Out: Friday, July 12, 2019
Due: Thursday, July 25, 2019 by 11:59 pm
Here you will use the LinkedList and HashTable modules that you built in Homework 1 in order to finish our implementation of a file system crawler, indexer, and search engine:
As before, please read through this entire document before beginning the assignment, and please start early! There is a fair amount of coding you need to do to complete this assignment, and it will definitely expose any conceptual weaknesses you have with the prior material on C, pointers, malloc/free, and the semantics of the LinkedList and HashTable implementations.
In Homework 1, we asked you to look for out-of-memory errors and return an error code to the caller. To make life a bit easier, in Homework 2, you can use Verify333() to test for an out-of-memory error; this way, running out of memory will cause the program to halt, and you don't need to deal with returning an error code if one occurs.
You're going to write a module that reads the contents of a text file into memory and then parses the text file to look for words within it. As it finds words, it will build up a HashTable that contains one record for each word. Each record will contain a lowercase copy of the word, and also a sorted linked list. Each record of the linked list contains an offset within the file that the word appeared in (the first character in the file has offset zero).
Our word parser won't be very smart. It will treat as a word any non-zero sequence of alphabetic characters separated by non-alphabetic characters.
So, graphically, what your module will do is take a text file that contains something like this:
My goodness! I love the course CSE333.\n
I'll recommend this course to my friends.\n
(where '\n' represents the newline control character that appears at the end of each line in the input file) and produces a data structure that looks like this:
Specifically, note a few things:
You should follow these steps to do this assignment:
git pull
to retrieve new folders containing
the starter code for hw2 and a directory containing test data for the
remaining parts of the project. Check that you have everything.bash$ git pull ...git output... bash$ ls clint.py gtest hw0 hw1 hw2 projdocs
(Note that you still need the hw1 directory; hw2 won't build properly without it. It's ok if you've deleted your hw0 directory.)
fileparser.c
. Start by reading through fileparser.h
and
make sure you understand the semantics of the functions. Also,
look at the WordPositions structure typedef'ed in fileparser.h and
compare it to the figure above. The function BuildWordHT() builds
a HashTable that looks like the figure, and each value in the
HashTable contains a heap-allocated WordPositions structure.
fileparser.c
to get a sense
of it's layout, and look for all occurrences of STEP X
(e.g., STEP 1, STEP 2, ...) for where you need to add code. Be
sure to read the full file before adding any code, so you can see
the full structure of what we want you to do. Once you're finished
adding code, run the test suite and you should see three
suites succeed: FPTestReadFile, FPTestBuildWordHT, and
FPTestBigBuildWordHT.
static
) functions
when that makes sense.
At a high-level, a search engine has three major components: a crawler, an indexer, and a query processor. A crawler explores the world, looking for documents to index. The indexer takes a set of documents found by the crawler, and produces something called an inverted index out of them. A query processor asks a user for a query, and processes it using the inverted index to figure out a list of documents that match the query.
File system crawler: Your file system crawler will be provided with the name of a directory in which it should start crawling. Its job is to look through the directory for documents (text files) and to hand them to the indexer, and to look for subdirectories; it recursively descends into each subdirectory to look for more documents and sub-sub-directories. For each document it encounters, it will assign the document a unique "document ID", or "docID". A docID is just a 64-bit unsigned integer.
Your crawler itself will build two hash tables in memory, adding to them each time it discovers a new text file. The two hash tables map from docID to document filename, and from document filename to docID:
For each document the crawler finds, it will make use of your part A code to produce a word hashtable using BuildWordHT().
Indexer: This is the heart of the search engine. The job of the indexer is to take each word hashtable produced by BuildWordHT(), and fold its contents in to an inverted index. An inverted index is easy to understand; it's just a hash table that maps from a word to a "posting list," and a posting list is just a list of places that word has been found.
Specifically, the indexer should produce an in-memory hash table that looks roughly like this:
Walking through it, the inverted index is a hash table that maps from a (hash of a) word to a structure. The structure (shown in green) contains the word as a string, and also a HashTable. The HashTable maps from the docID (not the hash of docID) to a LinkedList. The LinkedList contains each position that word appeared in that docID.
So, based on the figure, you can see that the word "course" appeared in a single document with docID 3, at byte offsets 25 and 62 from the start of file. Similarly, the word "love" appears in three documents: docID 1 at positions 7 and 92, docID 3 at position 16, and docID 4 at positions 18, 21, and 55.
The bulk of the work in this homework is in this step. We'll tackle it in parts.
doctable.h
; this is the public interface to
the module that builds up the docID-to-docname HashTable and the
docname-to-docID HashTable. Make sure you understand the
semantics of everything in that header file; note that a single
DocTable structure contains both of these tables, so when you
implement AllocateDocTable(), you'll end up malloc'ing a structure
that contains two HashTables, and you'll allocate each of those
HashTables.
doctable.c
; this is our partially
completed implementation. Be sure to read the full file. Your
job, as always, is to look for the "STEP X" comments and finish
our implementation. Once you've finished the implementation,
re-compile and re-run the test_suite executable to see if you pass
our DTTestSimple test suite. If not, go back and fix some bugs!
filecrawler.h
; this is the public
interface to our file crawler module. Make sure you understand
the semantics of everything in that header file. Next, read
through the full filecrawler.c
and then complete our
implementation. Once you're ready, re-compile and re-run the
test_suite executable to see if you pass our FCTestSimple test
suite. If not, go back and fix some bugs!
memindex.h
. This is the public
interface to the module that builds up the in-memory inverted
index. Make sure you understand the semantics of everything in
that header file. Next, read the full memindex.c
, and then complete
our implementation. (This is the most involved part of the
assignment.) Once you're ready, re-compile and re-run the
test_suite executable to see if you pass our MITestSimple test
suite. If not, go back and fix some bugs!
Once you've passed all of the test suites, re-rerun the test suites under valgrind and make sure you don't have any memory leaks.
Congrats, you've passed part B of the assignment!
Now that you have a working inverted index, you're ready to build your search engine. The job of a search engine is to receive queries from a user, and return a list of matching documents in some rank order.
For us, a query is just a string of words, such as:
course friends my
The goal of our search engine is to find all of the documents
that contain all of the words. So, if a document has
the words "course" and "my" but not the word "friends," it shouldn't
match.
To execute a query, first the query processor must split the query up into a list of words (the strtok_r() function is useful for this). Next, it looks up in the inverted index the list of documents that match the first word. This is our initial matching list.
Next, for each additional word in the query, the query processor uses the inverted index to get access to the HashTable that contains the set of matching documents. For each document in the matching list, the query processor tests to see if that document is also in the HashTable. If so, it keeps the document in the matching list, and if not, it removes the document from the matching list.
Once the processor has processed all of the words, it's done. Now, it just has to rank the results, sort the results by rank, and return the sorted result list to the user.
For us, our ranking function is very simple: given a document that matches against a query, we sum up the number of times each query word appears in the document, and that's our rank. So, if the user provides the query "foo" and that words appears on a document 5 times, the rank for that document given that query is 5. If the user provides a multi-word query, our sum includes the number of times each of the words in the query appears. So, if the user provides the query "foo bar", the word "foo" appears in the document 5 times, and the word "bar" appears 10 times, the rank of the document for that query is 15. The bigger the sum, the higher the rank, and therefore the better the search result.
We have provided a mostly empty skeleton file searchshell.c
.
It is up to you to complete it by implementing a program that uses
the Linux console to interact with the user. When you run the
processor (called searchshell -- you can try a working searchshell
in the solution_binaries/ directory), it takes from a command line
argument the name of a directory to crawl. After crawling the
directory and building the inverted index, it enters a query processing
loop, asking the user to type in a search string and printing the
results.
If the user signals end-of-file when the program asks for a search string
(control-D from the linux terminal keyboard),
the program should clean up any allocated resources - particularly memory - and shut down.
When you are ready, try running ./searchshell to interact with your program and see if your output matches the transcript from a search session with our solution. Alternatively, compare your searchshell to the searchshell we provided in the solution_binaries/ directory. Note that our ranking function does not specify an ordering for two documents with the same rank. Documents that have the same rank may be listed in any order, and that might be different from the ordering in the sample transcript or produced by the solution_binaries version of searchshell.
We're offering two bonus tasks for a tiny amount of extra credit. These are purely optional; if you choose not to do either, your grade won't be negatively affected at all. These are just if you happen to have the time and interest! You can do either or both of the bonus tasks.
If you do work on either of the bonus tasks, you must also include a hw2-bonus
tag in your repository. While grading, we will use whichever commit has that tag
to examine the bonus, so it may be the same or a different commit from the one that has hw2-final
.
If you do not have a hw2-bonus
tag in your repository, we will assume you did
not choose to submit anything for the bonus (again, which will not negatively affect your grade in any way!)
Implement a stop word filter. The search shell should accept a second, optional argument "-s" that will act as a flag for turning the filter on. When the flag is not specified, your program should not filter any stop words. It is up to you to decide how you will implement the stop word filter (and where you'll get your list of stop words), but be sure to explain in your README file what changes you had to make and how your filter works. Stop words that have apostrophes in them should be handled the same way that you've handled the words in the documents.
alice "cool fountains" flowersThis query would match all documents that contain all of the following:
Using the positions information in the inverted index postings list, implement support for phrase search. You'll have to also modify the query processor to support phrase syntax; phrases are specified by using quotation marks. Be sure to create a README.md file that describes what changes you had to make to get phrasing to work.
When you are ready to turn in your assignment, you should follow
the same procedures you used for hw0 and hw1, except this time tag
the repository with hw2-final
. Remember to clean up,
commit, and push all necessary files to your repository before you
add the tag. After you have created and pushed the
tag, remember to test everything by
creating a new clone of the repository in a separate, empty
directory, checkout the hw2-final
tag, and verify that
everything works as expected. Refer to the hw0 turnin instructions
for details, and follow those steps carefully.
When you clone your repository to check it, it normally will not include hw1/libhw1.a, which is needed to build hw2. Either run make in hw1 to recreate it, or copy the libhw1.a file in hw1/solution_binaries and place it in the hw1 folder.
If you fail to check your work and your project doesn't build properly when the same steps are done by the course staff to grade it, you may lose a huge amount of the possible credit for the assignment even if almost absolutely everything is actually correct.
We will be basing your grade on several elements:
test_suite.c
. If your code fails a
test, we won't attempt to understand why: we're planning on just
including the number of points that the test drivers print out.
clint.py
tool to check your code for style issues.