A tool for mining commits from Git repositories and diffs to automatically extract code change pattern instances and features with ast analysis
Home Page: https://hal.inria.fr/hal-00861883/file/paper-short.pdf
License: MIT License
I did not find any code related to this action type.
Does it mean persist?
These two snippets are from the NPEfix patch but I don't think that they could be generated by NPEfix repair tool regarding NPEfix's method. Am I right? @monperrus
Hi @martinezmatias ,
I have a doubt about designing test cases. I think I've two options right now:
Use @BeforeAll to process all the input files at once. After that, we just need to analyze the results in each of the test cases. Hence, the heavy processing is done only once.
The other is the usual choice of dividing files into different sets of test cases. This may result in a repetition of the same file as well as the processing of the same file more than once.
It's a tradeoff between testing time and modularity of test cases.
WDYT?
The Diff:
--- /src/main/java///org/apache/commons/math3/optimization/linear/SimplexSolver.java
+++ /src/main/java///org/apache/commons/math3/optimization/linear/SimplexSolver.java
@@ -120,7 +120,7 @@
for (int i = 0; i < tableau.getNumArtificialVariables(); i++) {
int column = i + tableau.getArtificialVariableOffset();
final double entry = tableau.getEntry(row, column);
- if (Precision.equals(entry, 1d, maxUlps) && row.equals(tableau.getBasicRow(column))) {
+ if (Precision.equals(entry, 1d, maxUlps) && !row.equals(tableau.getBasicRow(column))) {
return row;
}
}The pattern:
<pattern name="unary">
<entity id="2" type="UnaryOperator"/>
<action entityId="2" type="ANY"/>
</pattern>The commad:
coming -location ... -mode mineinstance -pattern unary.xml -input files -output ./unary_mineThe output:
{
"revision": "patch5-Math-28-jMutRepair",
"pattern_name": "unary",
"instance_detail": [
{
"pattern_action": "ANY",
"pattern_entity": {
"entity_type": "UnaryOperator",
"entity_new value": "*",
"entity_role": "*",
"entity_parent": "null"
},
"concrete_change": {
"operator": "INS",
"src_type": "UnaryOperator",
"dst_type": "null",
"src": "(!(row.equals(tableau.getBasicRow(column))))",
"dst": "null",
"src_parent_type": "BinaryOperator",
"dst_parent_type": "null",
"src_parent": "(org.apache.commons.math3.util.Precision.equals(entry, 1.0, maxUlps)) \u0026\u0026 (!(row.equals(tableau.getBasicRow(column))))",
"dst_parent": "null"
},
"file": "/test",
"line": 123
}
]
}In the above example the json output has source specified as (!(row.equals(tableau.getBasicRow(column)))) while that is actually the destination.
I observed this inversion present in most of the instances mined for the given pattern.
Any idea why this might be happening? @martinezmatias
When I attempted to input a git repository with 2000+ commits, I didn't get the same result as the test data set. Similarly, when input another git repository with 2000+commits, the results were the same.
java -classpath ./target/coming-0.1-SNAPSHOT-jar-with-dependencies.jar fr.inria.coming.main.ComingMain -location ./*/ -mode mineinstance -pattern ./pattern_INS_IF_MOVE_ASSIG.xml -output ./out/*
For sake of dissemination, it would be nice that coming has a 'default mode', taking a repo as input.
In this default mode, the output could be a nice graphical visualization à la Gource.
Hello @martinezmatias and thank you for your great library.
I am doing research on python language and I want to know if you added python support for coming.
I didn't see the language support in the README.
I would also be glad if I can contribute in adding Python support.
let me know if I can have your email address to contact you about this issue :)
As discussed with Samuel at KTH meetup, we should have the possibility to specify in the pattern specification the min distance between parent and children. The current distance corresponds to max
I think that something on the line of revision_format or input_format would be better. Just input seems to be a bit confusing(as it can be confused with location if a person hasn't read the documentation).
Can we make some sort of domain-specific language for specifying the changing pattern?
We can use some sort of modifiers to specify the actions. For example:
//: Move+/-: Update-: Remove+: AddExample specification 1
for(*) {
//a++;
}
This specifies a change in which a for was added(irrespective of the conditions) and a++ was moved.
Example specification 1
+/-for(*) {
// a++
}
This specifies a change in which a for was modified.
The main advantage of such an approach would be UX. I feel that this will make the process of specifying change pattern easier and hence improving the usability.
Thanks! Coming is very helpful for my research. Do you mind adding a few more descriptions about this tool?
thread to share paper links related to coming features, as well as tools
the Coming core should collect all JSON results, process them and write them to disk in coming_result.json.
For this, it'd be nice to have FinalResult be serializable to JSON
By default, Coming analyzes the last max_nb_commit_analyze commits (100 according to the configuration file).
The potential problem is that, if a user uses the command line with the min number of required parameters (location, inputs, a few more), he did not realize that the analysis is done over the partial portion of the input (e.g., git repo).
Should we put this feature as optional? Should be increase the value of max_nb_commit_analyze?
The tests of Coming are too verbose, and the console logs reach the Travis limit.
Consequently, it is impossible to debug the Travis failure, eg https://travis-ci.org/SpoonLabs/coming/builds/559463057
We must make the tests silent.
Hi, Is the textual diff already available incoming? @martinezmatias
when i am trying to upgrade GTSpoon to newer version (to solve issues on some datasets), one CI error caused by exceeding limit of log size was triggered.
Also, I suggest to move resources files only used by test files to .../src/test/resources
@ycaxgjd has done a first migration from the the Zhingxing's Coming folk to Coming.
However, it think we have still some differences i.e., Comings pushed to the mentioned folk after the migration.
marker issue for @sedflix' work
I modified this patch of code in fineGrainAnalyzer to find the difference between the source and target file. however, the source file address read by revision.getChildren(); is not available. Do you have any idea why this happens? @martinezmatias
an example of source file address printed out is src/test/java/org/apache/commons/lang3/builder/ToStringBuilderTest.java
public AnalysisResult<IRevision> analyze(IRevision revision) {
List<IRevisionPair> javaFiles = revision.getChildren();
Map<String, Diff> diffOfFiles = new HashMap<>();
log.info("\n*****\nCommit: " + revision.getName());
for (IRevisionPair<String> fileFromRevision : javaFiles) {
String left = fileFromRevision.getPreviousVersion();
String right = fileFromRevision.getNextVersion();
String leftName = fileFromRevision.getPreviousName();
String rightName = fileFromRevision.getName();
System.out.println("In fine grain analyzer...............................................................");
System.out.println(leftName);
System.out.println(rightName);
//build simple lists of the lines of the two testfiles
List<String> original = null;
try {
original = Files.readAllLines(new File(leftName).toPath());
} catch (IOException e) {
e.printStackTrace();
}
List<String> revised = null;
try {
revised = Files.readAllLines(new File(rightName).toPath());
} catch (IOException e) {
e.printStackTrace();
}
Diff diff = compare(left, right, leftName, rightName);
if (diff != null) {
diffOfFiles.put(fileFromRevision.getName(), diff);
//compute the patch: this is the diffutils part
Patch<String> patch = null;
try {
patch = DiffUtils.diff(original, revised);
} catch (DiffException e) {
e.printStackTrace();
}
System.out.println("patch .............");
System.out.println(patch);
}
}
return new DiffResult<IRevision, Diff>(revision, diffOfFiles);
}
New kind of input: to have two arguments corresponding to two files. Then, coming executes the analyzers over the pair.
Concerned patch:
--- /src/com/google/javascript/rhino/testing/Asserts.java
+++ /src/com/google/javascript/rhino/testing/Asserts.java
@@ -102,3 +102,3 @@
(a == null) == (b == null));
- if (a == null) {
+ if (message!=null) {
return;The pattern is not supposed to match the given change because a has been changed to message.
Pattern Specification:
<pattern name="binary">
<entity id="1" type="BinaryOperator"/>
<entity id="2" type="*" role="LEFT_OPERAND">
<parent parentId="1" distance="1"/>
</entity>
<entity id="3" type="*" role="RIGHT_OPERAND">
<parent parentId="1" distance="1"/>
</entity>
<action entityId="1" type="UPD"/>
<action entityId="2" type="UNCHANGED_HIGH_PRIORITY"/>
<action entityId="3" type="UNCHANGED_HIGH_PRIORITY"/>
</pattern>
Instance matched:
"instance_detail": [
{
"pattern_action": "UPD",
"pattern_entity": {
"entity_type": "BinaryOperator",
"entity_new value": "*",
"entity_role": "*",
"entity_parent": "null"
},
"concrete_change": {
"operator": "UPD",
"src_type": "BinaryOperator",
"dst_type": "BinaryOperator",
"src": "a == null",
"dst": "message != null",
"src_parent_type": "If",
"dst_parent_type": "If",
"src_parent": "if (a == null) {\n return;\n}",
"dst_parent": "if (message != null) {\n return;\n}"
},
"file": "/home/ubuntu/coming/patch1-Closure-7-Nopol2017",
"line": 103
}
]
The concerned files can be found here (private repo)
Hi Matias,
Do we support filtering by maximum number of AST changes?
I want to match the change of IF condition, i.e.
- if (a){
+ if (b){
System.out.println(a);
}
I have tried the specification like this one below:
<?xml version="1.0"?>
<pattern>
<entity id="1" type = "If"/>
<entity id="2" type = "*">
<parent parentId="1" distance="10" />
</entity>
<action entityId ="2" type = "*" />
</pattern>But it will also capture the change in the if block. For example, it will also match the change like this one.
if (a){
- System.out.println(a);
+ System.out.println(a);
}
How should I write the Pattern Specification to only match the changes in the condition part?
This is a proposal to add a check for properties like BinaryOperatorKind and UnaryOperatorKind in the Change Pattern Specification. For example: If entity="BinaryOperator" and kind="Plus", we only consider revisions in which Operator is of kind plus.
And will it be beneficial is add another abstraction over OperatorKinds like LogicalOperator etc?
In the run section of readme.md, I think there is a typo in the following line:
-parameters max_nb_hunks2::max_files_per_commit:1
@SophieHYe has done great OverfittingFeatures in Python, we can reimplement them in Java in Coming.
as proposed by Samuel at the KTH meetup, we can add multiple repos to analyze
I have written this test case for NPEfix and I can't figure out why gumtree does not find any difference between these two files? @martinezmatias
`
public static void main (String[] args) {
hello();
void hello(){
String ptr = null;
if (ptr.equals("gfg"))
System.out.print("Same");
else
System.out.print("Not Same");
}
}
`
`
public static void main (String[] args) {
hello();
void hello(){
String ptr = null;
return ;
if (ptr.equals("gfg"))
System.out.print("Same");
else
System.out.print("Not Same");
}
}
`
In the current implementation, I notice that the following pattern specification will also match the change with only one TypeReference entity insert.
<?xml version="1.0"?>
<pattern>
<entity id="1" type="TypeReference"/>
<entity id="2" type="TypeReference"/>
<action entityId="1" type="INS"/>
<action entityId="2" type="INS"/>
</pattern>This specification would match
+ int c;In other words, the entities with different ids may match the same operation in the AST.
Is this by design? I would prefer that the engine distinguishes entities with different IDs, which is more flexible.
I think the implementation may be here.
Hi @martinezmatias and @monperrus ,
This issue is regarding the question of how should we handle the patches for JGenProg.
Here most of the patches are:
These are very weak conditions by itself and all the commits lie in one of these categories.
And since the paper applies diff minimization, from my perspective, it would be hard(with only the current features of coming) to figure out whether the new statement inserted already existed in the code or not.
Q1. ) Does our implementation(the patches in DRR) use diff minimization?
Q2. ) Do we want to include other characteristics into coming? For example, running test cases and seeing the paths followed in the buggy version or other analysis based on test cases results. Or maybe accepting only git repo as an input for certain kind of repair-tool analysis so that we can search through code(like jGenProg). Or about including diff minimization dependencies?
Q3. ) How strong indicator our classification is supposed to be of that patch being generation by the concerned repair tool?
minesstub has an interesting list of patterns, together with their automated detectors. it would be a new feature to our feature list.
we could copy and integrate the code in coming
I was trying to run the module on commons-lang(5468 commits) and I got the following error:
Analyzing 1022/5468
2019-07-05 11:41:06,797 INFO fr.inria.coming.changeminer.analyzer.commitAnalyzer.FineGrainDifftAnalyzer -
*****
Commit: 5111ae7db08a70323a51a21df0bbaf46f21e072e
Exception in thread "main" java.lang.OutOfMemoryError: GC overhead limit exceeded
....
<this stack trace can be different>
Mode diff must be changed.
Mode mineinstance should be in plural
Hi @niloofartrg could you compute the current number of false positives and false negatives over @SophieHYe's drr?
We should be able to pass via Argument line several patterns
Hello, @martinezmatias @monperrus
I have been reading the Arja paper, but something that is not clear for me is that it seems the pattern files of Genprog and Arja would be the same. Am I right about this? Thanks!
Hello @monperrus @martinezmatias,
I have been trying to add support for SimFix repair tool but I can't understand something.
According to the below figure from the paper, to understand if a patch can be generated by SimFix I wanted to check that the change between my source and target file was in the frequent modifications set(S1) and the S2 set.
My problem is that I think that all changes between source and target files are in S2 by nature(!) because in order to produce S2 We have selected a faulty snippet of code from this same source file (which is now the line of code in my source file where there is a change with respect to the target file) and considered it's modification with all possible snippets from the target file and at some point we will generate the same modification that we are indeed looking for in S2.
in test logs, there are exceptions such as
--java.lang.ClassNotFoundException: vvvvv
java.lang.Exception: Error Loading Engine: java.lang.ClassNotFoundException: vvvvv
at fr.inria.coming.core.extensionpoints.PlugInLoader.loadPlugin(PlugInLoader.java:23)
at fr.inria.coming.main.ComingMain.loadAnalyzersFromCommand(ComingMain.java:313)
at fr.inria.coming.main.ComingMain.loadModelAnalyzers(ComingMain.java:301)
at fr.inria.coming.main.ComingMain.createEngine(ComingMain.java:218)
at fr.inria.coming.main.ComingMain.run(ComingMain.java:137)
at fr.inria.coming.spoon.patterns.InstanceMiningTest.testMainModeArg(InstanceMiningTest.java:101)
at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62)
at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
at java.lang.reflect.Method.invoke(Method.java:498)
at org.junit.runners.model.FrameworkMethod$1.runReflectiveCall(FrameworkMethod.java:50)
However, they do not indicate a test failure and as such are quite confusing.
We should not have any exception in the console test logs (related to #test)
fr.inria.coming.repairability.repiartools
should be
fr.inria.coming.repairability.repairtools
a refactoring is required
The repair analysis takes too much time on real repos. To make progress, we can add an option to analyze the last nth commits.
As you can see from here, on Win File.pathSeparator is ; rather than : which causes four failures when I ran testing on my Win machine.
In those errors, File.pathSeparator simply serves as the character : rather than pathSeparator that diffs on different systems. For example, README says you need the following command to combine filters,
-filter numberhunks:maxfiles -parameters max_nb_hunks2:max_files_per_commit:1
where : is used to separate different parameters and filter.
I will make a PR soon to fix problems like this.
Now, the callback can be only be register programmatically: Example:
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
Django
The Web framework for perfectionists with deadlines.
Laravel
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
Microsoft
Open source projects and samples from Microsoft.
Google
Google ❤️ Open Source for everyone.
Alibaba
Alibaba Open Source for everyone
D3
Data-Driven Documents codes.
Tencent
China tencent open source team.