defrecord and overriding "empty"

[Nathan Sorenson]

I'm using clojure.walk/postwalk to rewrite terms in nested data
structures. However, I am unable to do this with types as defined by
defrecord, because they specify that the function "empty" throw a not-
implemented exception.

If I were able to over-ride this default implementation of 'empty' I
believe I could still use postwalk to rewrite my custom records (as
they happily implement 'into' which works as expected.)

This was my obvious first attempt:

(defrecord TypeVar [guid name]
clojure.lang.IPersistentCollection
(empty [coll] (TypeVar. 0 nil)))

Which fails due to:

"Duplicate method name&signature in class file types/TypeVar"
[Thrown class java.lang.ClassFormatError]

I would prefer not to fall back on deftype as I do wish these objects
to behave as maps. I also could write my own version of postwalk that
doesn't call 'empty' on my custom-defined records, but it would be
nice to make my types adhere to the contract of IPersistentCollection
as much as possible.

[David Nolen]

The problem is what does empty mean in the context of defrecord? A new instance of a defrecord is not "empty", it has those fields and they are set to nil.

David

[Nathan Sorenson]

You're right that my use isn't strictly returning a collection with a size of zero-- I'm treating empty more like 'default'. I'm thinking of its use in clojure.walk, which simply creates a "blank" version of an arbitrary collection in which to place the altered sub-forms. I can't find any other usages of 'empty' in the std lib.

I agree it's a bit of a mismatch, as 'walk' won't be able to alter the built-in keys of a record, but you wouldn't expect that to happen if you were using records.

Should IPersistentCollection even be defining 'empty', if one of the language's key data types doesn't support it? I think it would be better to either pull 'empty' into it's own protocol so clojure.walk doesn't match on IPersistentCollection when walking a data structure with the false expectation that it can create an empty version, or rename 'empty' it to something that records can implement, like 'default.'

[David Nolen]

On Tue, Sep 27, 2011 at 1:54 PM, Nathan Sorenson <nd...@sfu.ca> wrote:

Should IPersistentCollection even be defining 'empty', if one of the language's key data types doesn't support it? I think it would be better to either pull 'empty' into it's own protocol so clojure.walk doesn't match on IPersistentCollection when walking a data structure with the false expectation that it can create an empty version, or rename 'empty' it to something that records can implement, like 'default.' 

This is a fair point. ClojureScript fixes this by providing IEmptyableCollection.

David 

[David Nolen]

In core.logic we have a protocol called IUninitialized that you can implement. That works well enough for our needs.

On Saturday, December 15, 2012, Joseph Smith wrote:

Well here I am over a year later with the same problem. I'm curious, what was your solution for the time being?

--
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