Hello Wouter!
I think we should split files and contents at least for performance reason. A database might cache a table without blob better and it is possible to put the content into another tablespace (oracle).
What about using inodes?
CREATE TABLE Files ( parent_inode numeric(22) not null, inode numeric(22) not null, name VARCHAR(200) NOT NULL, type numeric(1) not null, link_inode numeric(22) not null, lastModified TIMESTAMP NOT NULL, created TIMESTAMP NOT NULL, lastAccessed TIMESTAMP NOT NULL, );
PRIMARY KEY parent_inode, inode, ALTERNATE unique key parent_inode, name type: 0=file, 1=directory, 2=link The root inode is "0" per definition.
CREATE TABLE Contents (
inode numeric(22) not null,
content LONGBLOB NULL, );
PRIMARY KEY inode
CREATE TABLE MetaAttributes ( inode numeric(22) not null, id numeric(22) not null, key varchar(200), value blob );
PRIMARY KEY inode, id
I am not sure if the value should really be blob but what if one would store a serialized java object as attribute (filehandler ;-)
To answer Rami's mail, I think the url should be something like: dbfs://user:[EMAIL PROTECTED]:dbname/
So one filesystem per database instance. I think the schema could be taken from the user and if this is not sufficient we could add a filesystemconfig where one could explicitely set the schema.
And yes. All the database stuff should be abstracted and put into its own (replaceable) class. Thus we could try to provide specialised versions for different databases - or provide one using hibernate3 and get rid of the blob/clob handling too. This is something I could try to do.
Ciao, Mario
--------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
