adding metadata to java objects

[Ben Mabey]

Hi,
I would like to be able to add metadata to arbitrary java objects that
have already been instantiated. I know that you can use proxy to add
metadata to objects that you create but in my case the object already
exists (i.e. it is returned from another method call outside of my control).

It seems like the best solution would be to create a delegate
class/object that wraps the original one. Being able to write something
like this would be ideal:

(defn wrap-with-meta
([obj]
(wrap-with-meta obj nil))
([obj meta]
(delegate obj
clojure.lang.IObj
(withMeta [this new-meta] (wrap-with-meta
obj
new-meta))
(meta [this] meta))))

The delegate function would operate very similar to proxy, but instead
of taking a class it takes an object. A class would be created that
extends the object's class, just like a class is generated for proxy.
However, instead of stubs being generated that call super the stubs
would delegate to the given object. (Also note, I am also using
reify-like syntax since I prefer that to the syntax in proxy.)

I'm curious to hear people's thoughts on this approach. I'd also be
really interested to see how other people have addressed this in the
past (I doubt I'm the first one to run into this).

I know that clojure mentality is to avoid wrappers, but I don't see this
as being much different than what proxy already does. Of course, I may
be missing something... if so, please enlighten me. :)

Thanks,
Ben

[Ben Smith-Mannschott]

Here's an approach that may be of use: don't store the metadata in a
Map instead of decorating the Object with it. This map should use
object identity, not equality and should hold its keys weakly so that
it prevent collection of objects that otherwise would be garbage.
Safe to use concurrently would also be a plus. Conveniently, Google's
Guava library provides such a thing. Here's a sketch:

== project.clj ========================================

(defproject pojometa "0.0.1-SNAPSHOT"
:dependencies [[org.clojure/clojure "1.3.0"]
[com.google.guava/guava "10.0.1"]])

== src/pojometa/core.clj ==============================

(ns pojometa.core
(:import [com.google.common.collect MapMaker]))

(def meta-map
(-> (MapMaker.) .weakKeys .makeMap))

(defn meta* [o]
(if (instance? clojure.lang.IMeta o)
(clojure.core/meta o)
(.get meta-map o)))

(defn with-meta* [o m]
(if (instance? clojure.lang.IMeta o)
(clojure.core/with-meta o m)
(do (.put meta-map o m)
o)))

== usage ==============================================

pojometa.core=> (def o (Object.)) ;; arbitrary java object
pojometa.core=> (meta* (with-meta* o {:foo true}))
{:foo true}

;; also "does the right thing" for Clojure types that
;; already know how to have metadata.

pojometa.core=> (meta* (with-meta* {} {:bar 1}))
{:bar 1}

== END

Hope that helps,
Ben

[Alan Malloy]

On Nov 16, 11:53 am, Ben Smith-Mannschott <bsmith.o...@gmail.com>
wrote:

(def o (Object.))

(def om (with-meta* o {:foo true}))

(def whatever (with-meta* o {:foo false}))

(meta* om) ;=> {:foo false}

Doesn't really support Clojure's concept of metadata if it's shared
global mutable state.

[Ben Smith-Mannschott]

Yes, that's true. Though it wouldn't occur to me to expect some random
POJO to behave as if it were persistent WRT metadata. POJOs generally
conflate state and identity and are mutable. That's life. If that's a
problem, then you'll need to wrap it in something that supports
metadata.

// ben

[Chas Emerick]

On Nov 17, 2011, at 1:01 AM, Ben Smith-Mannschott wrote:

>> (def o (Object.))
>>
>> (def om (with-meta* o {:foo true}))
>>
>> (def whatever (with-meta* o {:foo false}))
>>
>> (meta* om) ;=> {:foo false}
>>
>> Doesn't really support Clojure's concept of metadata if it's shared
>> global mutable state.
>
> Yes, that's true. Though it wouldn't occur to me to expect some random
> POJO to behave as if it were persistent WRT metadata. POJOs generally
> conflate state and identity and are mutable. That's life. If that's a
> problem, then you'll need to wrap it in something that supports
> metadata.

Mechanisms like what Ben proposed have been discussed before, but declined for core (which is probably a good decision). However, something is generally better than nothing.

meta* isn't clojure.core/meta, so it can have weaker/different semantics without apologizing for it, IMO.

- Chas

[Chas Emerick]

Ben,

It's a good idea, but it's going to bite you:

- what if the class in question has fields or package-private methods that are touched by Java code?
- what if the class take ctor arguments?
- what if the ctor has side effects?

- Chas

> --
> You received this message because you are subscribed to the Google
> Groups "Clojure" group.
> To post to this group, send email to clo...@googlegroups.com
> Note that posts from new members are moderated - please be patient with your first post.
> To unsubscribe from this group, send email to
> clojure+u...@googlegroups.com
> For more options, visit this group at
> http://groups.google.com/group/clojure?hl=en

[Ben Mabey]

On 11/17/11 4:21 AM, Chas Emerick wrote:
> Ben,
>
> It's a good idea, but it's going to bite you:
>
> - what if the class in question has fields or package-private methods that are touched by Java code?

Yeah, there are some use cases where this would fall apart in. Assuming
that you use only public methods then everything should delegate as
expected. Point taken about the potential of unpleasant surprises.

> - what if the class take ctor arguments?
> - what if the ctor has side effects?

I don't think these last two points apply.. in my use case the object
has already been created and so the original constructor never is
called. The instantiated object is simply passed from one delegate to
another with the different metadata associated on to it.

-Ben