Is RBClassNotReferencedRule ~200x slower in Pharo 4 vs Pharo 3 ?
Locked
Is RBClassNotReferencedRule ~200x slower in Pharo 4 vs Pharo 3 ?
 
|
I think I'm setting this up wrong in Pharo 4 but am not sure. I use the following code for the test:
rule:=RBClassNotReferencedRule new. environment:=RBBrowserEnvironment new forPackageNames: #('MyPackage'). AndreasSystemProfiler spyOn:[RBSmalllintChecker runRule: rule onEnvironment: environment] In Pharo 4 it spends most of the time in Behavior>>#includesDefinedSelector:ofClassName: and takes 300s. In Pharo 3 its only 1.5 s Is there something that can be done to get Pharo 4 up to the speed of Pharo 3? Thanks Paul |
|
|
Thanks for the question.
It looks like we got a regression. Le 29/7/15 06:04, Paul DeBruicker a écrit : > I think I'm setting this up wrong in Pharo 4 but am not sure. I use the following code for the test: > > > rule:=RBClassNotReferencedRule new. > environment:=RBBrowserEnvironment new forPackageNames: #('MyPackage'). > > > AndreasSystemProfiler spyOn:[RBSmalllintChecker runRule: rule onEnvironment: environment] > > > In Pharo 4 it spends most of the time in Behavior>>#includesDefinedSelector:ofClassName: and takes 300s. In Pharo 3 its only 1.5 s > > Is there something that can be done to get Pharo 4 up to the speed of Pharo 3? > > > Thanks > > Paul > |
|
|
In reply to this post by Paul DeBruicker
Paul
I wanted to check on Pharo 50 but we need a comparison point. We should get the same package. Do you have an idea of a package that did not change much? Setf Le 29/7/15 06:04, Paul DeBruicker a écrit : > I think I'm setting this up wrong in Pharo 4 but am not sure. I use the following code for the test: > > > rule:=RBClassNotReferencedRule new. > environment:=RBBrowserEnvironment new forPackageNames: #('MyPackage'). > > > AndreasSystemProfiler spyOn:[RBSmalllintChecker runRule: rule onEnvironment: environment] > > > In Pharo 4 it spends most of the time in Behavior>>#includesDefinedSelector:ofClassName: and takes 300s. In Pharo 3 its only 1.5 s > > Is there something that can be done to get Pharo 4 up to the speed of Pharo 3? > > > Thanks > > Paul > |
Re: Is RBClassNotReferencedRule ~200x slower in Pharo 4 vs Pharo 3 ?
|
|
Hi Stef,
The slowdown is dependent upon the # of classes in the package. If you use Fuel then you should see a slowdown of 20x. I just have a lot of classes in this package. The source of the slowdown is in the changes to RBSmallLintChecker>>#checkClass: and RBClassNotReferencedRule>>#checkClass: . The old versions use the linter to check if the literal is used anywhere and the new one uses #allCallsOn. Hope this helps Paul |
Re: Is RBClassNotReferencedRule ~200x slower in Pharo 4 vs Pharo 3 ?
|
|
and by "use Fuel" I mean: rule:=RBClassNotReferencedRule new. environment:=RBBrowserEnvironment new forPackageNames: #('Fuel'). AndreasSystemProfiler spyOn:[RBSmalllintChecker runRule: rule onEnvironment: environment] |
|
|
In reply to this post by Paul DeBruicker
Paul
I'm running out of town. Could you open a bug entry with the information you have so that we do not forget? Stef Le 29/7/15 17:33, Paul DeBruicker a écrit : > Hi Stef, > > The slowdown is dependent upon the # of classes in the package. If you use > Fuel then you should see a slowdown of 20x. I just have a lot of classes in > this package. The source of the slowdown is in the changes to > RBSmallLintChecker>>#checkClass: and RBClassNotReferencedRule>>#checkClass: > . The old versions use the linter to check if the literal is used anywhere > and the new one uses #allCallsOn. > > > Hope this helps > > Paul > > > > -- > View this message in context: http://forum.world.st/Is-RBClassNotReferencedRule-200x-slower-in-Pharo-4-vs-Pharo-3-tp4839875p4839976.html > Sent from the Pharo Smalltalk Users mailing list archive at Nabble.com. > > |
Re: Is RBClassNotReferencedRule ~200x slower in Pharo 4 vs Pharo 3 ?
|
|
In reply to this post by Paul DeBruicker
I added an issue:
i think we need to put in some cashing… here if we would could all *used* classes once for an enviroment….
|
|
|
Hi marcus
do you get an idea why we got this slowdown? Stef Le 30/7/15 14:08, Marcus Denker a
écrit :
I added an issue: |
|
|
This may have happened when I was removing contexts from the rules. Uko
|
Re: Is RBClassNotReferencedRule ~200x slower in Pharo 4 vs Pharo 3 ?
|
|
I think the reason is that there was a cache in the context. But we need to check.
|
|
|
In reply to this post by Uko2
yuriy what was a context?
Stef Le 18/8/15 22:39, Yuriy Tymchuk a
écrit :
|
|
|
I need to double check it, for now I cannot access the issue tracker (probably local internet issues). As far as I remember, the problem was that SmallLint rules where checking contexts and not classes and methods. I.e. you had to create a context, set it’s “selected" class or method, pass it into the rule and then rule would take out class or method out of the context and check them. In my humble opinions this is 1) pain, 2) a bit confuting that #chackClass: method accepts context and not a class. But I guess at some point there was some caching happening during the check in a certain rules, and now it’s not there. What I was suggesting is to either define a context attribute for all rules, or even better, define a hierarchy of rules that besides the entity to check accept the context that is needed for the check, as not avery rule requires a context. By default it can be a live context of whale system, but on demand we can create a snapshot of a context (with caches) and pass it to different rules several times. Sadly I never managed to finish implementing it, it’s still in my todo list. Uko
|
«
Return to Pharo Smalltalk Users
|
1 view|%1 views
| Free forum by Nabble | Edit this page |