From wessels@COLORADO.EDU
Date: Thu, 02 Nov 1995 01:26:11 -0700
From: Duane Wessels <wessels@COLORADO.EDU>
To: harvest-users-real@burton.cs.colorado.edu
Subject: Harvest beta available


The Harvest version 1.3 patchlevel 3 beta release has been updated.
Both source and binary distributions are available at
    ftp://harvest.cs.colorado.edu/priv/pl3.beta/

Changes between release v1.3 pl3 (November, 1995) and v1.3:

- Changes to the Gatherer:
	- Added symbolic link loop detection to httpenum.
	- Added a GIF image summarizer (GIFImage.sum), requires netpbm.
	  The GIFImage type is still in the Essence stoplist by default.
	- Added 'C' version of ftpget.
	- Added ability to rewrite the SOIF template URL with Essence 
	  post-processing.  Could be used to gather file:// URLs and
	  have them exported as http:// URLs.
	- Fixed select() timeouts to POSIX semantics.
	- Fixed SGML summarizer to give error if input is empty.
	- Fixed a Makefile to actually build and install HTML-lax.sum.
	- Fixed liburl problem with AFS.  Must *copy* files into the
	  cache-liburl directory.
	- Fixed News gatherering: If 'newsget.pl' exits non-zero,
	  close the NNTP server socket.
	- Fixed newsget.pl with a major rewrite.
	- Fixed 'fileenum' to use URLs and not always return file://hostname/.
	- Fixed gatherd bug where child process would remove parent's
	  gatherd.pid file.
	- Changed NewsArticle.sum TTL to 7 days by default.
	- Changed Essence unnesting to occur in individual directories.
	- Removed confusing gatherd DNS mismatch warning message.

- Changes to the Broker:
	- Added #Restart-Index-Server command to broker admin command set.
	- Added error logging and debugging in Glimpse inline query code.
	- Fixed select() timeouts to POSIX semantics.
	- Fixed Glimpse minor malloc problems.
	- Fixed the broker on Linux; needs unbuffered input from gather
	  process.
	- Fixed broker query language bug for high-bit (international)
	  characters.
	- Changed Broker to allow specifically setting GlimpseServer_Port
	  again; if not set, port is chosen randomly.
	- Changed BrokerAdmin.cgi to use unbuffered output.
	- Changed Glimpse macros CLEANUP and RETURN to be functions.
	- Changed broker admin/LOG to log FQDN instead of IP address.
	- Remove glimpse version ambiguities in Glimpse/index.c.
	- Removed getpeername() call in the broker; get address from accept().

- Changes to the Cache:
	- The cache has been moved to a separate distribution.

- Miscellaneous Changes
	- Dont link with -lmalloc on Solaris.
	- Fixed User Manual and FAQ inconsistencies.




Changes between release v1.3 pl3 beta (26 Oct 1995) and v1.3 pl2:

- Changes to the Cache:

	- Add behind_firewall mode to cached.conf.
	- Add TCP_DONE logging.
	- Add C version of ftpget.
	- Add rotate_logs(), called upon SIGHUP.  Renames log files
	  with incremental digits 0 through 9.
	- Add code to disable UDP port if set to -1.
	- Add swap file bitmap to avoid stat(2) calls.
	- Add check for objects in DELETE_BEHIND mode; junk the object
	  when client goes away.
	- Add {FTP,HTTP,GOPHER}_EXPIRE logging.
	- Add User-Agent fix to include "via Harvest Cache..."
	- Add domain list to neighbor query algorithm--only ping neighbors
	  for URLs of the given domains.
	- Add checks for NULL StoreEntry key.
	- Add 'cache_hot_vm_factor' to cached.conf.
	- Add dead parent/neighbor detection.
	- Add single parent mode.  No ping or DNS lookups are done.
  	  Everything is retrieved from the parent.
	- Add cache_neighbor_obj on/off to cached.conf. 
	  If off, objects pulled from neighbors are not cached.
	- Add Linux patches from gitelson@chaph.usc.edu
	- Add configure check for IRIX, add -ansi if found.
	- Fix APPEND_DOMAIN coredumps.
	- Fix many things wrong with cachemgr/stats information.
	- Fix asciiProcessInput() to wait for end of HTTP request;
	  don't assume entire request comes in one read().
	- Fix having a cached parent/neighbor on the same machine.
	- Fix bindAddressList memory leak.
	- Fix client-goes-away coredump bug.
	- Fix object size limit problems.  Now can proxy any size object.
	- Fix bogus URL coredump bug (http://aaaaaaa.aaaaaaa.aaaaa....)
	- Fix local_ip to check server IP instead of client IP.
	- Fix gopher off-by-one bug.
	- Fix SECHO bug.  Added DECHO for dumb cache pinging.
	- Fix cached.conf inconsistencies and missing tags.
	- Fix getFromSource() coredump bug.
	- Fix cachemgr.c to not use /proxy/.
	- Change select() timeouts to POSIX style.
	- Changed NO_TESTNAME #define to a command line option, -D.
	- Changed stats to show objects in transit.
	- Changed select loop to check for FD timeouts less frequently.
	- Changed cached to stop accepting new connections when 75%
	  of available file descriptors have been used.
	- Changed store.c to continue swapfileno where
	  storeRebuildFromDisk() left off.
	- Changed HTTP and Gopher to not use lifetime handlers for
	  server FDs.  Use read timeouts instead.
	- Changed store.c to use multiple swap directories.
	- Changed stat.c to use storeAppend().
	- Changed strncmp()'s to strcmp() in cache_cf.c main loop.
	- Change DO_MALLINFO to work better.
	- Change configure to not link with -lmalloc on Solaris
	- Remove sys_errlist[] in each file.  Use xstrerror() instead.

Known Bugs:
	- cached loops if reading from the server-side is
	  more than *_DELETE_GAP bytes ahead of writing to the client.
	- "Server-Push" URLs not detected.


Duane W.
--
Duane Wessels <wessels@colorado.edu>
Harvest Developer      http://harvest.cs.colorado.edu/~wessels/