CLJ loop bindings hold on to head of lazy sequence

[Luke VanderHart]

Consider the following Clojure code:

(let [s (range 10000000)]

  (reduce + s))

This behaves nicely and returns the value 49999995000000

Now consider this code:

(loop [s (range 10000000)]

  (reduce + s))

In this case, the loop binding holds on to the head of the sequence,

resulting in an OutOfMemoryError.

Of course, using loop without recur is pointless: this is only a

degenerate case to illustrate the problem. The issue actually does

show up in real code, where a loop might have a conditional to either

recur or bail to an alternative handler. This is actually happening

right now in the clojure.lang.Object impelementation of

clojure.core.protocols/internal-reduce.

Should I file this as a bug, or is there a good reason for this that I

haven't thought of?

Thanks,

-Luke

[Tassilo Horn]

Luke VanderHart <luke.va...@gmail.com> writes:

> Consider the following Clojure code:
>
> (let [s (range 10000000)]
> (reduce + s))
>
> This behaves nicely and returns the value 49999995000000
>
> Now consider this code:
>
> (loop [s (range 10000000)]
> (reduce + s))
>
> In this case, the loop binding holds on to the head of the sequence,
> resulting in an OutOfMemoryError.

Running 1.4.0, both work fine for me. What version are you using?

Bye,
Tassilo

[Tassilo Horn]

Tassilo Horn <tas...@member.fsf.org> writes:

>> Consider the following Clojure code:
>>
>> (let [s (range 10000000)]
>> (reduce + s))
>>
>> This behaves nicely and returns the value 49999995000000
>>
>> Now consider this code:
>>
>> (loop [s (range 10000000)]
>> (reduce + s))
>>
>> In this case, the loop binding holds on to the head of the sequence,
>> resulting in an OutOfMemoryError.
>
> Running 1.4.0, both work fine for me. What version are you using?

Oh, too fast. I can reproduce the issue by using a hundred times longer
range...

Bye,
Tassilo

[Luke VanderHart]

Yes, the size of the range needed to cause the error varies depending how much memory you're running with.

The error is present on the master branch as of this morning.

[Tassilo Horn]

Luke VanderHart <luke.va...@gmail.com> writes:

Hi Luke,

> Yes, the size of the range needed to cause the error varies depending
> how much memory you're running with.
>
> The error is present on the master branch as of this morning.

LetExpr in Compiler.java is the implementation of both let* and loop*.
Not that I'd understand it right now, but in the loop* case it pushes
and pops some thread bindings and the LOOP_LOCALS dynamic var looks
suspicious...

Bye,
Tassilo