Skip to content

hephaestus-compiler-project/hephaestus-pldi-eval

Repository files navigation

Artifact for "Finding Typing Compiler Bugs" (PLDI'22)

This artifact is for the PLDI'22 paper titled "Finding Typing Compiler Bugs".

An archived version of the artifact is also available on Zenodo. See https://zenodo.org/record/6330357.

Table of Contents

Overview

The artifact contains the instructions and scripts to re-run the evaluation described in our paper. The artifact has the following structure:

  • scripts/: This directory contains the scripts needed to re-run the experiments presented in our paper.
  • data/: This is the directory that contains the precomputed results of our evaluation.
  • database/bug_schema.sql: This is the database schema that contains the bugs discovered by our approach.
  • database/bugdb.sqlite3: This is the sqlite3 database file corresponding to our bug database.
  • database/bugs.json: This JSON file contains the bugs of database/bugdb.sqlite.
  • hephaestus/: Contains the source code of the tool (provided as a git submodule) used for testing the compilers of Java, Kotlin, and Groovy. The name of our tool is Hephaestus.
  • installation_scripts/: Contains helper scripts used to install all dependencies (e.g., compiler versions from SDKMAN).
  • figures/: This directory will be used to save figure 8 of the paper.
  • Dockerfile: The Dockerfile used to create a Docker image of our artifact. This image contains all data and dependencies.

Hephaestus is available as open-source software under the GNU General Public License v3.0, and can also be reached through the following repository: https://github.com/hephaestus-compiler-project/hephaestus.

Inside the hephaestus/ directory, there are the following directories:

  • src/: The source code of Hephaestus, which is written in Python.
  • tests/: Contains the tests of Hephaestus.
  • deployment/: Contains configuration and scripts to install and run Hephaestus on a machine every second day.

Requirements

See REQUIREMENTS.md

Setup

See INSTALL.md

Getting Started

We will use the Docker image (namely hephaestus-eval) built by the instructions from the Setup guide to get started with Hephaestus. Recall that this image contains all the required environments for testing the three compilers (i.e., it includes installations of the corresponding compilers, as well as any other tool needed for processing the results).

You can enter a new container by using the following command:

docker run -ti --rm hephaestus-eval

Usage

hephaestus provides a rich CLI with many available options. Below, we explain the most important parameters you should specify when running hephaestus.

hephaestus@e0456a9b520e:~$ hephaestus.py --help
usage: hephaestus.py [-h] [-s SECONDS] [-i ITERATIONS] [-t TRANSFORMATIONS] [--batch BATCH] [-b BUGS] [-n NAME] [-T [{TypeErasure} [{TypeErasure} ...]]]
                     [--transformation-schedule TRANSFORMATION_SCHEDULE] [-R REPLAY] [-e] [-k] [-S] [-w WORKERS] [-d] [-r] [-F LOG_FILE] [-L] [-N] [--language {kotlin,groovy,java}]
                     [--max-type-params MAX_TYPE_PARAMS] [--max-depth MAX_DEPTH] [-P] [--timeout TIMEOUT] [--cast-numbers] [--disable-use-site-variance] [--disable-contravariance-use-site]
                     [--disable-bounded-type-parameters] [--disable-parameterized-functions]

optional arguments:
  -h, --help            show this help message and exit
  -s SECONDS, --seconds SECONDS
                        Timeout in seconds
  -i ITERATIONS, --iterations ITERATIONS
                        Iterations to run (default: 3)
  -t TRANSFORMATIONS, --transformations TRANSFORMATIONS
                        Number of transformations in each round (default: 0)
  --batch BATCH         Number of programs to generate before invoking the compiler
  -b BUGS, --bugs BUGS  Set bug directory (default: /home/hephaestus/bugs)
  -n NAME, --name NAME  Set name of this testing instance (default: random string)
  -T [{TypeErasure} [{TypeErasure} ...]], --transformation-types [{TypeErasure} [{TypeErasure} ...]]
                        Select specific transformations to perform
  --transformation-schedule TRANSFORMATION_SCHEDULE
                        A file containing the schedule of transformations
  -R REPLAY, --replay REPLAY
                        Give a program to use instead of a randomly generated (pickled)
  -e, --examine         Open ipdb for a program (can be used only with --replay option)
  -k, --keep-all        Save all programs
  -S, --print-stacktrace
                        When an error occurs print stack trace
  -w WORKERS, --workers WORKERS
                        Number of workers for processing test programs
  -d, --debug
  -r, --rerun           Run only the last transformation. If failed, start from the last and go back until the transformation introduces the error
  -F LOG_FILE, --log-file LOG_FILE
                        Set log file (default: /home/hephaestus/logs)
  -L, --log             Keep logs for each transformation (bugs/session/logs)
  -N, --dry-run         Do not compile the programs
  --language {kotlin,groovy,java}
                        Select specific language
  --max-type-params MAX_TYPE_PARAMS
                        Maximum number of type parameters to generate
  --max-depth MAX_DEPTH
                        Generate programs up to the given depth
  -P, --only-correctness-preserving-transformations
                        Use only correctness-preserving transformations
  --timeout TIMEOUT     Timeout for transformations (in seconds)
  --cast-numbers        Cast numeric constants to their actual type (this option is used to avoid re-occrrence of a specific Groovy bug)
  --disable-use-site-variance
                        Disable use-site variance
  --disable-contravariance-use-site
                        Disable contravariance in use-site variance
  --disable-bounded-type-parameters
                        Disable bounded type parameters
  --disable-parameterized-functions
                        Disable parameterized functions

CLI Options

--bugs (Optional)

Set the directory to save the results of the testing session.

NOTE: The default directory is $(pwd)/bugs.

Example: --bugs hephaestus-results

--name (Optional)

Name of the current testing session.

NOTE: The default name is a randomly generated 5-character long string (e.g., hl43S).

Example: --name test-javac-1

--language

When running hephaestus, you should specify which language's the compiler you want to test. The available options are kotlin, groovy, and java. Hephaestus will use the selected language's compiler that is on the PATH. If you want to test a specific compiler version, you should configure it as the current session's default compiler.

Example: --language kotlin -- hephaestus will test the Kotlin compiler (i.e., kotlinc)

--seconds and --iterations

You should always specify either --seconds or --iterations option. The former specifies how much time hephaestus should test a compiler in seconds, whereas the second specifies how many test cases should hephaestus generate and run.

Example 1: --seconds 120 -- hephaestus will run for 2 minutes.

Example 2: --iterations 60 -- hephaestus will generate and run 60 test programs.

--batch (Optional)

When running hephaestus, most of the testing time is spent compiling the test programs. Instead of generating one program at a time, you can specify the number of programs you want to generate and the compiling all of them as a batch.

NOTE: The default option is 1.

Example: --batch 30 -- First, create 30 programs and then compile them with a single compiler execution.

--workers (Optional)

When --batch option is larger than one, you can specify the number of workers that will generate and mutate programs in parallel.

NOTE: The default option is 1.

Example: --workers 4 -- Use four workers to generate and mutate test programs.

--transformation-types and --only-correctness-preserving-transformations

Hephaestus supports two transformations, those that produce well-typed test programs and those that produce ill-typed test programs. Currently, hephaestus implements Type Erasure Mutator (TEM) and Type Overwriting Mutator (TOM). The former constructs well-typed programs, while the latter yields ill-typed.

By default, TOM is always running after generating a test program. To disable TOM, you should use the option --only-correctness-preserving-transformations. --transformation-types option specifies which mutations that produce well-typed programs should be used during a testing session. Currently, hephaestus implements only one mutator that produces well-typed programs (i.e., TEM).

Note: Although you can use the mutators in combination, they have not been rigorously tested in combination.

Example: --transformation-types TypeErasure -- enable TypeErasure mutation.

--transformations and --transformation-schedule

You should always specify one of those options. --transformations specify the number of mutations that should be applied per test program. If the value is 0, hephaestus will run only the generator. Note that this option only specifies how many correctness-preserving mutations should be applied. --transformation-schedule expects a path for a file containing the schedule of transformations. This file should specify a mutator per line.

Example 1: --transformations 0 -- Do not perform any transformations.

Example 2: --transformation-schedule transformations.txt -- Perform the transformations declared in file transformations.txt. The transformations.txt file could contain the following.

TypeErasure

--keep-all (Optional)

hephaestus only saves programs that result in compiler bugs. When --keep-all is enabled, hephaestus will save all generated and mutated test programs regardless of whether they trigger compiler bugs or not.

Example: --keep-all

--dry-run (Optional)

When this option is used, hephaestus only produces and mutates test programs, i.e., it does not invoke the compiler under test.

Example: --dry-run

--log-file (Optional)

By default, hephaestus keeps logs of a testing session in a file called logs, which resides in the current working directory. However, with --log-file option, you can specify another file to save the logs.

Example: --log-file my_logs

--replay (Optional)

Use a seed program written in hephaestus' IR, instead of invoking hephaestus' program generator.

Note: The input program should be pickled.

Example: --replay bugs/idET7/generator/iter_1/Main.java.bin

--debug (Debugging option)

Print debug messages before every step (i.e., program generation, mutation, compilation).

Note: Use this option only when --workers option is set to 1 and -batch is set to 1.

Example: --debug

--examine (Debugging option)

Open a debugger session to inspect the IR of the generated program.

Note: This option can only be used with --replay option.

--print-stacktrace (Debugging option)

Print stacktaces when encountering hephaestus internal errors.

Example: --print-stacktrace

--cast-numbers (Optional)

This option is used to cast numeric constants to their actual type in Groovy programs. We use this option to avoid the re-occurrence of a specific Groovy bug.

Note: This option has an effect only when --language is set to groovy.

Example: --cast-numbers

--disable-use-site-variance (Optional)

Generate programs that do not use use-site variance.

Example: --disable-use-site-variance

--disable-contravariance-use-site (Optional)

Generate programs that do not use contravariance in use-site variance.

Example: --disable-contravariance-use-site

--disable-parameterized-functions (Optional)

Generate programs that do not declare parameterized functions.

Example: --disable-parameterized-functions

--disable-bounded-type-parameters (Optional)

Generate programs that do declare type parameters with upper bounds.

Example: --disable-bounded-type-parameters

--max-type-params (Optional)

Specify the maximum number of type parameters for a parameterized class or a parameterized function.

Note: the default value is 3.

Example: --max-type-params 5

Run Tests

To run hephaestus tests you should execute the following commands:

# Enter hephaestus directory
hephaestus@e0456a9b520e:~$ cd hephaestus
# Run tests
hephaestus@e0456a9b520e:~/hephaestus$ python setup.py test

The output of the previous command should be similar to the following:

tests/test_call_analysis.py::test_program1 PASSED                       [  0%]
...
tests/test_use_analysis.py::test_program7 PASSED                        [100%]
tests/test_use_analysis.py::test_program8 PASSED                        [100%]
============================ 154 passed in 0.55s =============================

Example: Testing the Groovy compiler

NOTE: At each run, hephaestus generates random programs. Therefore, you should expect to get different results at each run: some randomly generated programs might trigger unfixed compiler bugs.

Here, we will test the Groovy compiler by employing hephaestus' program generator. Specifically, we will produce 30 test programs in batches of 10 test programs using two workers with the following command. Our testing session is named groovy-session.

hephaestus@e0456a9b520e:~/hephaestus$ hephaestus.py \
    --language groovy --transformations 0 \
    --batch 10 --iterations 30 --workers 2 -P \
    --name groovy-session

The expected outcome is something of the form:

stop_cond             iterations (30)
transformations       0
transformation_types  TypeErasure
bugs                  /home/hephaestus/hephaestus/bugs
name                  groovy-session
language              groovy
compiler              Groovy compiler version 4.0.0
===============================================================================
Test Programs Passed 30 / 30 ✔          Test Programs Failed 0 / 30 ✘

Two files are generated inside /home/hephaestus/bugs/groovy-session: stats.json and faults.json.

stats.json contains the following details about the testing session.

{
  "Info": {
    "stop_cond": "iterations",
    "stop_cond_value": 30,
    "transformations": 0,
    "transformation_types": "",
    "bugs": "/home/hephaestus/bugs",
    "name": "groovy-session",
    "language": "groovy",
    "compiler"  "Groovy compiler version 4.0.0"
  },
  "totals": {
    "passed": 30,
    "failed": 0
  }
}

If there were some bugs detected, faults.json would look like the following JSON file.

{

  "3": {
    "transformations": [],
    "error": " 72: The type ? is not a valid substitute for the bounded parameter <F extends java.lang.Double>\n @ line 72, column 99.\n    extends Double> gifting(Door<? super Do\n                                 ^\n 100: The type ? is not a valid substitute for the bounded parameter <F extends java.lang.Double>\n @ line 100, column 90.\n    extends Double> gifting(Door<? super Do\n                                 ^\n 141: The type ? is not a valid substitute for the bounded parameter <F extends java.lang.Double>\n @ line 141, column 90.\n    extends Double> gifting(Door<? super Do\n                                 ^\n 202: The type ? is not a valid substitute for the bounded parameter <F extends java.lang.Double>\n @ line 202, column 90.\n    extends Double> gifting(Door<? super Do\n                                 ^\n 228: The type ? is not a valid substitute for the bounded parameter <F extends java.lang.Double>\n @ line 228, column 90.\n    extends Double> gifting(Door<? super Do\n                                 ^\n 299: The type ? is not a valid substitute for the bounded parameter <F extends java.lang.Double>\n @ line 299, column 90.\n    extends Double> gifting(Door<? super Do\n                                 ^\n 376: The type ? is not a valid substitute for the bounded parameter <Y extends java.lang.Short>\n @ line 376, column 44.\n   spoons(Sicken issued, Plumage<? super Sh\n                                 ^\n 393: The type ? is not a valid substitute for the bounded parameter <F extends java.lang.Double>\n @ line 393, column 90.\n    extends Double> gifting(Door<? super Do\n                                 ^\n 510: The type ? is not a valid substitute for the bounded parameter <F extends java.lang.Double>\n @ line 510, column 90.\n    extends Double> gifting(Door<? super Do\n                                 ^",
    "programs": {
      "/tmp/tmpt_x_l1wk/src/kettles/Main.groovy": true
    }
  },
  "11": {
    "transformations": [
        "TypeOverwriting"
    ],
    "error": "SHOULD NOT BE COMPILED: X <: N expected but Imagine <: (Playing<Function1<Boolean(groovy-builtin), Float(groovy-builtin)>>) found in node global/Reconcile/reflexes/soybeans/cellos",
    "programs": {
      "/tmp/tmpmtyy6u6q/src/spanners/Main.groovy": true,
      "/tmp/tmpmtyy6u6q/src/franker/Main.groovy": false
    }
  },
  "1050": {
    "transformations": [],
    "error": ">>> a serious error occurred: BUG! exception in phase 'instruction selection' in source unit '/tmp/tmphj006wfu/src/wack/Main.groovy' unexpected NullPointerException\n>>> stacktrace:\nBUG! exception in phase 'instruction selection' in source unit '/tmp/tmphj006wfu/src/wack/Main.groovy' unexpected NullPointerException\n\tat org.codehaus.groovy.control.CompilationUnit$IPrimaryClassNodeOperation.doPhaseOperation(CompilationUnit.java:905)\n\tat org.codehaus.groovy.control.CompilationUnit.processPhaseOperations(CompilationUnit.java:654)\n\tat org.codehaus.groovy.control.CompilationUnit.compile(CompilationUnit.java:628)\n\tat org.codehaus.groovy.control.CompilationUnit.compile(CompilationUnit.java:609)\n\tat org.codehaus.groovy.tools.FileSystemCompiler.compile(FileSystemCompiler.java:311)\n\tat org.codehaus.groovy.tools.FileSystemCompiler.doCompilation(FileSystemCompiler.java:240)\n\tat org.codehaus.groovy.tools.FileSystemCompiler.commandLineCompile(FileSystemCompiler.java:165)\n\tat org.codehaus.groovy.tools.FileSystemCompiler.commandLineCompileWithErrorHandling(FileSystemCompiler.java:205)\n\tat org.codehaus.groovy.tools.FileSystemCompiler.main(FileSystemCompiler.java:189)\n\tat java.base/jdk.internal.reflect.NativeMethodAccessorImpl.invoke0(Native Method)\n\tat java.base/jdk.internal.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:77)\n\tat java.base/jdk.internal.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)\n\tat java.base/java.lang.reflect.Method.invoke(Method.java:568)\n\tat org.codehaus.groovy.tools.GroovyStarter.rootLoader(GroovyStarter.java:112)\n\tat org.codehaus.groovy.tools.GroovyStarter.main(GroovyStarter.java:130)\nCaused by: java.lang.NullPointerException: Cannot invoke \"org.codehaus.groovy.ast.stmt.Statement.visit(org.codehaus.groovy.ast.GroovyCodeVisitor)\" because the return value of \"org.codehaus.groovy.ast.MethodNode.getCode()\" is null\n\tat org.codehaus.groovy.transform.stc.StaticTypeCheckingVisitor.isTypeSource(StaticTypeCheckingVisitor.java:4189)\n\tat org.codehaus.groovy.transform.stc.StaticTypeCheckingVisitor.checkForTargetType(StaticTypeCheckingVisitor.java:4160)\n\tat org.codehaus.groovy.transform.stc.StaticTypeCheckingVisitor.visitTernaryExpression(StaticTypeCheckingVisitor.java:4136)\n\tat org.codehaus.groovy.ast.expr.TernaryExpression.visit(TernaryExpression.java:44)\n\tat org.codehaus.groovy.transform.stc.StaticTypeCheckingVisitor.visitMethodCallExpression(StaticTypeCheckingVisitor.java:3303)\n\tat org.codehaus.groovy.transform.sc.StaticCompilationVisitor.visitMethodCallExpression(StaticCompilationVisitor.java:421)\n\tat org.codehaus.groovy.ast.expr.MethodCallExpression.visit(MethodCallExpression.java:77)\n\tat org.codehaus.groovy.ast.CodeVisitorSupport.visitExpressionStatement(CodeVisitorSupport.java:117)\n\tat org.codehaus.groovy.ast.ClassCodeVisitorSupport.visitExpressionStatement(ClassCodeVisitorSupport.java:204)\n\tat org.codehaus.groovy.transform.stc.StaticTypeCheckingVisitor.visitExpressionStatement(StaticTypeCheckingVisitor.java:2188)\n\tat org.codehaus.groovy.ast.stmt.ExpressionStatement.visit(ExpressionStatement.java:41)\n\tat org.codehaus.groovy.ast.CodeVisitorSupport.visitBlockStatement(CodeVisitorSupport.java:86)\n\tat org.codehaus.groovy.ast.ClassCodeVisitorSupport.visitBlockStatement(ClassCodeVisitorSupport.java:168)\n\tat org.codehaus.groovy.transform.stc.StaticTypeCheckingVisitor.visitBlockStatement(StaticTypeCheckingVisitor.java:3895)\n\tat org.codehaus.groovy.ast.stmt.BlockStatement.visit(BlockStatement.java:70)\n\tat org.codehaus.groovy.ast.CodeVisitorSupport.visitBlockStatement(CodeVisitorSupport.java:86)\n\tat org.codehaus.groovy.ast.ClassCodeVisitorSupport.visitBlockStatement(ClassCodeVisitorSupport.java:168)\n\tat org.codehaus.groovy.transform.stc.StaticTypeCheckingVisitor.visitBlockStatement(StaticTypeCheckingVisitor.java:3895)\n\tat org.codehaus.groovy.ast.stmt.BlockStatement.visit(BlockStatement.java:70)\n\tat org.codehaus.groovy.ast.CodeVisitorSupport.visitClosureExpression(CodeVisitorSupport.java:239)\n\tat org.codehaus.groovy.transform.stc.StaticTypeCheckingVisitor.visitClosureExpression(StaticTypeCheckingVisitor.java:2402)\n\tat org.codehaus.groovy.ast.expr.ClosureExpression.visit(ClosureExpression.java:110)\n\tat org.codehaus.groovy.transform.stc.StaticTypeCheckingVisitor.visitCastExpression(StaticTypeCheckingVisitor.java:4074)\n\tat org.codehaus.groovy.ast.expr.CastExpression.visit(CastExpression.java:96)\n\tat org.codehaus.groovy.transform.stc.StaticTypeCheckingVisitor.visitInitialExpression(StaticTypeCheckingVisitor.java:1931)\n\tat org.codehaus.groovy.transform.stc.StaticTypeCheckingVisitor.visitDefaultParameterArguments(StaticTypeCheckingVisitor.java:2616)\n\tat org.codehaus.groovy.transform.stc.StaticTypeCheckingVisitor.visitConstructorOrMethod(StaticTypeCheckingVisitor.java:2588)\n\tat org.codehaus.groovy.ast.ClassCodeVisitorSupport.visitMethod(ClassCodeVisitorSupport.java:110)\n\tat org.codehaus.groovy.transform.stc.StaticTypeCheckingVisitor.startMethodInference(StaticTypeCheckingVisitor.java:2573)\n\tat org.codehaus.groovy.transform.stc.StaticTypeCheckingVisitor.visitMethod(StaticTypeCheckingVisitor.java:2552)\n\tat org.codehaus.groovy.transform.sc.StaticCompilationVisitor.visitConstructorOrMethod(StaticCompilationVisitor.java:236)\n\tat org.codehaus.groovy.transform.sc.StaticCompilationVisitor.visitMethod(StaticCompilationVisitor.java:251)\n\tat org.codehaus.groovy.ast.ClassNode.visitMethods(ClassNode.java:1135)\n\tat org.codehaus.groovy.ast.ClassNode.visitContents(ClassNode.java:1128)\n\tat org.codehaus.groovy.ast.ClassCodeVisitorSupport.visitClass(ClassCodeVisitorSupport.java:52)\n\tat org.codehaus.groovy.transform.stc.StaticTypeCheckingVisitor.visitClass(StaticTypeCheckingVisitor.java:437)\n\tat org.codehaus.groovy.transform.sc.StaticCompilationVisitor.visitClass(StaticCompilationVisitor.java:197)\n\tat org.codehaus.groovy.transform.sc.StaticCompileTransformation.visit(StaticCompileTransformation.java:68)\n\tat org.codehaus.groovy.control.customizers.ASTTransformationCustomizer.call(ASTTransformationCustomizer.groovy:298)\n\tat org.codehaus.groovy.control.CompilationUnit$IPrimaryClassNodeOperation.doPhaseOperation(CompilationUnit.java:900)\n\t... 14 more\n",
    "programs": {
      "/tmp/tmphj006wfu/src/yarn/Main.groovy": true
    }
  }
}

The first error is an unexpected compile-time error detected using our generator (GROOVY-10153). The second is a compiler bug where the compiler accepts an ill-typed program (GROOVY-10370). Finally, the third one is an internal error of groovyc (GROOVY-10357).

In the above scenario, the structure of testing session directory (i.e., bugs/groovy-session/) would be like the following

|-- 7
|   |-- Main.groovy
|   `-- Main.groovy.bin
|-- 11
|   |-- incorrect.groovy
|   |-- incorrect.groovy.bin
|   |-- Main.groovy
|   `-- Main.groovy.bin
|-- 29
|   |-- Main.groovy
|   `-- Main.groovy.bin
|-- faults.json
`-- stats.json

Now, you can exit the Docker container by running:

hephaestus@e0456a9b520e:~$ exit

Step By Step Instructions

NOTE: Remember to run all the subsequent docker run commands from the root directory of the artifact (i.e., hephaestus-pldi-eval/).

To validate the main results presented in the paper, first create a new Docker container by running:

docker run -ti --rm \
  -v $(pwd)/database:/home/hephaestus/database \
  -v $(pwd)/data:/home/hephaestus/data \
  -v $(pwd)/scripts:/home/hephaestus/eval-scripts \
  -v $(pwd)/figures:/home/hephaestus/eval-figures \
  hephaestus-eval

Note that we mount four local volumes inside the newly-created container. The first volume (database/) contains the bug database that includes the bugs discovered by our approach, while the second volume (data/) provides the data collected during our evaluation. The third volume (eval-scripts/) includes some scripts to reproduce and validate the results of the paper. Finally, the fourth volume (eval-figures/) will be used to save Figure 8 of our paper.

Bug Database

We provide an SQLite database (see the file database/bugdb.sqlite3) that contains information about the bugs discovered by Hephaestus during the evaluation. This database is initialized based on the SQL script stored into database/bug_schema.sql. The bug database consists of three tables, namely CompilerBug, Characteristic, and CompilerBugCharacteristics.

Each record of the CompilerBug table consists of the following columns.

  • bid: A serial number corresponding to the ID of the bug.
  • bug_id: The bug id as displayed in the corresponding issue tracker.
  • language: The name of the programming language of the test program.
  • compiler: The name of the compiler where the bug was found.
  • title: The title of the bug report.
  • issue_tracker_link: A URL pointing to the bug report opened by us.
  • mutator: The component that detected the bug. There are four possible values: generator, soundness (i.e., TOM), inference (i.e., TEM), and inference/soundness (i.e., TEM and then TOM).
  • fix_link: A URL pointing to the fix of the bug.
  • severity: The severity of the bugs given by the developers.
  • status: The status of the bug.
  • resolution: The resolution of the bug (e.g., Fixed, Duplicate).
  • report_date: The date that we reported the bug.
  • resolution_date: The date that the developers resolved the bug.
  • symptom: The symptom of the bug. There are three possible values: unexpected compile-time error (UCTE), unexpected runtime behavior (URB), and crash. Note that the URB symptom corresponds to soundness bugs detected by TOM.
  • resolved_in: How long did it take to resolve this bug.
  • test: The test program that revealed the bug.
  • error_msg: The error message reported by the compiler, or the stacktrace of the crash, or the exception caused in the runtime.

The Characteristic table contains the following three fields.

  • cid: A serial number corresponding to the ID of the characteristic.
  • characteristic_name: The name of the characteristic (e.g., Parameterized class).
  • category: The category of the characteristic (e.g. Parametric polymorphism).

Finally, CompilerBugCharacteristics is a table implementing the many-to-many relationship between CompilerBug and CompilerBugCharacteristics, this table contains three fields: bcid, cid, bid.

Example Queries

From inside the container, we can perform some basic queries on this bug database.

Get the total number of the discovered bugs.

hephaestus@e0456a9b520e:~$ sqlite3 database/bugdb.sqlite3 "SELECT COUNT(*) FROM CompilerBug";
156

Find the number of groovyc bugs.

hephaestus@e0456a9b520e:~$ sqlite3 database/bugdb.sqlite3 "SELECT COUNT(*) FROM CompilerBug WHERE compiler = 'groovyc'";
113

Find the number of javac bugs that have UCTE as their symptom.

hephaestus@e0456a9b520e:~$ sqlite3 database/bugdb.sqlite3 "SELECT COUNT(*) FROM CompilerBug WHERE compiler = 'javac' AND symptom = 'Unexpected Compile-Time Error'";
7

For each Kotlin bug revealed by TEM (i.e., type erasure mutation), dump the URLs pointing to our bug reports.

hephaestus@e0456a9b520e:~$ sqlite3 database/bugdb.sqlite3 "SELECT issue_tracker_link FROM CompilerBug WHERE compiler = 'kotlinc' AND mutator = 'inference'";
https://youtrack.jetbrains.com/issue/KT-49024
https://youtrack.jetbrains.com/issue/KT-49092
https://youtrack.jetbrains.com/issue/KT-45118
https://youtrack.jetbrains.com/issue/KT-47184
https://youtrack.jetbrains.com/issue/KT-48764
https://youtrack.jetbrains.com/issue/KT-43846
https://youtrack.jetbrains.com/issue/KT-44082
https://youtrack.jetbrains.com/issue/KT-44742
https://youtrack.jetbrains.com/issue/KT-44595
https://youtrack.jetbrains.com/issue/KT-44551
https://youtrack.jetbrains.com/issue/KT-46684
https://youtrack.jetbrains.com/issue/KT-44651

Get the three most common characteristics used in the test cases of the reported bugs.

hephaestus@e0456a9b520e:~$ sqlite3 database/bugdb.sqlite3 "SELECT c.characteristic_name, COUNT(*) as total FROM CompilerBugCharacteristics as cbc JOIN Characteristic as c ON c.cid = cbc.cid GROUP BY cbc.cid ORDER BY total DESC LIMIT 3";
Parameterized class|98
Parameterized type|78
Bounded type parameter|51

RQ1: Bug-Finding Results (Section 4.2)

For the first research question, first, we will use database/bugs.json to reproduce Figure 7a, which shows how many bugs (and their status) were found in each tested compiler. To do so, run:

hephaestus@e0456a9b520e:~$ python eval-scripts/process_bugs.py database/bugs.json rq1
                         Figure 7a
============================================================
Status              groovyc   kotlinc   Java      Total
------------------------------------------------------------
Reported            0         3         0         3
Confirmed           34        14        3         51
Fixed               74        10        2         86
Wont fix            2         2         5         9
Duplicate           3         3         1         7
------------------------------------------------------------
Total               113       32        11        156

Next, run the following script to produce Figure 8 and compute the numbers of "Affected compiler versions" paragraph.

hephaestus@e0456a9b520e:~$ python eval-scripts/history_analysis.py \
    data/history/history_19_11_21.json eval-figures/bug_versions.pdf
groovyc
All versions are buggy: 35
The error exist only in master: 50
Regressions: 25

kotlinc
All versions are buggy: 14
The error exist only in master: 5
Regressions: 13

javac
All versions are buggy: 6
The error exist only in master: 2
Regressions: 3

   Compiler Affected stable versions  Bugs
0   kotlinc                    [1-3]     5
1   kotlinc                     > 12     1
2   kotlinc                    [7-9]     7
3   kotlinc                      All    14
4   kotlinc              master only     5
5   groovyc                    [1-3]    15
6   groovyc              master only    50
7   groovyc                      All    35
8   groovyc                    [4-6]     1
9   groovyc                     > 12     3
10  groovyc                  [10-12]     6
11    javac              master only     2
12    javac                      All     6
13    javac                    [1-3]     2
14    javac                    [4-6]     1

This script will also generate Figure 8 and save it at figures/bug_versions.pdf in your host machine.

The statements that we want to check from paragraph "Affected compiler versions" are the following:

  • "35 groovyc and 14 kotlinc bugs occur in all stable compiler versions".
  • "A large portion of groovyc bugs (50/110 -- 45%) are triggered only in the master branch of the compiler".

Re-run the "Affected Compiler Versions" experiment. (Optional)

To re-compute which compiler versions are affected, you can run the following command (it will take around 90 minutes):

python eval-scripts/history_run.py database/bugs.json history.json

NOTE: The results might be slightly different because (1) some compiler version might be no longer supported by SDKMAN, (2) the developers may have fixed some bugs.

RQ2: Bug and Test Case Characteristics (Section 4.3)

For the second research question, we will use database/bugs.json to reproduce Figure 7b, which shows the symptoms of the reported bugs. Furthermore, we will find the most popular features used in the bug-revealing test programs. To do so, run:

hephaestus@e0456a9b520e:~$ python eval-scripts/process_bugs.py database/bugs.json rq2
                         Figure 7b
============================================================
Symptoms            groovyc   kotlinc   Java      Total
------------------------------------------------------------
UCTE                80        17        7         104
URB                 19        3         0         22
Crash               14        12        4         30



================================================================================
Characteristics
--------------------------------------------------------------------------------
Parameterized class          98   Parametric polymorphism
Parameterized type           78   Parametric polymorphism
Bounded type parameter       51   Parametric polymorphism
Type argument inference      36   Type inference
Lambda                       35   Functional programming
Conditionals                 32   Standard language features
Inheritance                  31   OOP features
Subtyping                    30   Type system-related features
Function type                27   Functional programming
Variable type inference      25   Type inference
Parameterized function       24   Parametric polymorphism
Use-site variance            23   Parametric polymorphism
Flow typing                  13   Type inference
Array                        11   Standard language features
Primitive type               10   Type system-related features
Function reference           9    Functional programming
Overriding                   8    OOP features
SAM type                     5    Functional programming
Variable arguments           4    Standard language features
Return type inference        4    Type inference
Declaration-site variance    2    Parametric polymorphism
Named arguments              1    Other
Cast                         1    Standard language features
================================================================================

================================================================================
Categories
--------------------------------------------------------------------------------
Parametric polymorphism      106
Type inference               63
Standard language features   44
Functional programming       39
Type system-related features 38
OOP features                 31
Other                        1
================================================================================

Beyond Figure 7b, this script produces the numbers mentioned in Section 4.3. Specifically, Section 4.3 contains the following statements:

  • "Features related to parametric polymorphism (e.g., parameterized class) are in the list of features with the most bug-revealing capability".
  • "In total, 106/156 bugs are caused by programs containing at least one such feature (parametric polymorphism)".
  • "In 47% of test cases that use conditionals, type inference features are also included".

To verify the last statement, you can use the option --combinations and see which features are used with conditionals (i.e., 15 out of 32 programs that use conditionals, they also involve type inference).

hephaestus@e0456a9b520e:~$ python eval-scripts/process_bugs.py --combinations \
    database/bugs.json rq2
Combinations
================================================================================
SAM type                      Functional programming            5
SAM type                      Parametric polymorphism           3
SAM type                      Type system-related features      1
Function type                 Functional programming           26
Function type                 Type system-related features      6
Function type                 Type inference                    7
Function type                 Parametric polymorphism          10
Function type                 Standard language features        5
Function type                 OOP features                      3
Function reference            Functional programming            9
Function reference            Parametric polymorphism           4
Function reference            Type system-related features      1
Function reference            Type inference                    1
Function reference            Standard language features        1
Function reference            OOP features                      1
Lambda                        Functional programming           25
Lambda                        Parametric polymorphism          14
Lambda                        Type system-related features      7
Lambda                        Type inference                   16
Lambda                        Standard language features        6
Lambda                        OOP features                      2
Parameterized class           Parametric polymorphism          95
Parameterized class           Type inference                   43
Parameterized class           Type system-related features     18
Parameterized class           Functional programming           13
Parameterized class           Standard language features       23
Parameterized class           OOP features                     24
Parameterized type            Parametric polymorphism          78
Parameterized type            Type inference                   35
Parameterized type            Type system-related features     17
Parameterized type            Standard language features       22
Parameterized type            OOP features                     23
Parameterized type            Functional programming            6
Bounded type parameter        Parametric polymorphism          51
Bounded type parameter        Type inference                   21
Bounded type parameter        Type system-related features      9
Bounded type parameter        OOP features                     12
Bounded type parameter        Standard language features       13
Bounded type parameter        Functional programming            6
Declaration-site variance     Parametric polymorphism           2
Declaration-site variance     Type inference                    1
Declaration-site variance     OOP features                      1
Variable type inference       Parametric polymorphism          15
Variable type inference       Standard language features        6
Variable type inference       Functional programming           10
Variable type inference       Type inference                   13
Variable type inference       Type system-related features      3
Variable type inference       OOP features                      3
Use-site variance             Parametric polymorphism          22
Use-site variance             Type inference                    7
Use-site variance             Type system-related features      5
Use-site variance             Standard language features        3
Use-site variance             OOP features                      6
Use-site variance             Functional programming            1
Parameterized function        Functional programming            4
Parameterized function        Parametric polymorphism          19
Parameterized function        Type system-related features      3
Parameterized function        Type inference                   10
Parameterized function        Standard language features        6
Parameterized function        OOP features                      2
Type argument inference       Parametric polymorphism          36
Type argument inference       Type system-related features      7
Type argument inference       Type inference                    9
Type argument inference       OOP features                      7
Type argument inference       Standard language features        9
Type argument inference       Functional programming            4
Subtyping                     Parametric polymorphism          17
Subtyping                     Type inference                   14
Subtyping                     OOP features                     16
Subtyping                     Standard language features       14
Subtyping                     Functional programming            7
Subtyping                     Type system-related features      2
Array                         Standard language features        4
Array                         Type inference                    3
Array                         Type system-related features      5
Array                         OOP features                      2
Array                         Parametric polymorphism           4
Array                         Functional programming            2
Conditionals                  Standard language features        2
Conditionals                  Type inference                   15
Conditionals                  OOP features                      9
Conditionals                  Type system-related features     13
Conditionals                  Parametric polymorphism          18
Conditionals                  Functional programming            5
Return type inference         OOP features                      3
Return type inference         Type system-related features      3
Return type inference         Standard language features        2
Return type inference         Functional programming            1
Return type inference         Parametric polymorphism           2
Return type inference         Type inference                    1
Inheritance                   Type system-related features     18
Inheritance                   Type inference                   13
Inheritance                   Standard language features       11
Inheritance                   Parametric polymorphism          24
Inheritance                   OOP features                      8
Inheritance                   Functional programming            3
Flow typing                   Functional programming            6
Flow typing                   Type inference                    7
Flow typing                   Standard language features        4
Flow typing                   OOP features                      4
Flow typing                   Type system-related features      3
Flow typing                   Parametric polymorphism           1
Overriding                    Standard language features        4
Overriding                    Type system-related features      4
Overriding                    Type inference                    2
Overriding                    Parametric polymorphism           8
Overriding                    OOP features                      8
Variable arguments            Parametric polymorphism           4
Variable arguments            OOP features                      1
Variable arguments            Type inference                    2
Variable arguments            Functional programming            1
Variable arguments            Standard language features        1
Primitive type                Parametric polymorphism           2
Primitive type                OOP features                      2
Primitive type                Standard language features        3
Primitive type                Functional programming            2
Primitive type                Type system-related features      2
Cast                          Standard language features        1

RQ3: Effectiveness of Mutations (Section 4.4)

For the third research question, we will first use database/bugs.json to reproduce Figure 7c, which shows how many bugs each component revealed. To do so, run:

hephaestus@e0456a9b520e:~$ python eval-scripts/process_bugs.py database/bugs.json rq3
                         Figure 7c
============================================================
Component           groovyc   kotlinc   Java      Total
------------------------------------------------------------
Generator           55        16        7         78
TEM                 37        12        3         52
TOM                 20        3         1         24
TEM & TOM           1         1         0         2

Next, we are going to compute the results of Figure 9 per compiler. In the data/coverage/mutations/ directory, we provide the results of JaCoCo on 5k random test programs per compiler saved at data/test_programs/mutations.

Impact of mutations on groovyc

For computing the impact of TEM on code coverage, run

hephaestus@e0456a9b520e:~$ python eval-scripts/compute_coverage.py \
    g data/coverage/mutations/groovy/groovy-generator-inf.csv \
    data/coverage/mutations/groovy/groovy-combination-inf.csv \
    data/coverage/mutations/groovy/groovy_whitelist --increasepkg

This script will produce the following output:

                          Line Coverage  Function Coverage    Branch Coverage
Initial                           42.68              41.77              42.07
Combination                       43.14              42.14              42.52
% change                           0.46               0.37               0.45
Absolute change                     167                 27                752

org.codehaus.groovy.control.messages: Branch -- 44 (66.67), Line -- 12 (70.59), Func -- 3 (75.00)
org.codehaus.groovy.transform.stc: Branch -- 531 (4.58), Line -- 106 (4.25), Func -- 13 (3.58)
org.codehaus.groovy.transform: Branch -- 541 (4.05), Line -- 107 (3.69), Func -- 13 (3.04)
org.codehaus.groovy: Branch -- 722 (1.38), Line -- 158 (1.31), Func -- 26 (1.07)
org.codehaus: Branch -- 722 (1.38), Line -- 158 (1.31), Func -- 26 (1.07)
total: Branch -- 752 (1.07), Line -- 167 (1.09), Func -- 27 (0.87)
org: Branch -- 752 (1.07), Line -- 167 (1.09), Func -- 27 (0.87)
org.codehaus.groovy.control: Branch -- 76 (1.03), Line -- 22 (1.29), Func -- 6 (1.92)
org.codehaus.groovy.classgen.asm: Branch -- 48 (0.57), Line -- 15 (0.76), Func -- 3 (0.80)
org.codehaus.groovy.transform.sc: Branch -- 10 (0.57), Line -- 1 (0.25), Func -- 0 (0.00)
org.codehaus.groovy.ast: Branch -- 57 (0.54), Line -- 14 (0.56), Func -- 4 (0.50)
org.codehaus.groovy.classgen: Branch -- 48 (0.27), Line -- 15 (0.37), Func -- 3 (0.43)
org.codehaus.groovy.transform.sc.transformers: Branch -- 2 (0.22), Line -- 0 (0.00), Func -- 0 (0.00)
org.apache.groovy.parser.antlr4: Branch -- 30 (0.17), Line -- 9 (0.27), Func -- 1 (0.15)
org.apache.groovy.parser: Branch -- 30 (0.17), Line -- 9 (0.27), Func -- 1 (0.15)
org.apache.groovy: Branch -- 30 (0.17), Line -- 9 (0.27), Func -- 1 (0.15)
org.apache: Branch -- 30 (0.17), Line -- 9 (0.27), Func -- 1 (0.15)
org.codehaus.groovy.ast.expr: Branch -- 3 (0.16), Line -- 1 (0.19), Func -- 1 (0.49)

In the above output, please notice line:

org.codehaus.groovy.transform.stc: Branch -- 531 (4.58), Line -- 106 (4.25), Func -- 13 (3.58)

This line presents the absolute and percentage increase (shown inside parentheses) of branch, line, and function coverage caused by TEM on package org.codehaus.groovy.transform.stc.*.

Now, for computing the impact of TOM on code coverage, run:

hephaestus@e0456a9b520e:~$ python eval-scripts/compute_coverage.py \
    g data/coverage/mutations/groovy/groovy-generator-sound.csv \
    data/coverage/mutations/groovy/groovy-combination-sound.csv \
    data/coverage/mutations/groovy/groovy_whitelist
                          Line Coverage  Function Coverage    Branch Coverage
Initial                           43.30              42.21              42.72
Combination                       43.57              42.35              42.99
% change                           0.27               0.14               0.27
Absolute change                      99                 10                447

Impact of mutations on kotlinc

For computing the impact of TEM on code coverage, run:

hephaestus@e0456a9b520e:~$ python eval-scripts/compute_coverage.py k \
    data/coverage/mutations/kotlin/kotlin-generator-inf.csv \
    data/coverage/mutations/kotlin/kotlin-combination-inf.csv \
    data/coverage/mutations/kotlin/kotlin_whitelist --increasepkg

The above script will produce the following output:

                          Line Coverage  Function Coverage    Branch Coverage
Initial                           30.92              30.60              30.32
Combination                       31.38              31.00              30.78
% change                           0.46               0.39               0.46
Absolute change                     787                217               5431

org.jetbrains.kotlin.resolve.calls.inference.constraintPosition: Branch -- 83 (73.45), Line -- 6 (54.55), Func -- 5 (100.00)
org.jetbrains.kotlin.resolve.typeBinding: Branch -- 136 (60.44), Line -- 16 (50.00), Func -- 6 (54.55)
org.jetbrains.kotlin.resolve.calls.inference: Branch -- 1865 (20.09), Line -- 238 (17.75), Func -- 63 (14.89)
org.jetbrains.kotlin.resolve.calls.tower: Branch -- 938 (10.98), Line -- 137 (10.69), Func -- 36 (10.98)
org.jetbrains.kotlin.resolve.calls.inference.model: Branch -- 202 (9.36), Line -- 23 (6.97), Func -- 10 (7.35)
org.jetbrains.kotlin.resolve.calls.inference.components: Branch -- 466 (8.50), Line -- 63 (7.91), Func -- 5 (2.50)
org.jetbrains.kotlin.resolve.calls: Branch -- 3402 (8.38), Line -- 476 (8.11), Func -- 119 (6.94)
org.jetbrains.kotlin.psi.psiUtil: Branch -- 59 (8.35), Line -- 10 (12.66), Func -- 2 (4.26)
org.jetbrains.kotlin.types: Branch -- 957 (4.32), Line -- 147 (4.47), Func -- 69 (6.48)
org.jetbrains.kotlin.resolve: Branch -- 4086 (4.17), Line -- 572 (3.93), Func -- 135 (3.33)
org.jetbrains.kotlin.resolve.calls.tasks: Branch -- 14 (4.11), Line -- 3 (5.66), Func -- 1 (3.85)
org.jetbrains.kotlin.resolve.constants.evaluate: Branch -- 125 (3.86), Line -- 22 (4.33), Func -- 0 (0.00)
org.jetbrains.kotlin.resolve.calls.components: Branch -- 218 (3.53), Line -- 32 (3.88), Func -- 6 (2.27)
org.jetbrains.kotlin.resolve.calls.model: Branch -- 80 (3.18), Line -- 11 (2.70), Func -- 9 (5.49)
org.jetbrains.kotlin.types.checker: Branch -- 127 (2.95), Line -- 16 (3.54), Func -- 7 (2.73)
org.jetbrains.kotlin.resolve.constants: Branch -- 138 (2.89), Line -- 23 (3.17), Func -- 1 (0.57)
org.jetbrains.kotlin.descriptors.annotations: Branch -- 33 (2.83), Line -- 6 (4.23), Func -- 2 (3.77)
org.jetbrains.kotlin.types.expressions: Branch -- 164 (2.28), Line -- 28 (2.16), Func -- 6 (2.18)
org.jetbrains.kotlin.resolve.descriptorUtil: Branch -- 15 (2.24), Line -- 1 (1.32), Func -- 1 (2.04)
org.jetbrains.kotlin.resolve.calls.util: Branch -- 10 (1.90), Line -- 1 (1.06), Func -- 1 (1.72)
org.jetbrains.kotlin.resolve.calls.smartcasts: Branch -- 46 (1.64), Line -- 5 (1.20), Func -- 1 (0.88)
total: Branch -- 5431 (1.52), Line -- 787 (1.48), Func -- 217 (1.29)
org.jetbrains.kotlin: Branch -- 5431 (1.52), Line -- 787 (1.48), Func -- 217 (1.29)
org.jetbrains: Branch -- 5431 (1.52), Line -- 787 (1.48), Func -- 217 (1.29)
org: Branch -- 5431 (1.52), Line -- 787 (1.48), Func -- 217 (1.29)
org.jetbrains.kotlin.psi: Branch -- 104 (1.43), Line -- 17 (1.27), Func -- 6 (0.90)
org.jetbrains.kotlin.diagnostics: Branch -- 92 (1.04), Line -- 15 (0.73), Func -- 4 (1.74)
org.jetbrains.kotlin.cli.jvm.compiler: Branch -- 40 (0.76), Line -- 7 (0.90), Func -- 1 (0.70)
org.jetbrains.kotlin.cfg: Branch -- 74 (0.64), Line -- 13 (0.72), Func -- 0 (0.00)
org.jetbrains.kotlin.parsing: Branch -- 31 (0.52), Line -- 9 (0.77), Func -- 0 (0.00)
org.jetbrains.kotlin.cli.jvm: Branch -- 44 (0.49), Line -- 8 (0.63), Func -- 1 (0.36)
org.jetbrains.kotlin.resolve.jvm: Branch -- 35 (0.44), Line -- 4 (0.31), Func -- 1 (0.33)
org.jetbrains.kotlin.cli.common.messages: Branch -- 5 (0.40), Line -- 0 (0.00), Func -- 0 (0.00)
org.jetbrains.kotlin.descriptors: Branch -- 33 (0.39), Line -- 6 (0.45), Func -- 2 (0.34)
org.jetbrains.kotlin.cli: Branch -- 49 (0.27), Line -- 8 (0.38), Func -- 1 (0.16)
org.jetbrains.kotlin.resolve.calls.checkers: Branch -- 6 (0.20), Line -- 0 (0.00), Func -- 0 (0.00)
org.jetbrains.kotlin.psi2ir.transformations: Branch -- 2 (0.14), Line -- 0 (0.00), Func -- 0 (0.00)
org.jetbrains.kotlin.resolve.calls.results: Branch -- 2 (0.13), Line -- 1 (0.40), Func -- 0 (0.00)
org.jetbrains.kotlin.diagnostics.rendering: Branch -- 5 (0.11), Line -- 1 (0.11), Func -- 1 (1.30)
org.jetbrains.kotlin.renderer: Branch -- 3 (0.07), Line -- 0 (0.00), Func -- 0 (0.00)
org.jetbrains.kotlin.resolve.jvm.checkers: Branch -- 2 (0.06), Line -- 0 (0.00), Func -- 0 (0.00)
org.jetbrains.kotlin.cli.common: Branch -- 5 (0.06), Line -- 0 (0.00), Func -- 0 (0.00)
org.jetbrains.kotlin.resolve.checkers: Branch -- 2 (0.05), Line -- 0 (0.00), Func -- 0 (0.00)
org.jetbrains.kotlin.psi2ir: Branch -- 2 (0.02), Line -- 0 (0.00), Func -- 0 (0.00)

In the above output, please notice the following three lines:

org.jetbrains.kotlin.resolve.calls.inference: Branch -- 1865 (20.09), Line -- 238 (17.75), Func -- 63 (14.89)
org.jetbrains.kotlin.resolve: Branch -- 4086 (4.17), Line -- 572 (3.93), Func -- 135 (3.33)
org.jetbrains.kotlin.types: Branch -- 957 (4.32), Line -- 147 (4.47), Func -- 69 (6.48)

These lines present the absolute and percentage increase (shown inside parentheses) of branch, line, and function coverage caused by TEM on packages org.jetbrains.kotlin.resolve.calls.inference.*, org.jetbrains.kotlin.resolve.*, and org.jetbrains.kotlin.types.* respectively.

For computing the impact of TOM on code coverage, run:

hephaestus@e0456a9b520e:~$ python eval-scripts/compute_coverage.py \
    g data/coverage/mutations/kotlin/kotlin-generator-sound.csv \
    data/coverage/mutations/kotlin/kotlin-combination-sound.csv \
    data/coverage/mutations/kotlin/kotlin_whitelist

                          Line Coverage  Function Coverage    Branch Coverage
Initial                           31.47              31.01              30.87
Combination                       31.80              31.31              31.22
% change                           0.33               0.30               0.35
Absolute change                     572                166               4171

Impact of mutations on javac

For computing the impact of TEM on code coverage, run:

hephaestus@e0456a9b520e:~$ python eval-scripts/compute_coverage.py \
    g data/coverage/mutations/java/java-generator-inf.csv \
    data/coverage/mutations/java/java-combination-inf.csv \
    data/coverage/mutations/java/java_whitelist --increasepkg

The above command will produce the following output:

                          Line Coverage  Function Coverage    Branch Coverage
Initial                           36.99              39.68              34.56
Combination                       37.68              40.49              35.18
% change                           0.68               0.81               0.62
Absolute change                     396                 87               2150

com.sun.tools.javac.code: Branch -- 636 (3.27), Line -- 131 (3.31), Func -- 31 (3.16)
com.sun.tools.javac.comp: Branch -- 1200 (3.06), Line -- 204 (2.66), Func -- 47 (3.59)
total: Branch -- 2150 (1.79), Line -- 396 (1.85), Func -- 87 (2.05)
com.sun.tools.javac.util: Branch -- 129 (1.60), Line -- 29 (1.60), Func -- 5 (1.10)
com.sun.tools.javac.parser: Branch -- 59 (0.86), Line -- 14 (0.93), Func -- 0 (0.00)
com.sun.tools.javac.resources: Branch -- 109 (0.62), Line -- 13 (3.20), Func -- 2 (10.00)
com.sun.tools.javac.tree: Branch -- 12 (0.23), Line -- 4 (0.33), Func -- 2 (0.53)
com.sun.tools.javac.main: Branch -- 5 (0.11), Line -- 1 (0.12), Func -- 0 (0.00)

In the above output, please notice the following two lines:

com.sun.tools.javac.code: Branch -- 636 (3.27), Line -- 131 (3.31), Func -- 31 (3.16)
com.sun.tools.javac.comp: Branch -- 1200 (3.06), Line -- 204 (2.66), Func -- 47 (3.59)

These lines present the absolute and percentage increase (shown inside parentheses) of branch, line, and function coverage caused by TEM on packages com.sun.tools.javac.code.*, and com.sun.tools.javac.comp.* respectively.

To see, the impact of TEM on the code coverage of classes instead of packages, please run:

hephaestus@e0456a9b520e:~$ python eval-scripts/compute_coverage.py \
    g data/coverage/mutations/java/java-generator-inf.csv \
    data/coverage/mutations/java/java-combination-inf.csv \
    data/coverage/mutations/java/java_whitelist --increasecls

Now notice, the following lines, which presented in Figure 9.

com.sun.tools.javac.comp,Resolve: Branch -- 613 (15.67), Line -- 100 (14.12), Func -- 27 (17.20)
com.sun.tools.javac.code,Types: Branch -- 558 (7.54), Line -- 113 (8.11), Func -- 23 (7.67)

Finally, for computing the impact of TOM on javac's code coverage, run

hephaestus@e0456a9b520e:~$ python eval-scripts/compute_coverage.py \
    g data/coverage/mutations/java/java-generator-sound.csv \
    data/coverage/mutations/java/java-combination-sound.csv \
    data/coverage/mutations/java/java_whitelist

                          Line Coverage  Function Coverage    Branch Coverage
Initial                           37.63              40.46              35.18
Combination                       38.26              41.19              35.76
% change                           0.62               0.74               0.57
Absolute change                     362                 79               1990

Reproducing RQ3's Coverage Experiment (Optional)

Depending on your machine(s), re-running the complete experiment could take up to 5 days. Here, we will provide a complete example of how you get the coverage of 10 Java test programs for both TEM and TOM, using JaCoCo. To reproduce the full results, you should (1) produce 5k programs (instead of 10 programs) for both TOM and TEM, and (2) run the same experiments for the other compilers by replacing java with groovy and kotlin in the following commands.

Step 1: Generate the test programs.

The following command uses Hephaestus to generate 10 programs written in Java using our program generator. This command also produces one variant for each of those 10 programs using TEM. Note that this command does not invoke the compiler under test (see --dry-run option).

hephaestus@e0456a9b520e:~$ hephaestus.py --bugs coverage_programs \
    --name java_tem_10 --language java \
    --iterations 10 --batch 10 --workers 2 --transformations 1 \
    --keep-all --dry-run -P

The following command uses Hephaestus to generate 10 programs written in Java using our program generator. This command also produces one variant for each of those 10 programs using TOM. Note that this command does not invoke the compiler under test (see --dry-run option).

hephaestus@e0456a9b520e:~$ hephaestus.py --bugs coverage_programs \
    --name java_tom_10 --language java \
    --iterations 10 --batch 10 --workers 2 --transformations 0 \
    --keep-all --dry-run

Step 2: Produce the coverage reports for generator and TEM.

This command takes the programs generated in Step 1 and compiles them using javac. This script first compiles the programs produced by our program generator, and tracks the code coverage of javac. After that, this script takes the TEM variants, and compiles them using the instrumented version of javac.

hephaestus@e0456a9b520e:~$ ./eval-scripts/coverage/java_mutations.sh \
    $HOME/coverage \
    $HOME/coverage_programs/java_tem_10/ 10 inference 2> /dev/null

Step 3: Compute the results for generator and TEM.

hephaestus@e0456a9b520e:~$ python eval-scripts/compute_coverage.py g \
    results/java/java-generator-inference.csv \
    results/java/java-comb-inference.csv \
    data/coverage/mutations/java/java_whitelist
                          Line Coverage  Function Coverage    Branch Coverage
Initial                           35.12              38.10              32.31
Combination                       35.55              38.66              32.74
% change                           0.43               0.56               0.43
Absolute change                     250                 60               1511

Step 4: Produce the coverage reports for generator and TOM.

This command takes the programs generated in Step 1 and compiles them using javac. This script first compiles the programs produced by our program generator, and tracks the code coverage of javac. After that, this script takes the TOM variants, and compiles them using the instrumented version of javac.

hephaestus@e0456a9b520e:~$ ./eval-scripts/coverage/java_mutations.sh \
    $HOME/coverage \
    $HOME/coverage_programs/java_tom_10/ 10 2> /dev/null

Step 5: Compute the results for generator and TOM.

NOTE: You are expected to get slightly different results as the generated programs are different than those used in the paper.

hephaestus@e0456a9b520e:~$ python eval-scripts/compute_coverage.py g \
    results/java/java-generator-soundness.csv \
    results/java/java-comb-soundness.csv \
    data/coverage/mutations/java/java_whitelist

                          Line Coverage  Function Coverage    Branch Coverage
Initial                           36.18              39.23              33.23
Combination                       36.64              39.86              34.19
% change                           0.45               0.63               0.96
Absolute change                     263                 68               3327

RQ4: Code Coverage (Section 4.5)

For the fourth research question, we will use coverage data from data/coverage/compilers/ to reproduce Figure 10 that shows the code coverage improvement when adding 10k programs produced by Hephaestus to the test suites of each compiler. In the following, we compute the results per compiler.

Impact on groovyc

hephaestus@e0456a9b520e:~$ python eval-scripts/compute_coverage.py \
    g data/coverage/compilers/groovy/groovy-vanilla.csv \
    data/coverage/compilers/groovy/groovy-hephaestus.csv \
    data/coverage/compilers/groovy/groovy_whitelist --compiler

                          Line Coverage  Function Coverage    Branch Coverage
Initial                           82.00              71.77              78.38
Combination                       82.06              71.79              78.44
% change                           0.06               0.02               0.05

Impact on kotlinc

hephaestus@e0456a9b520e:~$ python eval-scripts/compute_coverage.py \
    g data/coverage/compilers/kotlin/kotlin-vanilla.csv \
    data/coverage/compilers/kotlin/kotlin-hephaestus.csv \
    data/coverage/compilers/kotlin/kotlin_whitelist --compiler

                          Line Coverage  Function Coverage    Branch Coverage
Initial                           80.80              72.99              74.08
Combination                       80.83              73.05              74.11
% change                           0.03               0.06               0.04

Note: Our results are slightly different from the submitted paper. We will update Figure 10 on the camera-ready paper to match the results of our artifact.

Impact on javac

hephaestus@e0456a9b520e:~$ python eval-scripts/compute_coverage.py \
    g data/coverage/compilers/java/java-vanilla.csv \
    data/coverage/compilers/java/java-hephaestus.csv \
    data/coverage/compilers/java/java_whitelist --compiler

                          Line Coverage  Function Coverage    Branch Coverage
Initial                           83.76              83.95              83.90
Combination                       83.94              83.99              84.12
% change                           0.18               0.03               0.22

Reproducing RQ4's Coverage Experiment (Optional)

Similar to RQ3, re-running the full experiments for this RQ might take days. These experiments require much time because you have to run the test-suites of the compilers to produce their code coverage. Note that if you run the experiments with a small number of generated test programs, the code coverage increase should be minimal.

Java, Kotlin

javac does not provide any script to generate code coverage metrics for its test-suite. Hence, we are going to compile every program in the test-suite using an instrumented version of javac to get the code coverage metrics of javac's test suite.

To compute the code coverage metrics of javac's test suite, run (estimated running time: 4 hours):

# This will generate a code coverage report for the test-suite in
# results/java_test_suite/java-test-suite.csv.
hephaestus@e0456a9b520e:~$ ./eval-scripts/coverage/java_test_suite.sh

To generate Java programs using Hephaestus, run the following commands. Note that these commands generate 10 Java programs using TEM, and 10 Java programs using TOM.

# You can skip this step if you have already generated some programs
# and their reports.
hephaestus@e0456a9b520e:~$ hephaestus.py --bugs coverage_programs \
    --name java_tem_10 --language java \
    --iterations 10 --batch 10 --workers 2 --transformations 1 \
    --keep-all --dry-run -P
hephaestus@e0456a9b520e:~$ hephaestus.py --bugs coverage_programs \
    --name java_tom_10 --language java \
    --iterations 10 --batch 10 --workers 2 --transformations 0 \
    --keep-all --dry-run

Compute code coverage metrics using the Hephaestus programs generated previously:

hephaestus@e0456a9b520e:~$ ./eval-scripts/coverage/java_mutations.sh \
    $HOME/coverage \
    $HOME/coverage_programs/java_tom_10/ 10 2> /dev/null
hephaestus@e0456a9b520e:~$ ./eval-scripts/coverage/java_mutations.sh \
    $HOME/coverage \
    $HOME/coverage_programs/java_tem_10/ 10 inference 2> /dev/null

Combine the code coverage reports

hephaestus@e0456a9b520e:~$ ./eval-scripts/coverage/combine.sh java \
    results/java/jacoco-comb-inference.exec \
    results/java/jacoco-comb-soundness.exec \
    results/java java-comb
hephaestus@e0456a9b520e:~$ ./eval-scripts/coverage/combine.sh java \
    results/java-test-suite/jacoco.exec \
    results/java/java-comb.exec \
    results/java java-hephaestus

Print results

hephaestus@e0456a9b520e:~$ python ~/eval-scripts/compute_coverage.py g \
    results/java-test-suite/java-test-suite.csv \
    results/java/java-hephaestus.csv \
    ~/data/coverage/compilers/java/java_whitelist
                          Line Coverage  Function Coverage    Branch Coverage
Initial                           82.84              83.09              83.07
Combination                       82.89              83.09              83.13
% change                           0.06               0.00               0.06
Absolute change                      13                  0                 75

Similar to javac, kotlinc does not provide any script to generate code coverage metrics for its test-suite. Hence, we are going to compile every program in the test-suite with an instrumented version of kotlinc to compute the code coverage of its test-suite.

You can perform the experiments by changing java to kotlin in the previous commands. Note that it would take up to 30 hours to compute the code coverage for Kotlin's test-suite.

Groovy

The build scripts of groovyc provide a command to generate code coverage reports; thus, we will use it instead of compiling each program in the test-suite individually with an instrumented compiler (This will take around 2 hours).

hephaestus@e0456a9b520e:~$ sdk use java 11.0.2-open
# Produce code coverage report for the test-suite of groovyc
hephaestus@e0456a9b520e:~$ cd ~/coverage/groovy
hephaestus@e0456a9b520e:~/coverage/groovy$ ./gradlew clean jacocoAllReport
hephaestus@e0456a9b520e:~/coverage/groovy$ cp build/jacoco/test.exec vanilla.exec
hephaestus@e0456a9b520e:~/coverage/groovy$ java \
    -jar $HOME/coverage/jacoco/lib/jacococli.jar report vanilla.exec \
    --classfiles $HOME/coverage/groovy/build/classes \
    --html vanilla --csv groovy-vanilla.csv
hephaestus@e0456a9b520e:~/coverage/groovy$ cd $HOME

To generate Groovy programs using Hephaestus, run the following commands. Note that these commands generate 10 Groovy programs using TEM, and 10 Groovy programs using TOM.

# You can skip this step if you have already generated some Groovy programs
hephaestus@e0456a9b520e:~$ hephaestus.py --bugs coverage_programs \
    --name groovy_tem_10 --language groovy \
    --iterations 10 --batch 10 --workers 2 --transformations 1 \
    --keep-all --dry-run -P
hephaestus@e0456a9b520e:~$ hephaestus.py --bugs coverage_programs \
    --name groovy_tom_10 --language groovy \
    --iterations 10 --batch 10 --workers 2 --transformations 0 \
    --keep-all --dry-run

Add the programs generated by Hephaestus to the test suite of groovyc:

hephaestus@e0456a9b520e:~$ ./eval-scripts/coverage/groovy-create-test-class.sh \
    $HOME/coverage \
    coverage_programs/groovy_tom_10/generator/ GeneratorSTCTest
hephaestus@e0456a9b520e:~$ ./eval-scripts/coverage/groovy-create-test-class.sh \
    $HOME/coverage \
    coverage_programs/groovy_tem_10/transformations/ InferenceSTCTest 0

Produce code coverage report for the test-suite of groovyc + the generated test cases by Hephaestus

hephaestus@e0456a9b520e:~$ cd ~/coverage/groovy
hephaestus@e0456a9b520e:~/coverage/groovy$ ./gradlew clean jacocoAllReport
hephaestus@e0456a9b520e:~/coverage/groovy$ cp build/jacoco/test.exec hephaestus.exec
hephaestus@e0456a9b520e:~/coverage/groovy$ java \
    -jar $HOME/coverage/jacoco/lib/jacococli.jar report hephaestus.exec \
    --classfiles $HOME/coverage/groovy/build/classes \
    --html hephaestus --csv groovy-hephaestus.csv

Print results

hephaestus@e0456a9b520e:~/coverage/groovy$ python ~/eval-scripts/compute_coverage.py \
    g groovy-vanilla.csv \
    groovy-hephaestus.csv \
    $HOME/data/coverage/compilers/groovy/groovy_whitelist
Line Coverage  Function Coverage    Branch Coverage
Initial                           81.98              71.86              78.52
Combination                       82.06              71.95              78.58
% change                           0.08               0.09               0.06
Absolute change                     129                 40                736

About

Replication package for the PLDI 2022 paper titled "Finding Typing Compiler Bugs".

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published