Hi,
I agree with what was said before that these annotations should also allow creation of other DB elements such as indexes and table spaces. I also like where the syntax is evolving to.
But I wonder how we would override one of multiple create statements in a deployment descriptor if I find that I want/need a slightly different statement from what’s in a hard-coded annotation.
E.g.:
@DDLs{
@DDL(suffix="engine=InnoDB", platforms="MySQL")
@DDL(create="CREATE TABLE blah ... tablespace X, initial size 512m", override=true, platforms="Oracle")
@DDL(create="CREATE INDEX", override=true, platforms="Oracle")
@DDL(drop="DROP TABLE blah CASCADE CONSTRAINTS", type=DROP, platforms={"Oracle"})
}
How would I put a <ddl> definition for a different CREATE INDEX for Oracle in my deployment descriptor (without affecting the CREATE TABLE)?
Basically, I suppose we’d need a unique identifier in there somewhere to match the two.
Or an all or nothing approach, where a single <ddl create=> overrides all @DDL(create=)’s so when you need to override one you’d have to copy the other ones to your deployment descriptor too.
But will programmers remember to fill in a unique name? Can it be defaulted automatically to something unique enough?
@DDL(create="CREATE INDEX", name="create_index1", override=true, platforms="Oracle")
I hope someone can come up with something nicer and better J
Regards,
Dies
From: eclipselink-dev-bounces@xxxxxxxxxxx [mailto:eclipselink-dev-bounces@xxxxxxxxxxx] On Behalf Of Goerler, Adrian
Sent: Wednesday, 6 April 2011 18:57
To: Dev mailing list for Eclipse Persistence Services
Subject: Re: [eclipselink-dev] bug 340329 - table creation prefix
Hi,
I like this proposal a lot because it promises ultimate flexibility. But I don’t think the @DDL annotation should have a member “ddl”. I’d prefer “drop” and “create”.
How about:
@DDLs{
@DDL(suffix="engine=InnoDB", platforms="MySQL")
@DDL(create="CREATE TABLE blah ... tablespace X, initial size 512m", override=true, platforms="Oracle")
@DDL(drop="DROP TABLE blah CASCADE CONSTRAINTS", type=DROP, platforms={"Oracle"})
}
Also, I am afraid that the @DDL annotations will get quite lengthy. Maybe one could alternatively refer to a file, e.g.
@DDL(createScript=”ddl-scripts/person_create_oracle.sql”, platforms=”Oracle”)
-Adrian
From: eclipselink-dev-bounces@xxxxxxxxxxx [mailto:eclipselink-dev-bounces@xxxxxxxxxxx] On Behalf Of James Sutherland
Sent: Dienstag, 5. April 2011 21:08
To: Dev mailing list for Eclipse Persistence Services
Subject: Re: [eclipselink-dev] bug 340329 - table creation prefix
DDL would allow more than just table creation, (indexes, constraints, types, tablespaces, etc.)
I think having multiple different nested annotations is too complex. Just have a single DDL and follow the normal collection pattern with an array annotation.
override is require as it may be overriding the table DDL, or may be DDL for a constraint or index on the existing table.
A type could be added to define when the DDL should be executed.
@DDLs{
@DDL(suffix="engine=InnoDB", platforms="MySQL")
@DDL(ddl="CREATE TABLE blah ... tablespace X, initial size 512m", override=true, platforms="Oracle")
@DDL(ddl="DROP TABLE blah CASCADE CONSTRAINTS", type=DROP, platforms={"Oracle"})
}
SQL = structured query language (querying, does not include ddl)
DDL = data definition language (creating database types/objects)
-----Original Message-----
From: douglas clarke
Sent: Tuesday, April 05, 2011 1:57 PM
To: eclipselink-dev@xxxxxxxxxxx
Subject: Re: [eclipselink-dev] bug 340329 - table creation prefix
James,
Interesting. Some feedback on this proposal:
I like the idea of being able to specify the full create table string similar to how @Column allows you to specify a replacement columnDefinition.
I don't love the idea of an annotation being an acronym. I would prefer to see an annotation that is more readable as to what it will do.
If you can provide a create string or suffix for one or more database platforms then you should be able to specify multiples
I prefer the platform names used in TARGET_DATABASE ("eclipselink.target-database") property and believe we should take strings and process them in a similar fashion using the string as a class name if a short name match is not found
Unsure what exactly you mean by override. If the full sql is provided it is used as is. If a suffix is provided it is used in conjunction with the generated sql. If no value is provided for the current platform then the default generated sql is used.
Should we include specifying the DROP TABLE syntax for specific platforms as well so I could get the CASCADE CONSTRAINTS included on platforms that support it.
As a rough idea we could support something like:
@Entity
@Table(name="BLAH")
@DDL(
create={
@CreatTable(suffix="engine=InnoDB" platforms="MySQL"),
@CreateTable(sql="CREATE TABLE blah ... tablespace X, initial size 512m" platforms="Oracle")
}
drop=@DropTable(sql="DROP TABLE blah CASCADE CONSTRAINTS", platforms={"Oracle"})
)
public class Blah {
...
On 04/04/2011 8:39 AM, James Sutherland wrote:
I don't think we should be using descriptor properties for this (or any) feature.
If we have a specific feature, we should have a specific annotation for it.
I would propose,
@DDL(ddl, table, suffix, override, databases)
.i.e
@DDL(suffix="engine=InnoDB", databases=MySQLPlatform.class)
@DDL(ddl="create hash index empname_inx on employee by name")
@DDL(ddl="create table employee (id numeric(20), name varchar2(512)) tablespace amce3, initial size 512m, overrude=true, databases=OraclePlatform.class)
I don't think we should add infix or prefix unless we have valid examples of these. I do not know of any, other than "temporary" which is not relevant.
Please give concrete examples for this feature.
-----Original Message-----
From: douglas clarke
Sent: Friday, April 01, 2011 2:43 PM
To: eclipselink-dev@xxxxxxxxxxx
Subject: Re: [eclipselink-dev] bug 340329 - table creation prefix
The original issue that lead to the suffix capability was very specific to the engine=InnoDB usage.
I believe what we are now discussing is a slightly broader set of requirements around augmenting the CREATE TABLE DDL statement.
I would say the requirements are now:
Allow a developer to specify a suffix for the CREATE DATABASE at the PU level or specifically on a given entity's table(s)
Allow a developer to specify a prefix for the CREATE DATABASE at the PU level or specifically on a given entity's table(s)
Allow the developer to specify the platform a suffix or prefix applies to and only use it in that case
One approach would be to deprecate the <table suffix support and extend the existing properties support to apply to entities and to allow the property to be specified as platform specific.
PU property (already exists):
"eclipselink.ddl-generation.table-creation-suffix" - indicates a suffix used on all CREATE TABLES for all platforms
Add platform support:
"eclipselink.ddl-generation.table-creation-suffix.MySQL" - indicates a suffix for all CREATE TABLES issued against a database platform with the short name 'MySQL'.
Then we could extend this support on a per-entity basis using EclipseLink's properties for cases where a developer needs to supply these values differently for different entity table(s).
@Entity
@Properties({
@Property(name="eclipselink.ddl-generation.table-creation-suffix.MySQL", value="engine=InnoDB"),
@Property(name="eclipselink.ddl-generation.table-creation-prefix.MaxDB", value="???")
})
public class Foo {
This of course could also be specified in the eclispelink-orm.xml within an entities's properties tags.
Doug
On 29/03/2011 8:29 AM, Tom Ware wrote:
That's a reasonable point. In the end, I guess it comes down to a choice about which is more important to us, the flexibility or the portability.
My vote is for the flexibility due to the fact that we currently only know of one suffix and it is not even necessary if you set your MySQL to use Innodb. DDL Generation, after all, is only a development-time feature.
I wonder if James' suggesting of a larger DDL config could be used to address this problem more generally. I guess it depends on if we want to worry about that larger feature now, or if it is more important to get the prefix work.
Other committers... What do you think?
-Tom
Goerler, Adrian wrote:
Hi,
We thought about the delegation model you suggest below. The weakness is, of course, that it requires a code change to deal with any property that might be passed in. (i.e. if MySQL added a new modifier other than INNODB, we'd have to actually change the MySQLPlatform to support it). So far, the features that are addressed with the suffix are fairly obscure, so we thought it would be better to leave it free form. I am, however open to arguments the other way.
an advantage of the delegation model would be that properties (or hints) not applicable for the current database platform would just be ignored. With the free from suffix, I can't run the same application on two different database platforms.
-Adrian
Adrian Görler
SAP AG
Pflichtangaben/Mandatory Disclosure Statements:
http://www.sap.com/company/legal/impressum.epx
-----Original Message-----
From: eclipselink-dev-bounces@xxxxxxxxxxx [mailto:eclipselink-dev-bounces@xxxxxxxxxxx] On Behalf Of Tom Ware
Sent: Montag, 28. März 2011 19:29
To: Dev mailing list for Eclipse Persistence Services
Cc: Singer, Reiner
Subject: Re: [eclipselink-dev] bug 340329 - table creation prefix
Hi Adrian,
I'd like to hear what other committers think but I think my preference is to specify this as a "table creation modifier" or some such thing. (your infix suggestion)
Unless we can think of a case where we would not have both the "CREATE" and "TABLE" strings, I think it would be cleaner to avoid requiring they be specified.
We thought about the delegation model you suggest below. The weakness is, of course, that it requires a code change to deal with any property that might be passed in. (i.e. if MySQL added a new modifier other than INNODB, we'd have to actually change the MySQLPlatform to support it). So far, the features that are addressed with the suffix are fairly obscure, so we thought it would be better to leave it free form. I am, however open to arguments the other way.
-Tom
Goerler, Adrian wrote:
Hi Tom,
yes, a "prefix" as propsed in the patch would substitute "CREATE TABLE". I agree that "prefix" does not appear to be the right terminology. "createTableStatement" isn't either as its only the first part of the statement. Maybe "createTableKeywords" would be better or "createTableStatementHeader".
It is an SAP-specific future feature we would like leverage, which allows to control some storage parameters of a table. The actual Syntax has the structure "CREATE <modifier> TABLE". Hence specifying the <modifier> as an "infix" would also be OK.
Still, I am afraid that this (prefix/suffix) opens a small can of worms and it might be cleaner to delegate writeCreateTable to the platform as scetched out below.
-Adrian
Adrian Görler
SAP AG
Pflichtangaben/Mandatory Disclosure Statements:
http://www.sap.com/company/legal/impressum.epx
-----Original Message-----
From: eclipselink-dev-bounces@xxxxxxxxxxx [mailto:eclipselink-dev-bounces@xxxxxxxxxxx] On Behalf Of Tom Ware
Sent: Montag, 28. März 2011 16:55
To: Dev mailing list for Eclipse Persistence Services
Cc: Xiang, Xu; Singer, Reiner
Subject: Re: [eclipselink-dev] bug 340329 - table creation prefix
What would a typical prefix be? (is it really a prefix, or a replacement for "CREATE TABLE"? Is PREFIX the right terminology?)
When would someone choose to use a prefix? Is this a MAXDB specific thing?
-Tom
Goerler, Adrian wrote:
Hi Chris, others,
https://bugs.eclipse.org/bugs/show_bug.cgi?id=340329
we got the requirement to allow overriding the CREATE TABLE keywords in DDL in a table-specific way to leverage special database features. Xu has proposed to introduce a creation-prefix attribute to the table-mappings of eclipselink-orm.xml - analogously to https://bugs.eclipse.org/bugs/show_bug.cgi?id=214519. Please find attached a revised proposal including test for this enhancement.
I you are OK with this feature, I would go ahead and check it in.
-Adrian
PS.
Alternatively, I could consider to specify additional requirements on the DDL using @Properties/@Property annotations. Then, one could add hese properties to the TableDefinition, redirect rendering of CREATE TABLE statements to the DatabasePlatform and render the statement in a database-vendor specific way according to the properties recognized by the vendor.
E.g.:
@Table(name="MY_TABLE")
@Property("mysql.jdbc.engine", "InnoDB")
@Entity
Public class MyEntity
This, however, would obsolete the creation-suffix just introduced in 2.2 ;-).
