PetitJava performance
Locked
 
|
While working on PetitDelphi, we noticed significant performance
problems. As PetitDelphi copied the design of PetitJava, we found the same root cause. When creating a new java parser, a significant amount of time is spend on initialization, especially the creation of parsers for keywords, separators and operators. The introduction of a lexicon cache allows the tests to run significantly faster. 2230 ms to 224ms the second time. PetitJava-DiegoLont.115 Yuriy, can you merge? Stephan _______________________________________________ Moose-dev mailing list [hidden email] https://www.iam.unibe.ch/mailman/listinfo/moose-dev |
|
|
wow :)
On Jun 19, 2013, at 4:09 PM, Stephan Eggermont <[hidden email]> wrote: > While working on PetitDelphi, we noticed significant performance > problems. As PetitDelphi copied the design of PetitJava, we found > the same root cause. > > When creating a new java parser, a significant amount of time is spend > on initialization, especially the creation of parsers for keywords, separators and > operators. The introduction of a lexicon cache allows the tests to run > significantly faster. 2230 ms to 224ms the second time. > > PetitJava-DiegoLont.115 > > Yuriy, can you merge? > > Stephan > > > _______________________________________________ > Moose-dev mailing list > [hidden email] > https://www.iam.unibe.ch/mailman/listinfo/moose-dev _______________________________________________ Moose-dev mailing list [hidden email] https://www.iam.unibe.ch/mailman/listinfo/moose-dev |
|
|
Hi,
I'd like to merge, but right now I'm on vacation and do not have laptop with me. I'll be back on next weekend. Maybe Benjamin Arezki can take a look as he will be working on FAST for Java which also means working with PetitJava. Uko Надіслано з iPhone 19 черв. 2013 о 16:17 Esteban Lorenzano <[hidden email]> написав(ла): > wow :) > > On Jun 19, 2013, at 4:09 PM, Stephan Eggermont <[hidden email]> wrote: > >> While working on PetitDelphi, we noticed significant performance >> problems. As PetitDelphi copied the design of PetitJava, we found >> the same root cause. >> >> When creating a new java parser, a significant amount of time is spend >> on initialization, especially the creation of parsers for keywords, separators and >> operators. The introduction of a lexicon cache allows the tests to run >> significantly faster. 2230 ms to 224ms the second time. >> >> PetitJava-DiegoLont.115 >> >> Yuriy, can you merge? >> >> Stephan >> >> >> _______________________________________________ >> Moose-dev mailing list >> [hidden email] >> https://www.iam.unibe.ch/mailman/listinfo/moose-dev > > > _______________________________________________ > Moose-dev mailing list > [hidden email] > https://www.iam.unibe.ch/mailman/listinfo/moose-dev _______________________________________________ Moose-dev mailing list [hidden email] https://www.iam.unibe.ch/mailman/listinfo/moose-dev |
Re: PetitJava performance
|
|
In reply to this post by Stephan Eggermont-3
Yuriy,
Ok, I'm not in a hurry. Please note that your latest version is out of sync since PetitJava-MartinDias.104, and there are lots of versions to merge by cbc & ChrisCunningham. Automerge fails. Stephan _______________________________________________ Moose-dev mailing list [hidden email] https://www.iam.unibe.ch/mailman/listinfo/moose-dev |
|
|
Yes, I know, I have to talk to Chris, as he introduced a lot of strange nodes to AST, and I couldn't figure out why.
Надіслано з iPhone 20 черв. 2013 о 14:09 Stephan Eggermont <[hidden email]> написав(ла): > Yuriy, > > Ok, I'm not in a hurry. Please note that your latest version is out of sync > since PetitJava-MartinDias.104, and there are lots of versions to merge > by cbc & ChrisCunningham. Automerge fails. > > Stephan > _______________________________________________ > Moose-dev mailing list > [hidden email] > https://www.iam.unibe.ch/mailman/listinfo/moose-dev _______________________________________________ Moose-dev mailing list [hidden email] https://www.iam.unibe.ch/mailman/listinfo/moose-dev |
|
|
Hi Yuriy, The why is easy: I couldn't make sense of the deeply nested arrays that were present in the parts of the parsed code that weren't modeled in the existing AST, so I tried to make a model that I could have a chance of navigating.
As to the nodes themselves: - some of them were already there, and I just hooked them up to the parsers. - some I would claim aren't strange (such as the ForStatementNode),and probably don't deserve the label 'strange'.
- several really do deserve the label 'strange', though. These were ones where I wasn't sure what was being parsed out, and just blindly followed the parser naming convention. This is where the odd/strange nodes like PrimaryWithSelector came from. That one, at least, has been removed with a better set of nodes to replace it (MethodInvocationNode and VariableAccessorNode). The rest have not been rethought to make their usage clearer.
Going forward, I plan on removing some of the nodes (PJArguments and PJTypeArguments should just be an array in the containing node; PJThisIdentifierNode and PJSuperIdentifierNode were probably bad ideas to begin with); and re-think at least one of the others: PJClassOrInterfaceNode. If there are others that are still strange, please let me know which.
Interestingly, Stephan has commented most of the other nodes - thanks! -Chris On Thu, Jun 20, 2013 at 5:52 AM, Yuriy Tymchuk <[hidden email]> wrote: Yes, I know, I have to talk to Chris, as he introduced a lot of strange nodes to AST, and I couldn't figure out why. _______________________________________________ Moose-dev mailing list [hidden email] https://www.iam.unibe.ch/mailman/listinfo/moose-dev |
Re: PetitJava performance
|
|
In reply to this post by Stephan Eggermont-3
Chris wrote:
>Interestingly, Stephan has commented most of the other nodes - thanks! Can't claim that. We tried merging with the last version and had conflicts. We then used an earlier version that merged. Good to see it working though. I noticed you're committing as both cbc and ChrisCunningham? Seems to confuse smalltalkhub (it can't find user ChrisCunningham). Stephan _______________________________________________ Moose-dev mailing list [hidden email] https://www.iam.unibe.ch/mailman/listinfo/moose-dev |
|
|
On Thu, Jun 20, 2013 at 10:44 AM, Stephan Eggermont <[hidden email]> wrote:
It depends on which image I'm making the changes in - usually cbc is from a Squeak image, while ChrisCunningham is from the Moose image (because the Pharo folks prompt heavily for a full name usage). It helps me keep track of where the change occurred.
I was going to ask why it matters - but I think I've figured it out. SmalltalkHub won't show a commit by a non-registered user. Interesting, that. While poking around, I noticed that I can't see who else is a 'contributer' for this package on SmalltalkHub - only myself. I can of course guess by the names on the packages that show up in the commits tab.
-Chris _______________________________________________ Moose-dev mailing list [hidden email] https://www.iam.unibe.ch/mailman/listinfo/moose-dev |
|
|
Hi Stephan, I have merged Yuriy's branch and your's together - and removed a couple of nodes that I introduced. I do enjoy the speed-up. Yuriy: the merge was to include your alterations into my expanded nodes branch. It wasn't the other way around - to include Stephan's speed up into your original branch. Just an FYI.
-Chris On Thu, Jun 20, 2013 at 1:52 PM, Chris Cunningham <[hidden email]> wrote:
_______________________________________________ Moose-dev mailing list [hidden email] https://www.iam.unibe.ch/mailman/listinfo/moose-dev |
|
|
Yes, I noticed that PetitJava was changed as FAST builds became red. Now Benjamin has to start his work from understanding the cause of the problem. Morale: things have to be done right way. But I just didn't have enough time. Надіслано з iPhone
_______________________________________________ Moose-dev mailing list [hidden email] https://www.iam.unibe.ch/mailman/listinfo/moose-dev |
«
Return to Moose
|
1 view|%1 views
| Free forum by Nabble | Edit this page |