	v.1.2.6

In this version we have fixed numerous small and not-so-small bugs, as well
as added some improvements and new features. Here some changes are described
in details; for full list of changes see the NEWS file.

To summarize, this version should be more stable, as we have found the
reasons and fixed a few coredump cases, as well as healed some memory
leaks. Also, some tricks were incorporated to boost performance: buddy heap
is implemented for storing word position vectors in "index", replacing the
standard STL vector class; CBufferedFile is implemented and is used instead
of stdio during inverted index merging process (also known as "Saving
delta files").

Some other items from NEWS is explained here.

* Added MultipleDBConnections parameter to searchd.conf

Multiple DB connections feature is implemented in searchd. This improves
concurrency between searchd threads, especially in the case when some threads
do pattern search (which is slow), and other thread wants a simple search.

* Exiting of resolver process is logged now
* Event of breaking of read pipe in resolver process is logged now
* SIGCHLD signals are caught by "index"

We have some reports from users of 1.2.5 with complains about stuck of
index process. The problem seems to be in the resolver process. So, we have
added several facilities that can be helpful to nail down the bug.

* Added -X and -H flags to "index"
Use "index -X1" to check inverted index for URLs for which "urlword.deleted"
field is non-zero. Use "index -X2" to fix it by appending information about
deleted keys to the delta files. So if you want to remove records where
"urlword.deleted" is non-zero, run index -X2; index -D, and finally perform
SQL statements to delete unnecessary records.

"index -H" is used to recreate citation indexes and ranks file from
"urlwordsNN.hrefs" fields in case of citation index corruption.

* Fixed -P and -A flags in "index"

"index -P" is used to show "path" to given URL.

"index -A" is used to add or delete site to/from web space.
