(Reactivating discussion. I guess it's been discussed more on the
committer list, but here you have my 2 cents anyway)
For the sake of the Record-JPA discussion, people will fall into two
categories when they are using lift:
* The first group of people have an existing, working, tested JPA/OR
based data access library written in Java and are looking to integrate
that with a webapp written in lift. They will usually be coming from
an enterprise background, and will have some constraints on what they
can develop from scratch ("no - we will not rewrite all the db access
code to support the new web framework").
* The second group doesn't have an existing data access library in
Java and would like to write all their new stuff with lift in Scala.
But maybe they have mapping/usage requirements that precludes using
mapper because it's too simple, or they just know their way around one
of the other JPA enabled OR libs and want to bank on that knowledge
(without writing the entire layer in Java).
* The third group also doesn't have an existing data access library,
but for some reasons, they want their OR model not only be accessible
from lift, but actually also from Java code. This means that they not
only have to be able use JPA, but that the Scala code they write has
to be _JPA compatible_ (Java usable JPA entities etc.).
For the first group, some sort of delegation strategy from the Record
entity to the JPA entity is necessary (unless you can mixin the record
stuff on top of an actual JPA entity?). The JPA/OR configuration is
already in place on the existing lib (annotations/xml). The only thing
that record is doing is adding another layer on top of it for
validation, binding to UI, maybe querying. Something like what was
proposed earlier by Derek should be enough
> class MyEntry extends Record[MyEntry] {
> var inner : RealEntry = _
>
> object name extends StringField(this) with Delegation
> name.delegate(inner.name, inner.name = _)
>
> ...
> }
Now for the second group, they probably would like to have all the
power associated with a JPA/OR framework without resolving to writing
that layer in Java. Mind you that this is quite a different problem
from what the first group wants to solve.
* They just want to have Record entities, no separate JPA entity, no
delegation.
* They don't want to duplicate configuration from what they already
specify on Record entities.
* Have the integration as hassle free as possible, because in their
eyes, there is no actual delegation.
What the second group needs is a bit trickier, and most of the
discussion has revolved around how to make it as transparent as
possible (compiler plugins, byte code weaving, etc.).
Remember that for the second group of people Record is attempting to
solve the _exact same problem_ as JPA. It's a unifying frontend, that
abstracts away differences in persistence libraries. And as with JPA,
Record has the configuration present in code (instead of annotations
it just uses actual subclasses and parameters, but still).
But there's one thing that has only been mentioned on the off by
David.
> I believe there are ways to use Record's deep knowledge of class layouts to
> generate data structures and/or XML to send to JPA to "do the right thing."
The key point here is XML. While those annotations are a nice feature
for the programmer to get rid of the XML configuration from pre EJB
3.0, it's less than sexy for attaching a unified API (record) to an
implementation library (jpa, which incidentially is another unified
API for the actual OR libs :/ ). But that's a feature for the
developer who doesn't want to specify the configuration in an XML. But
the XML configuration capability is still there. Generating XML from
Record entities (maybe even in memory) to boot the JPA/OR lib should
be pretty easy and doesn't require more opaque techniques (compiler
plugin, byte code weaving). And it can be overridden easily as well
even to integrate things that are not supported in Record.
I'm not well versed enough in JPA to make statements about whether the
XML configuration is feature complete compared to the annotations, and
how flexible the mapping can be (directly from the Record object
fields to the columns). I would assume that the mapping is not
flexible enough to work with the objects fields from Record, but maybe
we can start discussing strategies to solve that (using vars,
generating bytecode, using untyped maps as entities, generating OR lib
specific XML - And yes I know that with a couple of those you loose
some features provided by JPA).
For the third group of people, those that want to use reuse their
objects as Java compatible JPA entities: Do we really need that right
now? Are there enough real use cases to warrant the effort? Mind that
you can always write a Java compatible access layer for your entities
in Scala. In that layer you refrain from using stuff that Java doesn't
like and then just use that connector from your java code. IMO lift
shouldn't solve that problem. Although it's interesting, I don't
really see the benefit of having actual JPA object usable from Java.
As already discussed on the list, there's a whole bunch of
compatibility problems to make the Record entity be an actual JPA
entity. If we find out that solving the problem for the second group
is only possible if we solve this one, then fine, otherwise I
personally wouldn't bother to fix that thing now. You can code around
it and it shouldn't be that much of a problem to do so.
Pick your battles...
Daniel
On Dec 2, 1:22 am, "David Pollak" <feeder.of.the.be...@gmail.com>
wrote:
> My 2 cents:
>
> - I'm strongly opposed to any compiler plugins as they (1) mean that IDEs
> will work less well and (2) they require some sort of installation (if they
> can be rolled into the Maven building stuff, it makes this objection go
> away)
> - I'm strongly opposed to mixing annotations and the Record stuff. It'll
> just make for wicked ugly looking code.
>
> I believe there are ways to use Record's deep knowledge of class layouts to
> generate data structures and/or XML to send to JPA to "do the right thing."
>
> On Sun, Nov 30, 2008 at 7:45 AM, Derek Chen-Becker
> <dchenbec...@gmail.com>wrote:
>
>
>
> > The whole point of integrating (and I use the word integrating here
> > loosely) is so that there's a common form framework for people to use.
> > Really, the point of Record as far as I can tell is to loosen and/or remove
> > the tight coupling with the datastore, and in that sense Record is
> > succeeding. As Tim points out, there are going to be people with existing
> > JPA libraries and what we're trying to propose is a means for them to
> > leverage their existing libraries alongside the nice work done in Record. As
> > ugly as the boilerplate for delegation may seem, I don't see any better
> > solution without adding significant complexity. I've done bytecode
> > manipulation in the past and while it can be very handy, it makes things a
> > lot more complex on the backend and I'm really trying to avoid complexity
> > (this includes a compiler plugin, too).
>
> > Delegation provides the cleanest approach I can come up with; it actually
> > helps in the case where someone wants to use an existing JPA library because
> > you could simply wrap an instance and delegate to it:
>
> > class MyEntry extends Record[MyEntry] {
> > var inner : RealEntry = _
>
> > object name extends StringField(this) with Delegation
> > name.delegate(inner.name, inner.name = _)
>
> > ...
> > }
>
> > By adding a few lines of boilerplate in each class you get validation, form
> > generation, presentation HTML, JSON representation, etc. It's up to the
> > end-user to determine whether these features are worth the cost of a few
> > LoC.
>
> > I'm not saying this is a perfect solution, just that I think that this is
> > the best first cut at it. Also, no one is saying that people have to use JPA
> > with Record. I'm assuming that the majority of new Lift apps will probably
> > use Record/Mapper because they're simple, easy to use and quite capable of
> > handling most data models. But for the people with existing JPA libraries,
> > or for people who would like to use JPA for the additional features,
> > compatibility with Java apps, etc, I'd like to have a better answer than an
> > absolute "Don't use Record" or "Switch to Record".
>
> > Derek
>
> > On Sun, Nov 30, 2008 at 1:40 AM, Marius <marius.dan...@gmail.com> wrote:
>
> >> I totally agree with you Viktor! ... although it seems like a workable
> >> solution in this case a Record contains to distinct entities to
> >> represent the same data and of course the boilerplate.
>
> >> It was mentioned above compiler plugins which I expressed my opinion
> >> about it (I'd stay away) but there might be another way to "alter" the
> >> bytecode at runtime. I'm talking about dynamic class weaving.
> >> Basically it is a class loader that before returning the actual
> >> updated class. Kind of what AspectJ is doing. A while ago a wrote a
> >> bytecode level manipulation framework and combined with a "special"
> >> classloader I was able to modify a class dynamically and use it.
>
> >> Of course there are caveats:
>
> >> 1. Complexity induced by bytecode level manipulation
> >> 2. How a dynamic class loader cope with container's classloader that
> >> essentially loads the web application. Of course it would delegate the
> >> loading to the container's classloader but these "modifiable" classes
> >> should probably not live in WEB-INF/classes or WEB-INF/lib folder.
>
> >> But all in all I'm not convinced that this effort really worth it.
> >> AFAIC I still don;t get the whole point of integrating Record/Field
> >> with JPA. If someone wants to switch easily from JPA to Record or vice-
> >> versa, to have this flexibility perhaps consider to abstract these
> >> layers from the rest of the application and using other persistence
> >> technologies would not affect the application "business logic" ... but
> >> this is about how the appliation is designed etc.
>
> >> Br's,
> >> Marius
>
> >> On Nov 30, 9:56 am, "Viktor Klang" <viktor.kl...@gmail.com> wrote:
> >> > IMHO:
>
> >> > " �...@column{val name = "my_name"}
> >> > var name : String = _
> >> > object nameField extends StringField(this,100) with Delegated
> >> > nameField.delegate(name, name = _)"
>
> >> > That's just too much boilerplate. For me, who went to scala to lose
> >> > boilerplate, this is not a viable solution.
>
> >> > Unfortunately the JPA spec is kind of weak when it comes to adaptation,
> >> so
> >> > my suggestion to make it work with Hibernate first still stands.
>
> >> > What do you guys think?
>
> >> > Cheers,
> >> > Viktor
>
> >> > On Sat, Nov 29, 2008 at 10:23 PM, Derek Chen-Becker
> >> > <dchenbec...@gmail.com>wrote:
>
> >> > > No big deal. In the example code I did a by-name in the delegate
> >> method so
> >> > > at least it's transparent to the end-user.
>
> >> > > Derek
>
> >> > > On Sat, Nov 29, 2008 at 11:06 AM, Jorge Ortiz <jorge.or...@gmail.com
> >> >wrote:
>
> >> > >> Just wanted to chime in real quick...
>
> >> > >>> type Getter = () => MyType // Can this be by-name in any way?
>
> >> > >> No. By-names are not first-class types. It's gotta be () => MyType
>
> >> > >> --j
>
> >> > --
> >> > Viktor Klang
> >> > Senior Systems Analyst
>
> --
> Lift, the simply functional web frameworkhttp://liftweb.net
> Collaborative Task Managementhttp://much4.us
> Follow me:http://twitter.com/dpp
> Git some:http://github.com/dpp
--~--~---------~--~----~------------~-------~--~----~
You received this message because you are subscribed to the Google Groups
"Lift" group.
To post to this group, send email to liftweb@googlegroups.com
To unsubscribe from this group, send email to
liftweb+unsubscr...@googlegroups.com
For more options, visit this group at
http://groups.google.com/group/liftweb?hl=en
-~----------~----~----~----~------~----~------~--~---
