In this assignment and the next, you will develop a search engine for text documents. Your engine will crawl through a directory on a local disk looking for documents. When it finds them, it will index the words that appear in those documents. Then it will answer queries posed by users.
In A4, you will build an working but inefficient version of the search engine. In A5, you will upgrade the data structures involved to make the engine scale up to large collections of documents.
What you’ll do: Implement some functors, use a testing tool called Bisect, and give your teammates a performance evaluation.
- Learn more about the OCaml module system.
- Implement two data structures, dictionaries and sets.
- Practice documenting and implementing abstraction functions and representation invariants.
- Gain experience with glass-box testing.
- Optionally, learn about regular expressions.
Table of contents:
- Step 1: Team Maintenance
- Step 2: Explore the Starter Code
- Step 3: Dictionaries
- Step 4: Sets
- Step 5: Index
- Step 6: Query
- Step 7: Bisect
- Documentation and Testing
Step 1: Team Maintenance
Your A4 team will be the same as your A3 team. At your first team meeting about A4, please do two things:
Assign new roles for this assignment. Make sure to give everyone a different role than they have had before.
Before the team meeting, every team member should print out and fill out an Internal Evaluation for all other team members. These are meant to be anonymous, and to provide advice to team members on how they are fulfilling their teamwork obligations, as well as how to improve. At the end of the meeting, distribute the evaluations. Each team member should shuffle the papers they receive, so as to maintain anonymity. Team members should review the evaluations on their own after the meeting.
Note that at the end of A5, you will submit a similar evaluation to the professor, and that those evaluations will become part of team members’ grades. So it would be unethical to give a team member an evaluation that does not correspond to your actual opinion of their teamwork performance. If you are disturbed by a team member’s lack of contribution to your team, now is the time to make that clear. Recall the Hitchhiker assignment: you need to be clear about your expectations and not enable hitchhikers. On the other hand, if you are impressed by a team member’s contributions, now is the time to make that clear. Let them know how much they are valued.
Step 2: Explore the Starter Code
There is a makefile provided with the usual targets: build, test, check, finalcheck, zip, docs, and clean.
We are back to an autograded assignment, hence your submission must pass
make check. Now that you’re used to working with
.mli files, we have
omitted the comments about what you are allowed to change. Any names
and types that we provide in an interface may not be changed, but you
may of course add new declarations and definitions. If you’re ever in
doubt about whether a change is permitted or not, just run
it will tell you whether you have changed the interface in a prohibited
Step 3: Dictionaries
The most important data structure that your search engine will need is a dictionary—that is, a mapping from keys to values. In this assignment, we’ll keep it simple and implement a dictionary with an association list. In A5, we’ll upgrade to a more efficient implementation.
ListDictionary, following the specifications and comments
provided in the starter code. Using test-driven development, also
implement unit tests for
AF and RI: Note that you need to document and implement an
abstraction function and any representation invariants. The documentation
goes above the representation type. The implementations go in
Format functions are discussed in the textbook in the section on abstract types under the heading “Custom Printers”. There is a helper function provided for you in the starter code for
format; you are free to improve its output as you wish.
How to implement
rep_okis discussed in the textbook in the section on implementing representation invariants. You do not have to insert calls to
ListDictionaryimplementation, though you might find it useful for debugging purposes.
You do not need to have OUnit tests for either of these functions. Indeed, it would be hard or perhaps even impossible to write such tests.
Example: the file
exampleDictionary.ml contains an example of how to
create a dictionary. That file is intended to be loaded in utop with
not to be compiled with
Step 4: Sets
Your search engine will also need a data structure for sets. In this
assignment, we’ll use a dictionary to represent a set. The core idea of
this representation is that the elements of the set are the keys in the
dictionary. The values in the dictionary are thus irrelevant, so they
might as well simply be
(), the unit value. Although there is a bit of
extra space overhead using this idea, because of storing the unit value,
it nonetheless provides a way to turn any dictionary data structure
into a set data structure. We will profit from that in A5, when we will
automatically get an improvement in the performance of our set data
structures by upgrading our dictionary data structures.
Use that idea to implement
DictionarySet, following the specifications
and comments provided in the starter code. Also implement unit tests for
This is the stopping point for a satisfactory solution.
Step 5: Index
Now that we have the data structures we need, it’s time to implement the search engine. We’ll start with indexing the words found in files.
Implement the functions
Engine.Make. It is a functor that produces an engine from
dictionaries and sets. The
ListEngine module uses
to create an engine based on the association list dictionaries you
created earlier. Implement tests for
provide more instructions and hints, below.
Files. The prototypical kind of file you should have in mind is a book stored as an ASCII-encoded plain text file, such as this edition of Alice’s Adventures in Wonderland. It would be reasonable to assume that the individual lines of files are not overly long, but that there might be many lines in a file.
Start by creating your own small test files. As a source for larger test files, we recommend Project Gutenberg. Project Gutenberg files are often encoded in UTF-8 instead of ASCII. On most Unix systems, including the 3110 VM, you can convert UTF-8 to ASCII with the following command:
iconv -f UTF-8 -t ASCII -c in.txt >out.txt
in.txt is the name of the input UTF-8 file and
out.txt is the name of the output ASCII file.
As large test cases, we will use directories containing up to 3 MB of files. That’s a high enough number of words that you will want to be careful about tail recursion.
Words: For purposes of this assignment, we define a word as follows:
A whitespace character is any space, tab, or newline character (i.e., carriage return or line feed).
A preword is any maximal length sequence of characters in a file that does not contain any whitespace.
A boundary character is any lowercase letter, uppercase letter, or digit.
A word is the maximal length subsequence of a preword that begins and ends with a boundary character. In between those there may be any number of any other characters. A preword that contains no boundary characters does not correspond to any word.
There will no doubt be some weird corner cases resulting from this definition of words. But we need a standard definition; this one is relatively simple for us all to use, and it gets many common cases right.
For example, given a file containing the following text:
"I would found an institution where any person can find instruction in any study." ---E. Cornell (b. 1807)
The words in that file would be: 1807, an, any, b, can, Cornell, E, find, found, I, in, institution, instruction, person, study, where, would.
Hint: there are 3755 words in Alice’s Adventures in Wonderland, and after converting them all to lowercase, only 3278 distinct words remain.
Paths and filenames:
Anywhere you need to use a path separator in your code, use
Filename.dir_seprather than hard-coding a forward or backward slash; this will ensure that your code works on Windows and Unix platforms.
Never change the case of any file or directory names that the user passes to your engine or that the operating system returns to your engine. Some underlying file systems are case sensitive whereas others are case insensitive, so your code should preserve whatever case it receives.
For processing directories, you will find the
Unixmodule helpful, in particular the
For processing files, you will find the
Pervasivesmodule helpful, in particular the
input_linefunction will handle newlines for you, so that you don’t have to worry about stripping out those particular whitespace characters.
To parse a line of a file into its component words, you can either directly implement it yourself using the
Stringmodule’s functions, or you can implement it with regular expressions using the
Strmodule. The latter is recommended but certainly not required. Regular expressions are a kind of small programming language; you would find it immensely useful to learn them if you haven’t picked it up somewhere already. Here’s a good tutorial: http://regexone.com/.
The modules mentioned above are specifically permitted by this assignment even if they have side effects or are imperative. When dealing with I/O, side effects are unavoidable.
Step 6: Query
The queries that users pose will have one of two forms. Abstractly those two forms are as follows, in which the NOT clause is always optional:
“and-not” queries: AND (w1, w2, …, wn), NOT (u1, u2, … um)
“or-not” queries: OR (w1, w2, …, wn), NOT (u1, u2, … um)
For example, “AND (far, away)” would return all files that contain both the words “far” and “away”, whereas “AND (far, away), NOT (day, night)” would return all files that do contain both “far” and “away” but do not contain “day” nor “night”. Likewise, “OR (all, nothing)” would return any file that contains “all” or “nothing” (or both), and “OR (all, nothing), NOT (between)” would return any file that contains “all” or “nothing” (or both) but does not contain “between”.
Queries must be case insensitive, which is the behavior you expect from Google. That means queries for the words “far”, “Far”, and “FAR” should all return the same results. Likewise, a file containing “far”, “Far”, or “FAR” would be returned by any of those queries.
Engine.Make, and implement unit
This is the stopping point for a good solution.
Step 7: Bisect
The textbook contains a tutorial on a tool called Bisect, which is a code coverage testing tool. Do that tutorial.
make bisect on your solution. Open
examine the code coverage you have achieved in
engine.ml. It’s unlikely you are at
100%, and probably it’s impossible to achieve that. For example, unit
tests are unlikely to ever exercise (all or any of) your
rep_ok functions. But outside of those, look for any lines colored
red by the Bisect report. Do your best to invent new unit tests that
will cause those lines to be executed. Add those tests to
How high do you need to get your coverage? In an ideal world you would
cover every line that you know it’s possible to cover, or at least that
is feasible to write unit tests to cover. With modest effort, the staff
solution to this assignment was able to achieve 90-100% code coverage in
those three files, excluding
rep_ok implementations. Use
(*BISECT-IGNORE-END*) comments described in the textbook to exclude
those functions from analysis.
engine.ml worth 8 points each, and
dictionarySet.ml worth 4 points. We’ll look at your code-coverage
percentages and round any that are at least 90% to up a full 100%. Then
you’ll get a number of points that is your percent code coverage on a
file times the number of points the file is worth, rounded to the
There is a potential abuse that we need to discuss. A team might insert
(*BISECT-IGNORE*) around code that is not part of
rep_ok. In some places this would be reasonable—for example, if
there is some defensive code that checks a precondition and raises an
exception if it is violated, and it turns out to be impossible or
infeasible to write a unit test to trigger that exception. In such a
case, you should add additional source code comments to explain why it
is reasonable to ignore that code in the Bisect report. Graders will
read your source code to make sure you have done this. If you do not
give reasonable justifications, or if you have used
to unfairly bump up your code coverage percent, the grader will assess a
This is the first time we’ve ever asked the entire class to use Bisect. Who knows what might happen? Please approach this—as the course staff will—with a spirit of fun about learning a new tool.
Here’s what we consider a satisfactory, good, and excellent solution:
Satisfactory: The solution passes
make check. The list dictionary and set data structures are implemented and unit tested.
Good: The search engine is also implemented.
Excellent: The test suite also achieves at least 90% code coverage as measured by Bisect on
Looking ahead to A5: Your A4 grade will be based on what you complete by the time A4 is due. A5 will then ask you to add new functionality. Any functionality that you leave incomplete in A4 will become part of the Satisfactory scope of A5.
Documentation and Testing
As in A2 and A3, the graders will look at the documentation produced
But the graders will not run
make test on your submission. That’s
because you will no doubt have some large test directories, and
we don’t want to clutter CMS with large files. So the graders will
test.ml rather than run it.
Make sure your team’s NetIDs are in
authors.mli, and set the
hours_worked variable at the end of
make zip to construct the ZIP file you need to submit on CMS.
Any mal-constructed ZIP files will receive a penalty of 20 points.
The size of the the files is limited in CMS to 1 MB. Please stay
within the size allotted, and do not submit large (or even any)
test directories. The
make zip command will run Bisect to
produce your code coverage report and include it in the zipfile.
If the report isn’t in your zipfile, we won’t be able to give you any
credit for the Excellent scope, so please make sure it’s there.
Submit your zipfile on CMS. Double-check before the deadline that you have submitted the intended version of your file.
Congratulations! Your search is off to a good start.
Acknowledgement: Adapted from Prof. Greg Morrisett, Dean of Cornell CIS.