<< Back to previous view

[LOGIC-166] Reification exponential in depth of tree terms? Created: 27/Jan/15  Updated: 28/Jan/15

Status: Open
Project: core.logic
Component/s: None
Affects Version/s: None
Fix Version/s: None

Type: Defect Priority: Minor
Reporter: Tom Jack Assignee: David Nolen
Resolution: Unresolved Votes: 0
Labels: performance
Environment:

[org.clojure/clojure "1.7.0-alpha5"]
[org.clojure/core.logic "0.8.9"]


Attachments: Text File 0001-Walk-less.patch    

 Description   

By 'reification' I mean e.g.:

(run 1 [t]
  (== t <some-constant-to-be-reified>))

Demonstration of exponential performance, using nested vectors ([], [[]], [[[]]]...):
https://www.refheap.com/1e30c198d528300fcba9ef24a

Is this expected, or a bug? I'm guessing the former, but hoping the latter. Feel free to close without comment if this is just the way it is.



 Comments   
Comment by Tom Jack [ 28/Jan/15 4:31 AM ]

Attached patch 0001-Walk-less.patch gets me this nice result. It also makes my real case work quite speedily (25ms now, where before it took so long that I never saw it finish). That's a huge relief!

The problem wasn't reification, but walk-term. In the patch I attempt to avoid doing some walking which appears to be redundant. I hope the fact that most of the tests still pass is good evidence that it's redundant.

However, the patch breaks some unifier tests. I noticed that my refheap test 'passes' as of bd65104ec3~, but fails as of bd65104ec3 (the patch for LOGIC-69). It seems the fix made the unifier rely on the current behavior of walk-term.

I'm going to press on with my patched ./checkouts copy for now, since I'm not using the unifier. I may come back to try to fix it sometime.





Generated at Thu Jan 29 06:36:19 CST 2015 using JIRA 4.4#649-r158309.