Optimizing Parser Combinators

1

PetitParser CompilerJan Kurš & Jan Vraný

Mohammad Ghafari, Mircea Lungu and

Oscar Nierstrasz

2

3

Parsing Contexts

Bounded Seas

Furthest Failure

Layout-senstivity

4

5

6

id ← #letter #word*

block ← #INDENT expr / block /

… #DEDENT

expr ← term /‘(‘ expr ‘)’ /…

7

id ← #letter #word*

block ← #INDENT expr / block /

… #DEDENT

expr ← term /‘(‘ expr ‘)’ /…

8

id ← #letter #word*

block ← #INDENT expr / block /

… #DEDENT

expr ← term /‘(‘ expr ‘)’ /…

9

id ← #letter #word*

expr ← term /‘(‘ expr ‘)’ /…

10

id ← #letter #word*

expr ← term /‘(‘ expr ‘)’ /…

11

id ← #letter #word*

block ← #INDENT expr / block /

… #DEDENT

expr ← term /‘(‘ expr ‘)’ /…

12

id ← #letter #word*

block ← #INDENT expr / block /

… #DEDENT

expr ← term /‘(‘ expr ‘)’ /…

13

id ← #letter #word*

block ← #INDENT expr / block /

… #DEDENT

expr ← term /‘(‘ expr ‘)’ /…

14

id ← #letter #word*

block ← #INDENT expr / block /

… #DEDENT

expr ← term /‘(‘ expr ‘)’ /…

15

id ← #letter #word*

block ← #INDENT expr / block /

… #DEDENT

expr ← term /‘(‘ expr ‘)’ /…

16

17

id ← #letter #word*

block ← #INDENT expr / block /

… #DEDENT

expr ← term /‘(‘ expr ‘)’ /…

18

PetitParserCompiler

19

10k methods of Pharo code

3 parsers

Compared to Compiled PetitParser

Speedup

20PetitParser SmaCC RBParser

0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

21Expressions Python Smalltalk Java Ruby

0

2

4

6

8

10

12

14

22

PetitParser SmaCC RBParser0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

Compiler

23

map:

Plusprogram

TokenclassToken

TokenidToken

Sequencebody

Actionclass

Sequenceidentifier

Sequence

ClassNode new  id: idToken value  body: body 

CharClass#letter

ChoiceletterOrDigit

CharClass#digit

Star

… Indentindent

Star Dedentdedent

Choice

Actionmethod

…

24

0

0.2

0.4

0.6

0.8

1

1.2Ti

me [

mic

rose

conds]

Time per Character of Smalltalk

Compiled ParserGC time