How to convert Oracle charset database

Question

I have a Oracle database which set charset "WE8ISO8859P1". how can I convert to "SJIS" charset that dont need create new database

Answer 1

You can't.
You can successfully convert an Oracle database's charset without database recreation if and only if 1) the original charset is a strict subset of the final charset (the letters in the set) and 2) the codepoints (the numbers that encode the letters) in the original charset don't change in the final charset and 3) all of the data in the database actually is encoded in the original charset (or at least is correctly encoded in the final charset). This is because this type of conversion is accomplished by simply slapping a new charest lable on the database as a whole; none of the binary values in the data get looked at, much less touched. If you "convert" a database and all three are not true, then some or all of your string data becomes complete and utter garbage.
For WE8ISO8859P1 to SJIS, both 1) and 2) are false.
3) becomes an issue because of what Oracle calls an optimization and I call a stupidity; when a database recieves string data from a client that asserts it's using the same charset, nothing checks that the data passed is actually in that charset; thus is it's trivially easy to get 8-bit data in a 7-bit US7ASCII database; it will read and write correctly until the day you try to convert to a superset of US7ASCII, such as UTF8. Then WHAM! Data corruption.
Oracle customers with support contracts can get a lot of documentation about character set conversions from the Metalink support site. Oracle also strongly recommends that customers file a support request for conversions (although someone who reads all the docs and understands character encoding and the NLS settings can muddle through on their own). Oracle also has lot of free documentation about globalization and character sets on the OTN site, which just requires a registration, but I'm not sure the nitty gritty details are there.
For non-subset conversions, the way to do it is to export the data, create a new database and import the data. Conversion to the new encoding will occur on import.
There is a utility, CSSCAN, that can be used to prepare for a conversion. It will scan the database for strings that will have difficulty converting.
Besides the possible issue of data that is outside of the original charset, it will also check for data that will be truncated on conversion. By default, the size specification of a VARCHAR2 in Oracle is the number of bytes; when a character is converted, it may expand the number of bytes that it requires, for example when switching form a 7 or 8 bit encoding to a 16 bit encoding. If enough characters in a value are expanded, the data exceeds the size of the column and is truncated or dropped (I forget). Recent Oracle versions lets you change from having the DB count size in bytes to counting size in characters; I don't yet know if this plays nice with imports with character set conversion (I'll know in a few months).