* known problems

* Limitations

** Limits about index

In one index, logical limits in Senna are below.

-max record count : 268,435,455
-max lexicon count : 268,435,455
-max index size : 256GByte

(Actually, other conditions may interfere for Senna to attain these limits.)

** About MySQL bindings

*** We support only MyISAM

InnoDB is not supported.

*** Japanese encoding method is EUC

If you want to handle UTF-8 without explicitly specify encoding on creating index,
make /var/senna/senna.conf,
and write below.
 DEFAULT_ENCODING utf8

*** About 2ind-patch

2ind-patch may not be able to handle some type of query.

*** MySQL with 2ind patch returns results which is not sorted by score

ex.) with FORCE INDEX, sorting order is different from what you want.
 SELECT 
 SQL_NO_CACHE cur_title FROM cur
 WHERE match(cur_text) against('united states') LIMIT 0, 100;

 SELECT 
 SQL_NO_CACHE cur_title FROM cur FORCE INDEX(PRIMARY)
 WHERE match(cur_text) against('united states') LIMIT 0, 100;

*** Replications, Boolean Mode are supported.

*** To get results faster.

MySQL's data reading is slow compared to search in Senna.
So, you can get results quick with limit like 'limit 0, 10'.

'Select count(*)' query which contains many results is slow without 2ind-patch.
Because MySQL without 2ind-patch cannot use two index at one time.
You can get results quick with 2ind-patch.

*** After 'optimize table', you have to rebuild the index of Senna.

After you execute 'optimize table',

you have to execute either one.
-drop index and create index
-stop MySQL daemon, and execute 'myisamchk -r' for the tables indexed by Senna.

*** After Senna's major version up, rebuild the index of Senna.
Senna's major version up may change the format of fulltext index.

you have to execute either one.
-drop index and create index
-stop MySQL daemon, and execute 'myisamchk -r' for the tables indexed by Senna.

** use in multi-process/multi-thread

You cannot 'update' one index at a time in multi-process/multi-thread environments.
If you do so, you have to update with exclusive access control.

** About SEN_SPLIT_ALPHA,DIGIT,SYMBOL

SEN_SPLIT_ALPHA,DIGIT,SYMBOL options have no effect in absense of SEN_INDEX_NORLALIZE.

** About sen_records_find, sen_recods_at

You cannot get valid results with sen_records_next 
after you use sen_records_find or sen_records_at.
You have to execute sen_records_rewind and initialize the cursor.

** About opentag and closetag in snippet API

If you use snippet functions,
you cannot modify data in a buffer which is passed as opentag or closetag 
before you call sen_snip_close.

** About sen_sel_partial option, sen_sym_suffix_search

In below case, 
you cannot execute 'suffix' search in searching with sen_sel_search or sen_sym_suffix_search.

-alphabet, number, symbol in euc-jp/sjis/utf-8 text
-index is created with SEN_INDEX_DELIMTED option.

Because, English words generally vary in it's suffix but prefix.

** Indexes in Windows and Linux are different.

** stddef.h is required
senna.h uses some C99 types like size_t,
you have to include 'stddef.h' before you include 'senna.h'.

* Bugs

** Now, you cannot use Senna with AIO/DIO.

Senna has a user level cache with AIO/DIO which is implemented in Linux 2.6.
But, this feature is not stable. Segmentation fault may occur.

This problem is reported by Yappo.

** In 64 bit environment, you cannot make Senna with AIO/DIO.

** In snippet API, some snippets overlap.

** Bugs in searching with sen_sel_near option

Words of which df is 0 are searched preferentially.
