Use of UTF8 and AL32UTF8 for database character set

Question

I will be implementing Unicode on a 10g database, and am considering using AL32UTF8 as the database character set, as opposed to AL16UTF16 as the national character set, primarily to economize storage requirements for primarily English-based string data.
Is anyone aware of any issues, or tradeoffs, for implementing AL32UTF8 as the database character set, as opposed to using the national character set for storing Unicode data? I am aware of the fact that UTF-8 may require 3 bytes where UTF-16 would only require 2, so my question is more specific to the use of the database character set vs. the national character set, as opposed to differences between the encoding itself. (I realize that I could use UTF8 as the national character set, but don't want to lose the ability to store supplementary characters, which UTF8 does not support, as this Oracle character set supports up to Unicode 3.0 only.)
Thanks in advance for any counsel.

Answer 1

I don't have a lot of experience with SQL Server, but my belief is that a fair number of tools that handle SQL Server NCHAR/ NVARCHAR2 columns do not handle Oracle NCHAR/ NVARCHAR2 columns. I'm not sure if that's because of differences in the provided drivers, because of architectural differences, or because I don't have enough data points on the SQL Server side.
I've not run into any barriers, no. The two most common speedbumps I've seen are
- I generally prefer in Unicode databases to set NLS_LENGTH_SEMANTICS to CHAR so that a VARCHAR2(100) holds 100 characters rather than 100 bytes (the default). You could also declare the fields as VARCHAR2(100 CHAR), but I'm generally lazy.
- Making sure that the client NLS_LANG properly identifies the character set of the data going in to the database (and the character set of the data that the client wants to come out) so that Oracle's character set conversion libraries will work. If this is set incorrectly, all manner of grief can befall you. If your client NLS_LANG matches your database character set, for example, Oracle doesn't do a character set conversion, so if you have an application that is passing in Windows-1252 data, Oracle will store it using the same binary representation. If another application thinks that data is really UTF-8, the character set conversion will fail, causing it to display garbage, and then you get to go through the database to figure out which rows in which tables are affected and do a major cleanup. If you have multiple character sets inadvertently stored in the database (i.e. a few rows of Windows-1252, a few of Shift-JIS, and a few of UTF8), you'll have a gigantic mess to clean up. This is a concern whether you're using CHAR/ VARCHAR2 or NCHAR/ NVARCHAR2, and it's actually slightly harder with the N data types, but it's something to be very aware of.
Justin