    - Cleanups

  * OCMOSJ: Clear caches at restart
  * UPnP:
    - Fix device rediscovery and port opening after restart

          TunnelBuildReplyMessage into a common base class

    - Use system SHA-256 MessageDigest instead of Sha256Standalone in PRNG
    - Deprecate DataHelper functions using Sha256Standalone arguments;
      used only by Syndie
    - Note deprecation in javadocs

    - Log error on out-of-order spans
    - Log error on corrupt level and discard
    - Fix SkipSpan non-flush bug causing null levels
    - Add level fixer that repairs nulls and out-of-order levels
    - Fix bug on backwards span construction, not hit in practice
    - Lots of log level changes

  * Shutdown:
    - Clear more resources in peer manager, netdb, stat manager,
      session key manager, naming service, tunnel dispatcher,
      OCMOSJ (result of testing with jvisualvm)
    - Don't call wrapper on shutdown (starting two threads) if we
      were started with runplain

    - Fix Polish po file
    - Install as a service by default on Windows again
    - Change CPUID getters to package private
    - Split new jbigi install messages into two lines
    - Javadocs

(either wrong architecture or i2psvc is just plain missing) and give
a more useful error message.

    jbigi and wrapper files for arm.
    Compiled on trimslice with gcc version 4.4.5 (Ubuntu/Linaro 4.4.4-14ubuntu5) 
    Log postinstall errors to postinstall.log.
    java version "1.6.0_18"
    OpenJDK Runtime Environment (IcedTea6 1.8.7) (6b18-1.8.7-0ubuntu2.1)
    OpenJDK Zero VM (build 14.0-b16, mixed mode)
    GMP 4.3.2 (half the size of 5.0.2, and no speed difference) LGPLv3
    Wrapper 3.5.9 GPLv2
    All binaries stripped.

    As on the Android emulator, the stock BigInteger.modPow()
    has some serious bug on arm.
    Without the libjbigi checked in here, the JVM crashes almost immediately.

  you don't have gettext and want the build to continue.

Fixed single quotes.
Email follows:

...


I attached file "snark-messages_pl.po.gz". I give you permission to use this file in I2PSnark distribution.
If you don't believe me that it is true polish translation, you can check it with some web translator.

Fist I used ordinary text editor to translate. Later I used Poedit <http://www.poedit.net/>.
And I don't know why it gives strange error messages when I save the file.

17:43:29: snark-messages_pl.po:172: a format specification for argument {0} doesn't exist in 'msgstr'
17:43:29: snark-messages_pl.po:177: a format specification for argument {0} doesn't exist in 'msgstr'
17:43:29: msgfmt: found 2 fatal errors

_______
greetz,
polacco

Tested in Windows 7.
As has been customary, these were linked with gmp-5.0.2.

Built using 
"x86_64-w64-mingw32-gcc (GCC) 4.5.3" on Debian Sid with debug symbols stripped
using x86_64-w64-mingw32-strip.

The Tanuki Software website states "64-bit Windows versions of the Java Service Wrapper
are not currently being made available in the Community Edition." The Makefile
for x86_64 is missing from the upstream tarball as well. 

Well...included in this checkin is a diff against 
$WRAPPER-3.5.9-SRC/src/c/Makefile-windows-x86-32.nmake (see the README in
installer/libs/wrapper/win64.

bump version to -19 to reflect the upgraded wrapper and jcpuid/jbigi binaries for windows and freebsd

    * Update wrapper license

      - Add a lock for reading/writing the router.info file
      - Check our RouterInfo validity after reading and before saving,
        to catch fatal errors sooner

- Disable reusing Deflater on Android

      - Use java.security.MessageDigest instead of bundled GNU SHA-256 code
        if available, which it should always be.
        5 to 20% faster on Oracle JVM; 40 to 60% on Harmony;
        5 to 15% on JamVM; 20x (!) on GIJ.
      - Use java.security.MessageDigest instead of bundled Bitzi SHA-1 code
        if available on non-Oracle JVMs, which it should always be.
        Not faster on Oracle JVM; 30 to 60% faster on Harmony;
        15 to 20% on JamVM; 10-15x (!) on GIJ.