[jira] [Commented] (IVY-1465) ModuleRules.getRule is O(n) leading to resolution slowness

["Shikhar Bhushan (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/IVY-1465?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13973105#comment-13973105 ] 

Shikhar Bhushan commented on IVY-1465:
--------------------------------------

+1 on this -- 

we are using plain ant+ivy (not sbt) and we had a major build time regression after upgrading from Solr 4.6 -> 4.7, it got 3-4x slower

this was all time spent in ivy:resolve

but using sbt's ivy 2.4.0 pre-release built with this patch (https://repo.typesafe.com/typesafe/releases/org.scala-sbt.ivy/ivy/2.4.0-sbt-d6fca11d63402c92e4167cdf2da91a660d043392/jars/) fixes this slowness

> ModuleRules.getRule is O(n) leading to resolution slowness
> ----------------------------------------------------------
>
>                 Key: IVY-1465
>                 URL: https://issues.apache.org/jira/browse/IVY-1465
>             Project: Ivy
>          Issue Type: Bug
>          Components: Core
>    Affects Versions: 2.3.0
>         Environment: OpenJDK 6,7, MacOSX + Ubuntu
>            Reporter: Joshua Suereth
>            Priority: Minor
>              Labels: performance
>
> So, for the sbt project we noticed significant resolution time improvements with the following patch:
> https://github.com/sbt/ivy/pull/1
> The reasoning:
> A lot of artifacts being resolved use Maven's "dependencyManagement" conventions
> Ivy appears to turn these into "exact matcher" rules
> A ton of resolution time is spent filtering through these rules
> The existing solution is O(n) for all overrides
> What the patch does:
> Creates a key'd store for all "exact matcher" rules
> When executing rules, ensure that we only traverse what we have to (non-exact, exact specific to our key and "default").
> As I said, this represents a significant speed bump for sbt builds using Ivy.  All existing tests pass, and I think they cover this aspect of ivy pretty well, from what I could see.



--
This message was sent by Atlassian JIRA
(v6.2#6252)