We have an environment where we use both SQL*Server and Oracle.

Currently I have 1 set of mapping files, for oracle.

The SQL*Server mapping is largely the same, but, I need to change things like

<property name="Data" type="com.mine.hbm.OracleXmlUserType" not-null="false">

to be

<property name="Data" type="string" not-null="false">
<column name="DATA" sql-type="xml"/>
</property>

I can do this by having a parallel set of mapping files. Or I could generate the 2 files from a 'master file' at build time (e.g. ANT property replacement, XSLT, etc). Both feel a bit icky - is there any easier way to do this, perhaps property replacement in the hbm mapping file, conditional includes or something like that?

use <typedef> and change the typedef for each deployment.

That's ok for items where it's merely the type mapping that changes, but when I looked it didn't help when I actually need to change the underlying column.

For example, in Oracle I have

<id name="Id" type="com.cswgroup.kms.hbm.OracleOIDUserType">
<column name="ID" length="36"/>
</id>

in SQL*Server I need something like (

<id name="Id" type="com.cswgroup.kms.hbm.SQLServerOIDUserType">
<column name="ID" sql-type="uniqueidentifier"/>
</id>

I can't see how to return uniqueidentifier for a usertype, so I have to specify it, and if I don't say (in the former) length=36, then because it's mapped to types.VARCHAR, then the schema generator makes it something huge like VARCHAR(255) which isn't what I want...

well then some clever search-and-replace is probably the only way.

(or simply don't use the mapping files for schemaexport and just map it without the db specific type mappings ;)

Ok - that's what I thought, it's useful to have it confirmed :-)

What I might actually do then is customise the schema generator, since this is the only part (from what I can see) that is causing me a problem and I can do the rest with a typedef.

Ta.