<< Back to previous view

[CLJ-1606] Transducing an eduction finishes twice Created: 27/Nov/14  Updated: 09/Jan/15  Resolved: 09/Jan/15

Status: Closed
Project: Clojure
Component/s: None
Affects Version/s: Release 1.7
Fix Version/s: Release 1.7

Type: Defect Priority: Major
Reporter: Herwig Hochleitner Assignee: Unassigned
Resolution: Completed Votes: 0
Labels: transducers
Environment:

1.7.0-alpha4


Attachments: Text File CLJ-1606-2.patch     Text File CLJ-1606-3.patch     Text File CLJ-1606-4.patch     Text File CLJ-1606-5.patch     Text File CLJ-1606.patch    
Patch: Code and Test
Approval: Ok

 Description   
> (transduce (map identity)
             (fn
               ([s] (println "Finishing") s)
               ([s i] s))
             nil
             (eduction (map identity) []))
Finishing
Finishing
nil

Cause: transduce passes (xf f) into .reduce of Eduction, which calls transduce, causing completing xf to be called more than once.

Proposed: Eduction reduce should use (completing f) instead of f to isolate completion of inner xf from outer xf.

Patch: CLJ-1606-5.patch

Screened by: Alex Miller



 Comments   
Comment by Alex Miller [ 27/Nov/14 11:01 PM ]

identity is not a valid xf - changed to (map identity)

Comment by Ghadi Shayban [ 27/Nov/14 11:34 PM ]

identity is a valid though nonsensical transducer. fix & test added.

Comment by Ghadi Shayban [ 28/Nov/14 12:06 AM ]

Simple reproduction similar to into:

(transduce (map dec)
           (completing conj! persistent!)
           (transient [])
           (eduction (map inc) (range 6)))

;; ClassCastException clojure.lang.PersistentVector cannot be cast to clojure.lang.ITransientCollection

into doesn't use completing, and conj! has an arity that hides the problem.

Comment by Alex Miller [ 28/Nov/14 8:54 AM ]

I removed trailing whitespace in the patch so it applies cleanly.

Comment by Ghadi Shayban [ 14/Dec/14 11:16 PM ]

This patch is a little more subtle than I thought. Completion of the eduction's rfn needs to be handled separately from the "outer" transduce's xform. Patch coming.

Comment by Ghadi Shayban [ 14/Dec/14 11:32 PM ]

New patch with tests that completes the inner xform without completing the passed in rfn

Comment by Ghadi Shayban [ 15/Dec/14 1:19 AM ]

both -3 and -2 are equivalent. -3 is probably better stylistically.

Comment by Alex Miller [ 15/Dec/14 8:37 AM ]

Added CLJ-1606-4.patch - identical to -3, just fixed whitespace error.

Comment by Andy Fingerhut [ 08/Jan/15 6:09 PM ]

There are two identically named attachments here (containing -2). It looks like it isn't the one under consideration, but it might be nice to remove or rename to avoid the name conflict.

Comment by Ghadi Shayban [ 08/Jan/15 6:24 PM ]

Andy, not sure how to do that, but in any case I just added -5 clarifying language in the comment

Comment by Alex Miller [ 08/Jan/15 6:41 PM ]

Ghadi, that was super confusing. Did you just add a new -5 patch? The -4 patch has already been screened, and you have not removed the duplicate -2 patch so I don't get what the -5 is. Can we just delete the -5 and older -2 patches?

Comment by Andy Fingerhut [ 08/Jan/15 6:44 PM ]

Sorry for adding to the confusion. Ghadi, instructions for deleting patches are in the "Removing patches" section on this wiki page: http://dev.clojure.org/display/community/Developing+Patches

Comment by Ghadi Shayban [ 08/Jan/15 6:50 PM ]

Sorry. Fine by me, though permissions prevent me from deleting one of the patches.

As I read through the screened patch I just tried to clarify the wording. This:

;; NB (completing f) isolates completion of inner xfns from outer xfns
became:
;; NB (completing f) isolates completion of inner rf from outer rf

Feel free to nix that -5 patch if that's worthless

Comment by Alex Miller [ 08/Jan/15 7:12 PM ]

Gotcha. I will take care of the further changes later tonight.

In the future, please don't modify screened patches without letting me know.

Generated at Tue Nov 13 12:54:25 CST 2018 using JIRA 4.4#649-r158309.