CSE320/hw2-doc
2022-04-16 11:33:13 -04:00
..
DebuggingRef.md feat: added files 2022-04-16 11:33:13 -04:00
README.md feat: added files 2022-04-16 11:33:13 -04:00

Homework 2 Debugging and Fixing - CSE 320 - Spring 2022

Professor Eugene Stark

Due Date: Friday 3/4/2022 @ 11:59pm

Introduction

In this assignment you are tasked with updating an old piece of software, making sure it compiles, and that it works properly in your VM environment.

Maintaining old code is a chore and an often hated part of software engineering. It is definitely one of the aspects which are seldom discussed or thought about by aspiring computer science students. However, it is prevalent throughout industry and a worthwhile skill to learn. Of course, this homework will not give you a remotely realistic experience in maintaining legacy code or code left behind by previous engineers but it still provides a small taste of what the experience may be like. You are to take on the role of an engineer whose supervisor has asked you to correct all the errors in the program, plus add additional functionality.

By completing this homework you should become more familiar with the C programming language and develop an understanding of:

  • How to use tools such as gdb and valgrind for debugging C code.
  • Modifying existing C code.
  • C memory management and pointers.
  • Working with files and the C standard I/O library.

The Existing Program

Your goal will be to debug and extend an old program called par, which was written by Adam M. Costello and posted to Usenet in 1993. I have rearranged the original source code and re-written the Makefile to conform to the format we are using for the assignments in this course. Besides a bug that was present in the original version, I have introduced a few additional bugs here and there to make things more interesting and educational for you 😉. Although you will need to correct these bugs in order to make the program function, they do not otherwise change the program behavior from what the author intended.

The par program is a simple paragraph reformatter. It is basically designed to read text from the standard input, parse the text into paragraphs, which are delimited by empty lines, chop each paragraph up into a sequence of words (forgetting about the original line breaks), choose new line breaks to optimize some criteria that are designed to produce a pleasing result, and the finally output the paragraph with the new line breaks. There are several parameters that can be set which affect the result: the width of the output text, the length of a "prefix" and a "suffix" to be prepended and appended to each output line, a parameter "hang", which affects the default value of "prefix", and a boolean parameter "last", which affects the way the last line of a paragraph is treated.

What you have to do is to first get the program to compile (for the most part, I did not modify the original code, which requires some changes for it to compile cleanly with the compiler and settings we are using). Then, you need to test the program and find and fix the bugs that prevent it from functioning properly. Some of the bugs existed in the original version and some I introduced for the purposes of this assignment. Finally, you will make some modifications to the program.

As you work on the program, limit the changes you make to the minimum necessary to achieve the specified objectives. Don't rewrite the program; assume that it is essentially correct and just fix a few compilation errors and bugs as described below. You will likely find it helpful to use git for this (I did). Make exploratory changes first on a side branch (i.e. not the master branch), then when you think you have understood the proper changes that need to be made, go back and apply those changes to the master branch. Using git will help you to back up if you make changes that mess something up.

Getting Started - Obtain the Base Code

Fetch base code for hw2 as you did for the previous assignments. You can find it at this link: https://gitlab02.cs.stonybrook.edu/cse320/hw2.

Once again, to avoid a merge conflict with respect to the file .gitlab-ci.yml, use the following command to merge the commits:

  git merge -m "Merging HW2_CODE" HW2_CODE/master --strategy-option=theirs

:nerd: I hope that by now you would have read some git documentation to find out what the --strategy-option=theirs does, but in case you didn't 😠 I will say that merging in git applies a "strategy" (the default strategy is called "recursive", I believe) and --strategy-option allows an option to be passed to the strategy to modify its behavior. In this case, theirs means that whenever a conflict is found, the version of the file from the branch being merged (in this case HW2_CODE/master) is to be used in place of the version from the currently checked-out branch. An alternative to theirs is ours, which makes the opposite choice. If you don't specify one of these options, git will leave conflict indications in the file itself and it will be necessary for you to edit the file and choose the code you want to use for each of the indicated conflicts.

Here is the structure of the base code:

.
├── .gitlab-ci.yml
└── hw2
    ├── doc
    │   ├── par.1
    │   ├── par.doc
    │   └── protoMakefile
    ├── hw2.sublime-project
    ├── include
    │   ├── buffer.h
    │   ├── debug.h
    │   ├── errmsg.h
    │   └── reformat.h
    ├── Makefile
    ├── rsrc
    │   ├── banner.txt
    │   ├── gettysburg.txt
    │   └── loremipsum.txt
    ├── src
    │   ├── buffer.c
    │   ├── errmsg.c
    │   ├── main.c
    │   ├── par.c
    │   └── reformat.c
    ├── test_output
    │   └── .git-keep
    └── tests
        ├── basecode_tests.c
        ├── rsrc
        │   ├── banner.txt
        │   ├── basic.in -> gettysburg.txt
        │   ├── basic.out
        │   ├── blank_lines.txt
        │   ├── EOF.in
        │   ├── EOF.out
        │   ├── gettysburg.txt
        │   ├── loremipsum.txt
        │   ├── prefix_suffix.in -> banner.txt
        │   ├── prefix_suffix.out
        │   ├── valgrind_leak.in -> gettysburg.txt
        │   ├── valgrind_leak.out
        │   ├── valgrind_uninitialized.err
        │   ├── valgrind_uninitialized.in -> loremipsum.txt
        │   └── valgrind_uninitialized.out
        ├── test_common.c
        └── test_common.h

The src directory contains C source code files buffer.c. par.c, reformat.c, and errmsg.c, which were part of the original code. In addition, I have added a new file main.c, with a single main() function that simply calls original_main() in par.c. This is to satisfy our requirement (for Criterion) that main() is the only function in main.c.

The include directory contains C header files buffer.h, reformat.h, and errmsg.h, which were part of the original source code. I have also added our debug.h header file which may be of use to you.

The doc directory contains documentation files that were part of the original distribution of par. The file par.1 is in the format traditionally used for Unix manual pages. This file par. is intended to be processed with the the formatting program nroff with argument -man; for example: nroff -man doc/par.1 | less could be used to format and view its contents.

The tests directory contains C source code (in file basecode_tests.c) for some Criterion tests that can help guide you toward bugs in the program. These are not guaranteed to be complete or exhaustive. The test_common.c and test_common.h contain auxiliary code used by the tests. The subdirectory tests/rsrc contains input files and reference output files that are used by the tests. The par program was not designed to be particularly conducive to unit testing, so all the tests we will make (including the tests used in grading) will be so-called "black box" tests, which test the input-output behavior of the program running as a separate process from the test driver. The test_common.c file contains helper functions for launching an instance of par as a separate process, redirecting stdin from an input file, collecting the output produced on stdout and stderr, checking the exit status of the program, and comparing the output against reference output.

The test_output directory is a "dummy" directory which is used to hold the output produced when you run the Criterion tests. Look there if you want to understand, for example, why the tests reported that the output produced by your program was not as expected.

Before you begin work on this assignment, you should read the rest of this document. In addition, we additionally advise you to read the Debugging Document. One of the main goals of this assignment is to get you to learn how to use the gdb debugger, so you should right away be looking into how to use this while working on the tasks in the following sections.

Part 1: Debugging and Fixing

You are to complete the following steps:

  1. Clean up the code; fixing any compilation issues, so that it compiles without error using the compiler options that have been set for you in the Makefile. Use git to keep track of the changes you make and the reasons for them, so that you can later review what you have done and also so that you can revert any changes you made that don't turn out to be a good idea in the end.

  2. Fix bugs.

    Run the program, exercising the various options, and look for cases in which the program crashes or otherwise misbehaves in an obvious way. We are only interested in obvious misbehavior here; don't agonize over program behavior that might just have been the choice of the original author. You should use the provided Criterion tests to help point the way, though they are not guaranteed to be exhaustive.

  3. Use valgrind to identify any memory leaks or other memory access errors. Fix any errors you find.

    Run valgrind using a command of the following form:

       $ valgrind --leak-check=full --show-leak-kinds=all --undef-value-errors=yes [PAR PROGRAM AND ARGS]
     

    Note that the bugs that are present will all manifest themselves in some way either as incorrect output, program crashes or as memory errors that can be detected by valgrind. It is not necessary to go hunting for obscure issues with the program output. Also, do not make gratuitous changes to the program output, as this will interfere with our ability to test your code.

    😱 The author of this program was pretty fastidious about freeing memory before exiting the program. Once you have fixed the bugs, the program should exit without any type of memory leak reported by valgrind, including memory that is "still reachable" at the time of exit. "Still reachable" memory corresponds to memory that is in use when the program exits and can still be reached by following pointers from variables in the program. Although some people consider it to be untidy for a program to exit with "still reachable" memory, it doesn't cause any particular problem. For the present program, however, there should not be any "still reachable" memory.

    😱 You are NOT allowed to share or post on PIAZZA solutions to the bugs in this program, as this defeats the point of the assignment. You may provide small hints in the right direction, but nothing more.

Part 2: Changes to the Program

Rewrite/Extend Options Processing

The basecode version of par performs its own ad hoc processing of command-line options. This is likely due to the fact that there did not exist a commonly accepted library package for performing this function at the time the program was written. However, as options processing is a common function that is performed by most programs, and it is desirable for programs on the same system to be consistent in how they interpret their arguments, there have been more elaborate standardized libraries that have been written for this purpose. In particular, the POSIX standard specifies a getopt() function, which you can read about by typing man 3 getopt. A significant advantage to using a standard library function like getopt() for processing command-line arguments, rather than implementing ad hoc code to do it, is that all programs that use the standard function will perform argument processing in the same way rather than having each program implement its own quirks that the user has to remember.

For this part of the assignment, you are to replace the original argument-processing code in main() by code that uses the GNU getopt library package. In addition to the POSIX standard getopt() function, the GNU getopt package provides a function getopt_long() that understands "long forms" of option arguments in addition to the traditional single-letter options. In your revised program, main() should use getopt_long() to traverse the command-line arguments, and it should support the following option syntax (in place of what was originally used by the program):

  • --version (long form only): Print the version number of the program.

  • -w WIDTH (short form) or --width WIDTH (long form): Set the output paragraph width to WIDTH.

  • -p PREFIX (short form) or --prefix PREFIX (long form): Set the value of the "prefix" parameter to PREFIX.

  • -s SUFFIX (short form) or --suffix SUFFIX (long form): Set the value of the "suffix" parameter to SUFFIX.

  • -h HANG (short form) or --hang HANG (long form): Set the value of the "hang" parameter to HANG.

  • -l LAST (short form) or either --last or --no-last (long form): Set the value of the boolean "last" parameter. For the short form, the values allowed for LAST should be either 0 or 1.

  • -m MIN (short form) or either --min or --no-min (long form). Set the value of the boolean "min" parameter. For the short form, the values allowed for MIN should be either 0 or 1.

You will probably need to read the Linux "man page" on the getopt package. This can be accessed via the command man 3 getopt. If you need further information, search for "GNU getopt documentation" on the Web.

😱 You MUST use the getopt_long() function to process the command line arguments passed to the program. Your program should be able to handle cases where the (non-positional) flags are passed IN ANY order. Make sure that you test the program with prefixes of the long option names, as well as the full names.

Revise the Error Message Scheme

The original program uses a very ad hoc scheme for error-message reporting: if an error occurs, a string describing the error is stored into a global character array errmsg with a hard-coded maximum size. (This hard-coded size has an occurrence in the fprintf() format string in par.c, which creates undesirable implicit coupling between par.c and errmsg.c.) At various points in the program, the existence of an error condition is checked by looking to see if the first character of the error message string is a null character '\0'. Before the program terminates, if an error message exists, then it is printed and the program exits with an error status, otherwise it exits with a success indication.

Your job is to revise the error message scheme to make it somewhat more general and to eliminate the hard-coded limitation on the length of an error message. In particular, you should replace the interface defined in errmsg.h by the following function prototypes (exactly as shown):

/**
 * @brief  Set an error indication, with a specified error message.
 * @param msg Pointer to the error message.  The string passed by the caller
 * will be copied.
 */
void set_error(char *msg);

/**
 * @brief  Test whether there is currently an error indication.
 * @return 1 if an error indication currently exists, 0 otherwise.
 */
int is_error();

/**
 * @brief  Issue any existing error message to the specified output stream.
 * @param file  Stream to which the error message is to be issued.  
 * @return 0 if either there was no existing error message, or else there
 * was an existing error message and it was successfully output.
 * Return non-zero if the attempt to output an existing error message
 * failed.
 */
int report_error(FILE *file);

/**
 * Clear any existing error indication and free storage occupied by
 * any existing error message.
 */
void clear_error();

The global array errmsg should be removed from errmsg.h and replaced by a pointer variable declared as static char * in errmsg.c. The functions whose prototypes are given above should be implemented so that there is no fixed maximum imposed on the length of an error message. This means that error messages should be dynamically allocated on the heap (for example, using strdup()). The implementation should take care not to leak any memory used for error messages; for example if a new error message is set when one already exists. Before exiting, the program should call clear_error() to cause any existing error message to be freed.

Part 3: Testing the Program

For this assignment, you have been provided with a basic set of Criterion tests to help you debug the program.

In the tests/basecode_tests.c file, there are five test examples. You can run these with the following command:

    $ bin/par_tests

To obtain more information about each test run, you can supply the additional option --verbose=1. You can also specify the option -j1 to cause the tests to be run sequentially, rather than in parallel using multiple processes, as is the default. The -j1 flag is necessary if the tests could interfere with each other in some way if they are run in parallel (such as writing the same output file). You will probably find it useful to know this; however the basecode tests have been written so that they each use output files named after the test and (hopefully) will not interfere with each other.

The tests have been constructed so that they will point you at most of the problems with the program. Each test has one or more assertions to make sure that the code functions properly. If there was a problem before an assertion, such as a "segfault", the test will print the error to the screen and continue to run the rest of the tests. The basecode test cases check the program operation by reading input from a pre-defined input file, redirecting stdout and stderr to output files, and comparing the output produced against pre-defined reference files. Some of the tests use valgrind to verify that no memory errors are found. If errors are found, then you can look at the log file that is left behind (in the test_output directory) by the test code. Alternatively, you can better control the information that valgrind provides if you run it manually.

The tests included in the base code are not true "unit tests", because they all run the program as a black box using system(). You should be able to follow the pattern to construct some additional tests of your own, and you might find this helpful while working on the program. You are encouraged to try to write some of these tests so that you learn how to do it. Note that in the next homework assignment unit tests will likely be very helpful to you and you will be required to write some of your own. Criterion documentation for writing your own tests can be found here.

😱 Be sure that you test non-default program options to make sure that the program does not crash or otherwise misbehave when they are used.

Hand-in Instructions

Ensure that all files you expect to be on your remote repository are committed and pushed prior to submission.

This homework's tag is: hw2

$ git submit hw2