antlr/codebuff antlr/codebuff

  • Stars
    star
    440
  • Rank 99,050 (Top 2 %)
  • Language
    Java
  • License
    BSD 2-Clause "Sim...
  • Created almost 9 years ago
  • Updated over 4 years ago
Twitter logo Share LinkedIn logo Share Facebook logo Share
antlr/codebuff
github

antlr

Reviews

There are no reviews yet. Be the first to send feedback to the community and the maintainers!

Repository Details

Language-agnostic pretty-printing through machine learning (uh, like, is this possible? YES, apparently).

CodeBuff smart formatter

By Terence Parr (primary developer), Fangzhou (Morgan) Zhang (help with initial development), Jurgen Vinju (co-author of academic paper, help with empirical results and algorithm discussions).

kaby76 has done a C# port.

Abstract

Code formatting is not particularly exciting but many researchers would consider it either unsolved or not well-solved. The two well-established solutions are:

  1. Build a custom program that formats code for specific a language with ad hoc techniques, typically subject to parameters such as "always put a space between operators".
  2. Define a set of formal rules that map input patterns to layout instructions such as "line these expressions up vertically".

Either techniques are painful and finicky.

This repository is a step towards what we hope will be a universal code formatter that uses machine learning to look for patterns in a corpus and to format code using those patterns.

It requires Java 8. See pom.xml for dependencies (e.g., ANTLR 4.x, ...).

Whoa! It appears to work. Academic paper, Towards a Universal Code Formatter through Machine Learning accepted to SLE2016. Sample output is in the paper or next section. Video from Terence's presentation.

Sample output

All input is completed squeezed of whitespace/newlines so only the output really matters when examining CodeBuff output. You can check out the output dir for leave-one-out formatting of the various corpora. But, here are some sample formatting results.

SQL

SELECT *
FROM DMartLogging
WHERE DATEPART(day, ErrorDateTime) = DATEPART(day, GetDate())
      AND DATEPART(month, ErrorDateTime) = DATEPART(month, GetDate())
      AND DATEPART(year, ErrorDateTime) = DATEPART(year, GetDate())
ORDER BY ErrorDateTime
    DESC
SELECT
    CASE WHEN SSISInstanceID IS NULL
        THEN 'Total'
    ELSE SSISInstanceID END SSISInstanceID
    , SUM(OldStatus4) AS OldStatus4
    , SUM(Status0) AS Status0
    , SUM(Status1) AS Status1
    , SUM(Status2) AS Status2
    , SUM(Status3) AS Status3
    , SUM(Status4) AS Status4
    , SUM(OldStatus4 + Status0 + Status1 + Status2 + Status3 + Status4) AS InstanceTotal
FROM
    (
        SELECT
            CONVERT(VARCHAR, SSISInstanceID)             AS SSISInstanceID
            , COUNT(CASE WHEN Status = 4 AND
                              CONVERT(DATE, LoadReportDBEndDate) <
                              CONVERT(DATE, GETDATE())
                        THEN Status
                    ELSE NULL END)             AS OldStatus4
            , COUNT(CASE WHEN Status = 0
                        THEN Status
                    ELSE NULL END)             AS Status0
            , COUNT(CASE WHEN Status = 1
                        THEN Status
                    ELSE NULL END)             AS Status1
            , COUNT(CASE WHEN Status = 2
                        THEN Status
                    ELSE NULL END)             AS Status2
            , COUNT(CASE WHEN Status = 3
                        THEN Status
                    ELSE NULL END)             AS Status3
--, COUNT ( CASE WHEN Status = 4 THEN Status ELSE NULL END ) AS Status4
            , COUNT(CASE WHEN Status = 4 AND
                              DATEPART(DAY, LoadReportDBEndDate) = DATEPART(DAY, GETDATE())
                        THEN Status
                    ELSE NULL END)             AS Status4
        FROM dbo.ClientConnection
        GROUP BY SSISInstanceID
    ) AS StatusMatrix
GROUP BY SSISInstanceID

Java

public class Interpreter {
    ...
    public static final Set<String> predefinedAnonSubtemplateAttributes = new HashSet<String>() {
                                                                              {
                                                                                  add("i");
                                                                                  add("i0");
                                                                              }
                                                                          };
...
    public int exec(STWriter out, InstanceScope scope) {
        final ST self = scope.st;
        if ( trace ) System.out.println("exec("+self.getName()+")");
        try {
            setDefaultArguments(out, scope);
            return _exec(out, scope);
        }
        catch (Exception e) {
            StringWriter sw = new StringWriter();
            PrintWriter pw = new PrintWriter(sw);
            e.printStackTrace(pw);
            pw.flush();
            errMgr.runTimeError(this,
                                scope,
                                ErrorType.INTERNAL_ERROR,
                                "internal error: "+sw.toString());
            return 0;
        }
    }
...
    protected int _exec(STWriter out, InstanceScope scope) {
        final ST self = scope.st;
        int start = out.index(); // track char we're about to write
        int prevOpcode = 0;
        int n = 0; // how many char we write out
        int nargs;
        int nameIndex;
        int addr;
        String name;
        Object o, left, right;
        ST st;
        Object[] options;
        byte[] code = self.impl.instrs;        // which code block are we executing
        int ip = 0;
        while ( ip<self.impl.codeSize ) {
            if ( trace|| debug ) trace(scope, ip);
            short opcode = code[ip];
            //count[opcode]++;
            scope.ip = ip;
            ip++; //jump to next instruction or first byte of operand
            switch ( opcode ) {
                case Bytecode.INSTR_LOAD_STR:
                    // just testing...
                    load_str(self, ip);
                    ip += Bytecode.OPND_SIZE_IN_BYTES;
                    break;
                case Bytecode.INSTR_LOAD_ATTR:
                    nameIndex = getShort(code, ip);
                    ip += Bytecode.OPND_SIZE_IN_BYTES;
                    name = self.impl.strings[nameIndex];
                    try {
                        o = getAttribute(scope, name);
                        if ( o== ST.EMPTY_ATTR ) o = null;
                        }
                    catch (STNoSuchAttributeException nsae) {
                        errMgr.runTimeError(this, scope, ErrorType.NO_SUCH_ATTRIBUTE, name);
                        o = null;
                    }
                    operands[++sp] = o;
                    break;
...

ANTLR

referenceType : classOrInterfaceType | typeVariable | arrayType ;

classOrInterfaceType
    :   (   classType_lfno_classOrInterfaceType
        |   interfaceType_lfno_classOrInterfaceType
        )
        (   classType_lf_classOrInterfaceType
        |   interfaceType_lf_classOrInterfaceType
        )*
    ;

classModifier
    :   annotation
    |   'public'
    |   'protected'
    |   'private'
    |   'abstract'
    |   'static'
    |   'final'
    |   'strictfp'
    ;

typeSpecifier
    :   (   'void'
        |   'char'
        |   'short'
        |   'int'
        |   'long'
        |   'float'
        |   'double'
        |   'signed'
        |   'unsigned'
        |   '_Bool'
        |   '_Complex'
        |   '__m128'
        |   '__m128d'
        |   '__m128i'
        )
    |   '__extension__' '(' ('__m128' | '__m128d' | '__m128i') ')'
    |   atomicTypeSpecifier
    |   structOrUnionSpecifier
    |   enumSpecifier
    |   typedefName
    |   '__typeof__' '(' constantExpression ')' // GCC extension
    ;

Build complete jar

To make a complete jar with all of the dependencies, do this from the repo main directory:

$ mvn clean compile install

This will leave you with artifact target/codebuff-1.4.19.jar or whatever the version number is and put the jar into the usual maven local cache.

Formatting files

To use the formatter, you need to use class org.antlr.codebuff.Tool. Commandline usage:

  • -g grammar-name. The grammar must be run through ANTLR and be compiled (and in the CLASSPATH). For example, for Java8.g4, use -g Java8, not the filename. For separated grammar files, like ANTLRv4Parser.g4 and ANTLRv4Lexer.g4, use -g ANTLRv4. If the grammar is in a package, use fully-qualified like -g org.antlr.codebuff.ANTLRv4.
  • -rule start-rule. Start rule of the grammar where parsing of a full file starts, such as compilationUnit in Java.g4.
  • -corpus root-dir-of-samples
  • [-files file-extension]. E.g., use java, g4, c, ...
  • [-indent num-spaces]. This defaults to 4 spaces indentation.
  • [-comment line-comment-name]. As a failsafe, CodeBuff allows you to specify the token name for single-line comments, such as LINE_COMMENT, within the grammar so that it can ensure there is a line break after a single line,.
  • [-o output-file]. Filename with optional path to where output should go.
  • file-to-format. Filename (with optional path) must be last.

Output goes to standard out unless you use -o.

$ java -jar target/codebuff-1.4.19.jar  \
       -g org.antlr.codebuff.ANTLRv4 \
       -rule grammarSpec \
       -corpus corpus/antlr4/training \
       -files g4 \
       -indent 4 \
       -comment LINE_COMMENT \
       T.g4
$ java -jar target/codebuff-1.4.19.jar \
       -g org.antlr.codebuff.Java \
       -rule compilationUnit \
       -corpus corpus/java/training/stringtemplate4 \
       -files java \
       -comment LINE_COMMENT \
       T.java

These examples work for the grammars specified because they are already inside the complete jar. For parsers compiled outside of the jar, you might need to do something like:

java java -cp target/codebuff-1.4.19.jar:$CLASSPATH \
       org.antlr.codebuff.Tool  \
       -g org.antlr.codebuff.ANTLRv4 \
       -rule grammarSpec -corpus corpus/antlr4/training \
       -files g4 -indent 4 -comment LINE_COMMENT T.g4

Grammar requirements

All whitespace should go to the parser on a hidden channel. For example, here is a rule that does that:

WS  :	[ \t\r\n\f]+ -> channel(HIDDEN)	;

Comments should also:

BLOCK_COMMENT
	:	'/*' .*? ('*/' | EOF)  -> channel(HIDDEN)
	;

LINE_COMMENT
	:	'//' ~[\r\n]*  -> channel(HIDDEN)
	;

You can have line comments match newlines if you want.

Speed tests

The paper cites some speed tests for training and formatting time for

First, here is my machine configuration:

Memory speed seems to make a big difference given how much we have to trawl through memory---The tests shown below were done with 1867 MHz DDR3 RAM. We set an initial 4G RAM, 1M stack size. First build everything:

$ mvn clean compile install

Then you can run the speed tests as shown in following subsections.

ANTLR corpus

$ java -Xmx4G -Xss1M -cp target/codebuff-1.4.19.jar org.antlr.codebuff.validation.Speed -antlr corpus/antlr4/training/Java8.g4
Loaded 12 files in 172ms
antlr training of /Users/parrt/antlr/code/codebuff/corpus/antlr4/training/Java8.g4 = 353ms formatting = 340ms
antlr training of /Users/parrt/antlr/code/codebuff/corpus/antlr4/training/Java8.g4 = 188ms formatting = 161ms
antlr training of /Users/parrt/antlr/code/codebuff/corpus/antlr4/training/Java8.g4 = 145ms formatting = 153ms
antlr training of /Users/parrt/antlr/code/codebuff/corpus/antlr4/training/Java8.g4 = 130ms formatting = 129ms
antlr training of /Users/parrt/antlr/code/codebuff/corpus/antlr4/training/Java8.g4 = 123ms formatting = 113ms
antlr training of /Users/parrt/antlr/code/codebuff/corpus/antlr4/training/Java8.g4 = 114ms formatting = 116ms
antlr training of /Users/parrt/antlr/code/codebuff/corpus/antlr4/training/Java8.g4 = 93ms formatting = 90ms
antlr training of /Users/parrt/antlr/code/codebuff/corpus/antlr4/training/Java8.g4 = 80ms formatting = 90ms
antlr training of /Users/parrt/antlr/code/codebuff/corpus/antlr4/training/Java8.g4 = 73ms formatting = 88ms
antlr training of /Users/parrt/antlr/code/codebuff/corpus/antlr4/training/Java8.g4 = 72ms formatting = 71ms
antlr training of /Users/parrt/antlr/code/codebuff/corpus/antlr4/training/Java8.g4 = 71ms formatting = 69ms
antlr training of /Users/parrt/antlr/code/codebuff/corpus/antlr4/training/Java8.g4 = 71ms formatting = 73ms
antlr training of /Users/parrt/antlr/code/codebuff/corpus/antlr4/training/Java8.g4 = 76ms formatting = 63ms
antlr training of /Users/parrt/antlr/code/codebuff/corpus/antlr4/training/Java8.g4 = 70ms formatting = 70ms
antlr training of /Users/parrt/antlr/code/codebuff/corpus/antlr4/training/Java8.g4 = 70ms formatting = 69ms
antlr training of /Users/parrt/antlr/code/codebuff/corpus/antlr4/training/Java8.g4 = 73ms formatting = 70ms
antlr training of /Users/parrt/antlr/code/codebuff/corpus/antlr4/training/Java8.g4 = 70ms formatting = 68ms
antlr training of /Users/parrt/antlr/code/codebuff/corpus/antlr4/training/Java8.g4 = 71ms formatting = 66ms
antlr training of /Users/parrt/antlr/code/codebuff/corpus/antlr4/training/Java8.g4 = 70ms formatting = 70ms
antlr training of /Users/parrt/antlr/code/codebuff/corpus/antlr4/training/Java8.g4 = 73ms formatting = 72ms
median of [5:19] training 72ms
median of [5:19] formatting 70ms

Guava corpus, Java grammar

$ java -Xms4G -Xss1M -cp target/codebuff-1.4.19.jar org.antlr.codebuff.validation.Speed -java_guava corpus/java/training/guava/cache/LocalCache.java
Loaded 511 files in 1949ms
java_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1984ms formatting = 2669ms
java_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1747ms formatting = 3166ms
java_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1784ms formatting = 2811ms
java_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1507ms formatting = 1742ms
java_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1499ms formatting = 2832ms
java_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1582ms formatting = 2663ms
java_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1499ms formatting = 2807ms
java_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1561ms formatting = 2815ms
java_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1521ms formatting = 2136ms
java_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1545ms formatting = 2811ms
java_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1501ms formatting = 2800ms
java_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1506ms formatting = 2581ms
java_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1494ms formatting = 2838ms
java_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1494ms formatting = 2789ms
java_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1497ms formatting = 2621ms
java_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1501ms formatting = 2714ms
java_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1506ms formatting = 2816ms
java_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1512ms formatting = 2733ms
java_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1515ms formatting = 2587ms
java_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1508ms formatting = 2430ms
median of [5:19] training 1506ms
median of [5:19] formatting 2733ms

Guava corpus, Java8 grammar

Load time here is very slow (2.5min) because the Java8 grammar is meant to reflect the language spec. It has not been optimized for performance. Once the corpus is loaded, training and formatting times are about the same as for Java grammar.

$ java -Xms4G -Xss1M -cp target/codebuff-1.4.19.jar \
       org.antlr.codebuff.validation.Speed \
       -java8_guava corpus/java/training/guava/cache/LocalCache.java
Loaded 511 files in 159947ms
java8_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 2238ms formatting = 23312ms
java8_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1913ms formatting = 2368ms
java8_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1855ms formatting = 2277ms
java8_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1856ms formatting = 2267ms
java8_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1868ms formatting = 2348ms
java8_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1890ms formatting = 2263ms
java8_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1866ms formatting = 2328ms
java8_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1855ms formatting = 2247ms
java8_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1856ms formatting = 2243ms
java8_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1871ms formatting = 2204ms
java8_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1863ms formatting = 2244ms
java8_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1850ms formatting = 2212ms
java8_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1861ms formatting = 2215ms
java8_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1877ms formatting = 2257ms
java8_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1843ms formatting = 2249ms
java8_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1842ms formatting = 2205ms
java8_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1869ms formatting = 2343ms
java8_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1864ms formatting = 2225ms
java8_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1851ms formatting = 2260ms
java8_guava training of /Users/parrt/antlr/code/codebuff/corpus/java/training/guava/cache/LocalCache.java = 1871ms formatting = 2200ms
median of [5:19] training 1863ms
median of [5:19] formatting 2244ms

Generating graphs from paper

In the Towards a Universal Code Formatter Through Machine Learning paper, we have three graphs to support our conclusions. This sections shows how to reproduce them. (Note that these jobs take many minutes to run; maybe up to 30 minutes for one of them on a fast box.)

The Java code generates python code that uses matplotlib. The result of running the python is a PDF of the graph (that also pops up in a window).

Box plot with median error rates

To generate:

do this:

$ mvn clean compile install
$ java -Xms8G -Xss1M -cp target/codebuff-1.4.19.jar org.antlr.codebuff.validation.LeaveOneOutValidator
...
wrote python code to python/src/leave_one_out.py
$ cd python/src
$ python leave_one_out.py &

Plot showing effect of corpus size on error rate

To generate:

do this:

$ mvn clean compile install
$ java -Xms8G -Xss1M -cp target/codebuff-1.4.19.jar org.antlr.codebuff.validation.SubsetValidator
...
wrote python code to python/src/subset_validator.py
$ cd python/src
$ python subset_validator.py &

Plot showing effect of varying model parameter k

To generate:

do this:

$ mvn clean compile install
$ java -Xms8G -Xss1M -cp target/codebuff-1.4.19.jar org.antlr.codebuff.validation.TestK
...
wrote python code to python/src/vary_k.py
$ cd python/src
$ python vary_k.py &

More Repositories

1

antlr4

ANTLR (ANother Tool for Language Recognition) is a powerful parser generator for reading, processing, executing, or translating structured text or binary files.
Java
17,043
star
2

grammars-v4

Grammars written for ANTLR v4; expectation that the grammars are free of actions.
ANTLR
10,114
star
3

stringtemplate4

StringTemplate 4
Java
943
star
4

intellij-plugin-v4

An IntelliJ plugin for ANTLR v4
Java
467
star
5

antlrcs

The C# port of ANTLR 3, StringTemplate 3, and StringTemplate 4
C#
281
star
6

antlr3

antlr v3 repository (pulled from p4 with history from //depot/code/antlr/antlr3-main
Java
241
star
7

antlr4-intellij-adaptor

A library to support the use of ANTLR grammars in jetbrains IDE plugins for building custom languages.
Java
209
star
8

jetbrains-plugin-sample

A sample plugin for jetbrains IDEs that uses an ANTLR grammar for a nontrivial custom language.
Java
118
star
9

antlrworks

AntlrWorks tool for ANTLR
Java
97
star
10

antlr4-tools

Tools to run antlr4 w/o needing to install java or antlr4!
Python
83
star
11

antlr-php-runtime

PHP Runtime for ANTLR4
PHP
81
star
12

antlr4-cpp

C++ target for ANTLR 4, including templates and runtime
C++
75
star
13

examples-v3

Java
59
star
14

antlr5

Java
52
star
15

symtab

Generic symbol table for lexically/statically scoped languages
Java
48
star
16

Antlr4Formatter

Formatter for ANTLR 4 grammars
ANTLR
40
star
17

antlr4-lab

A client/server for trying out and learning about ANTLR
Java
35
star
18

grammars-v3

Grammars written for ANTLR v3; expectation that the grammars are free of actions but it's not a hard and fast rule. there is no common license!
GAP
24
star
19

jetbrains-plugin-st4

An intellij/jetbrains plugin for StringTemplate .st and .stg files
Java
23
star
20

antlr4test-maven-plugin

A maven plugin used to test the grammars-v4 repo grammars
Java
17
star
21

stringtemplate3

StringTemplate v3
Java
15
star
22

website-antlr3

The ANTLR 3 Website
HTML
13
star
23

website-antlr4

The ANTLR 4 Website
HTML
11
star
24

antlr-php-runtime-phpstan

Static analysis for ANTLR4 PHP Runtime.
PHP
6
star
25

antlr5-specs

Specs for antlr5
4
star
26

grammars-v5

3
star
27

_grammar-test

Historical files from grammars-v4 repo
Java
2
star
28

performance

Test the performance of ANTLR parsers (initially just Java target)
Java
2
star
29

website-antlr2

The ANTLR 2 Website
Java
1
star
30

website-st4

stringtemplate.org content
HTML
1
star
31

antlr5-lab

TypeScript
1
star