petit parser performance issues
Locked
 
|
Hi,
Recent changes in PetitParser have caused parsing to slow down by a factor of 15-20. Here is an example run on freshly downloaded Moose image:
[|string allStyles text| string := (PharoFilesOpener default changesFileOrNil contents copyFrom: 1 to: 60000). text := string asText. allStyles := { 'Announcement' -> TextColor green.
'Collections' -> TextColor blue. 'File' -> TextColor blue. 'Metacello' -> TextColor red. 'Monticello' -> TextColor magenta. 'Morphic' -> TextColor orange.
'Mooose' -> TextColor green. 'FAMIX' -> TextColor green. 'Roassal' -> TextColor green. }. allStyles do: [ :assoc | ((assoc key asParser, #newline asParser negate star) matchingRangesIn: string) do: [:aRange | text addAttribute: (assoc value) from: aRange first to: aRange last] ].
text] timeToRun ==> 15 sec ============== now do this: Gofer new url: 'http://smalltalkhub.com/mc/Moose/PetitParser/main'; version:'PetitParser-DiegoLont.233'; load. and run the example again. timeToRun ===> 2 sec So, we need to have a look at what is added recently in PetitParser that is making it slow down quite significantly because the for the larger systems, this can be enormous. regards, Usman P.S: Tests run on OS X 10.9.4, 2.3 GHz, 8 GB RAM, HD SSD.
_______________________________________________ Moose-dev mailing list [hidden email] https://www.iam.unibe.ch/mailman/listinfo/moose-dev |
Re: petit parser performance issues
|
|
Hi, Thanks for raising this. We did not have time to announce these changes properly. We integrated Jan's work on supporting island parsing and enhanced traceability. The features are really cool. To this end, the stream of petit parser was replaced with a first class context that holds the stream.
We did expect a slight slowdown, but not a factor of 7 (which is what I get from your measurements) though. Doru On Fri, Aug 22, 2014 at 1:30 PM, Usman Bhatti <[hidden email]> wrote:
"Every thing has its own flow"
_______________________________________________ Moose-dev mailing list [hidden email] https://www.iam.unibe.ch/mailman/listinfo/moose-dev |
|
|
Hey, I can confirm the slowdown. The latest change (JanKurs.238, committed today) should return most of the performance back. Please, let me know, if you have any other issues
Cheers, Jan On 22 August 2014 16:42, Tudor Girba <[hidden email]> wrote:
_______________________________________________ Moose-dev mailing list [hidden email] https://www.iam.unibe.ch/mailman/listinfo/moose-dev |
Re: petit parser performance issues
|
|
Hi Jan, Thanks a lot for looking into this. I checked, but I think the improvement you mention did not make it in the repo of PetitParser. Version JanKurs.238 is from August 20 (before the slowdown report):
Name: PetitParser-JanKurs.238 Author: JanKurs Time: 20 August 2014, 10:13:55.838908 am UUID: a7316ed3-0751-4703-80c9-2790add1cc00 Ancestors: PetitParser-JanKurs.237
The latest version is JanKurs.239 and it is also from the same day: Name: PetitParser-JanKurs.239 Author: JanKurs Time: 20 August 2014, 12:13:26.196695 pm
UUID: 8985ce0d-8d40-47ef-bd82-ead0d8695be2 Ancestors: PetitParser-JanKurs.238 Could you please double check? Cheers, Doru On Sun, Aug 24, 2014 at 3:56 PM, Jan Kurš <[hidden email]> wrote:
"Every thing has its own flow"
_______________________________________________ Moose-dev mailing list [hidden email] https://www.iam.unibe.ch/mailman/listinfo/moose-dev |
|
|
Hey, sorry for a complication, the branching system is a too tricky for me. The version JanKurs.240 should do the thing. I just checked in a build Moose-50 #2222 (https://ci.inria.fr/moose/job/moose-5.0/2222/) Cheers, Jan On 24 August 2014 15:02, Tudor Girba <[hidden email]> wrote:
_______________________________________________ Moose-dev mailing list [hidden email] https://www.iam.unibe.ch/mailman/listinfo/moose-dev |
Re: petit parser performance issues
|
|
Great! I tested it. The example of Usman went from 15s to 3s. @Usman: could you test and see if it is acceptable? Cheers, Doru
On Sun, Aug 24, 2014 at 5:16 PM, Jan Kurš <[hidden email]> wrote:
"Every thing has its own flow"
_______________________________________________ Moose-dev mailing list [hidden email] https://www.iam.unibe.ch/mailman/listinfo/moose-dev |
Re: petit parser performance issues
|
|
Hi,
On Sun, Aug 24, 2014 at 5:30 PM, Tudor Girba <[hidden email]> wrote:
I tested the example and we are down to roughly 3-4 seconds. Thank you Doru and Jan for looking into it.
_______________________________________________ Moose-dev mailing list [hidden email] https://www.iam.unibe.ch/mailman/listinfo/moose-dev |
Re: petit parser performance issues
|
|
In reply to this post by Tudor Girba-2
On Fri, Aug 22, 2014 at 5:42 PM, Tudor Girba <[hidden email]> wrote:
I understand that context is important when parsing a program for certain parsing related information. In my case, I am using PetitParser more as a syntax highlighter on the lines of PPTextHighlighter and in that case you dont need much of contextual information. All you need is token positions. Can we envisage, for such usage, a simplistic but more robust parser?
usman
_______________________________________________ Moose-dev mailing list [hidden email] https://www.iam.unibe.ch/mailman/listinfo/moose-dev |
Re: petit parser performance issues
|
|
2014-08-25 11:12 GMT+02:00 Usman Bhatti <[hidden email]>:
If you're more into that, then a lexer may be enough and a very efficient one can be built with SmaCC. What you're doing doesn't seems to require a parser: it is just matching Keyword [^\n] * isn't it? Thierry
[|string allStyles text| string := (PharoFilesOpener default changesFileOrNil contents copyFrom: 1 to: 60000). text := string asText. allStyles := { 'Announcement' -> TextColor green.
'Collections' -> TextColor blue. 'File' -> TextColor blue. 'Metacello' -> TextColor red. 'Monticello' -> TextColor magenta. 'Morphic' -> TextColor orange.
'Mooose' -> TextColor green. 'FAMIX' -> TextColor green. 'Roassal' -> TextColor green. }. allStyles do: [ :assoc | ((assoc key asParser, #newline asParser negate star) matchingRangesIn: string) do: [:aRange | text addAttribute: (assoc value) from: aRange first to: aRange last] ].
text] timeToRun ==> 15 sec _______________________________________________ Moose-dev mailing list [hidden email] https://www.iam.unibe.ch/mailman/listinfo/moose-dev |
Re: petit parser performance issues
|
|
On Mon, Aug 25, 2014 at 12:50 PM, Thierry Goubier <[hidden email]> wrote:
It was just a illustrative example to show the slowness of PParser due to the recent modifications. What I want to achieve is to be able to express keyword or lines matching (may be some day complete code blocks as well). We found that doing it with the parser is quite easy.
It would be worth trying SmaCC to compare its performance with PP-based solution but I am not knowledgable about SmaCC. Any pointers to where should I start from? regards,
Usman
_______________________________________________ Moose-dev mailing list [hidden email] https://www.iam.unibe.ch/mailman/listinfo/moose-dev |
Re: petit parser performance issues
|
|
2014-08-25 16:11 GMT+02:00 Usman Bhatti <[hidden email]>:
If there is little context around (deep tree structures, backtracking) then a lexer may be enough. If you give me a bit of context (or a petit parser description), I can work out the equivalent SmaCC grammar.
I updated the SmaCC chapter in the PharoForTheEnterprise book; it is also available as help inside Pharo itself. https://ci.inria.fr/pharo-contribution/job/PharoForTheEnterprise/ws/Smacc/Smacc.pier.pdf Thierry
_______________________________________________ Moose-dev mailing list [hidden email] https://www.iam.unibe.ch/mailman/listinfo/moose-dev |
Re: petit parser performance issues
|
|
Tx Thierry. I'd prefer to read and understand the basics of SmaCC first from the Pharo book for intelligent communication :). I'll let you know.
Usman On Mon, Aug 25, 2014 at 4:48 PM, Thierry Goubier <[hidden email]> wrote:
_______________________________________________ Moose-dev mailing list [hidden email] https://www.iam.unibe.ch/mailman/listinfo/moose-dev |
Re: petit parser performance issues
|
|
Le 25/08/2014 18:36, Usman Bhatti a
écrit :
Well, I teach parsing with SmaCC, so I may help :) Thierry
_______________________________________________ Moose-dev mailing list [hidden email] https://www.iam.unibe.ch/mailman/listinfo/moose-dev |
|
|
In reply to this post by Thierry Goubier
I'm really happy that you did it ;)
_______________________________________________ Moose-dev mailing list [hidden email] https://www.iam.unibe.ch/mailman/listinfo/moose-dev |
|
|
In reply to this post by Thierry Goubier
On 25/8/14 19:02, Thierry Goubier
wrote:
Thierry do you have slides and related material? Stef
_______________________________________________ Moose-dev mailing list [hidden email] https://www.iam.unibe.ch/mailman/listinfo/moose-dev |
Re: petit parser performance issues
|
|
Le 27/08/2014 23:12, stepharo a écrit :
I should write some this year. At least if I manage to get some time between the current projects and the project proposals :( Thierry _______________________________________________ Moose-dev mailing list [hidden email] https://www.iam.unibe.ch/mailman/listinfo/moose-dev |
|
|
>> Thierry do you have slides and related material? > I should write some this year. At least if I manage to get some time > between the current projects and the project proposals :( ahh that..... I'm writing a new ERC proposals so I'm allowed to open Pharo for a couple of weeks. No fun. _______________________________________________ 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 |