CSE 13S
Assignment 7
Lempel-Ziv Compression
CSE 13S – Winter 2021
Due: March 14th at 11:59 pm PST
Everyday, we create 2.5 quintillion bytes of data – so much that 90% of the data in
the world today has been created in the last two years alone.
—IBM report on Big Data (2011)
1 Introduction
Compressing data means reducing the number of bits needed to represent it. Compressed data, due to its reduced
size, is a lot faster to transfer and less expensive to store, freeing up storage and increasing network capacity. Algorithms
that perform data compression are known as data compression algorithms. Data compression algorithms
are divided into two categories: lossy and lossless. Lossy compression algorithms compress more than lossless
compression algorithms, but at the cost of losing some information, which can’t be recovered. Lossy compression
algorithms are typically used for audio and video files, where a loss in quality is tolerable and often not noticeable.
Lossless compression algorithms on the other hand don’t compress as much, but do not lose any data, meaning
compressed information can be exactly reconstructed back into its uncompressed form. Lossless compression
algorithms are used for data that must maintain integrity, such as binaries, text documents, and source code.
2 Lempel-Ziv Compression
Abraham Lempel and Jacob Ziv published papers for two lossless compression algorithms, LZ77 and LZ78, published
in 1977 and 1978, respectively. The core idea behind both of these compression algorithms is to represent
repeated patterns in data with using pairs which are each comprised of a code and a symbol. A code is an unsigned
16-bit integer and a symbol is a 8-bit ASCII character.
We will explain this algorithm by working through an example. Assume we have some string “abab”. Assume we
also have a dictionary where the key is a prefix, also known as a word, and the value is a code that we can utilize
for fast look-ups. Since codes are 16-bit unsigned integers, the dictionary is limited to 216 −1 possible codes. Why
don’t we have 216 codes? Because we will be reserving a code to be used as a stop code which indicates the end of
our encoded data. We will define the maximal usable code as MAX_CODE, which has the value 216−1. The dictionary
is initialized with the empty word, or a string of zero length, at the index EMPTY_CODE, which is a macro for 1. We
will specify the first index in which a new word can be added to as the macro START_CODE, which has the value 2.
We now consider the first character: ‘a’. Since this is the first character, we know that the string of all previously
seen characters up to this point must be the empty word; we haven’t considered any characters prior to this point.
We then append the current character ‘a’ to the empty word, which yields the word “a”. We check the dictionary
in vain to see if we’ve encountered this word before. Since we haven’t yet seen this word, we will add this first word
© 2021 Darrell Long 1
Compression Example
Initialized dictionary
MAX_CODE
to the dictionary and assign it the first available code of 2, or START_CODE. We set our previously seen word to be
the empty word and output the pair (EMPTY_CODE, ‘a’).
We continue on, now considering the second character: ‘b’. We append this character to our previously seen word,
which yields “b” as our current word. We again check the dictionary in vain to see if we’ve encountered this word
before. Since we haven’t, we add this word to the dictionary and assign it the next available code of 3. We set the
previously seen word to the empty word and output the pair (EMPTY_CODE, ‘b’).
The next character we see is ‘a’. We append this character to our previously seen word, which yields “a” as our
current word. We check the dictionary to see if we’ve encountered this word before. Indeed we have. We set the
previously seen word to “a” and proceed to the next character without any further output.
We read in the last character, ‘b’. We append this character to our previously seen word, which yields “ab”. Clearly,
this word isn’t in the dictionary, so pair comprised of the current symbol, ‘b’, and the code of the previously we add
it and assign it the next available code of 4. What should we output? The seen word. Since our previously seen
word at this stage was “a”, this code will be 2. Thus, we output (2, ‘b’). To finish off compression, we output the
final pair (STOP_CODE, 0). As you can imagine, this pair signals the end of compressed data. The symbol in this
final pair isn’t used and the value is of no significance. The macro STOP_CODE has the value of 0.
Of course, a compression algorithm is useless without a corresponding decompression algorithm. Assume we’re
reading in the output from the preceding compression example. The output was comprised of the following pairs,
in order: (EMPTY_CODE, ‘a’), (EMPTY_CODE, ‘b’), (2, ‘b’), (STOP_CODE, 0). Similar to the compression algorithm, the
decompression algorithm initializes a dictionary containing only the empty word. The catch is that the key and
value for the decompressing dictionary is swapped; each key is instead a code and each value a word. As you
might imagine, the decompression algorithm is the inverse of the compression algorithm, and thus from the output
pairs, the decompression algorithm will recreate the same dictionary used during compression to output the
decompressed data.
We consider the first pair: (EMPTY_CODE, ‘a’). We append the symbol ‘a’ to the word denoted by EMPTY_CODE,
which is the empty word. Thus, the current word is “a”. We add this word to the dictionary and assign it the next
available code of 2, then output the current word “a”.
We consider the next pair: (EMPTY_CODE, ‘b’). We append the symbol ‘b’ to the word denoted by EMPTY_CODE,
which is the empty word. Thus, the current word is “b”. We add this word to the dictionary and assign it the next
available code of 3, then output the current word “b”.
We now consider the next pair: (2, ‘b’). We append the symbol ‘b’ to the word denoted by the code 2, which we
previously added to our dictionary. The word denoted by this code is “a”, whence we obtain our current word of
“ab”. We add this word to the dictionary and assign it the next available code of 4, then output the current word
“ab”. Finally, we read in the last pair: (STOP_CODE, 0). Since the code is STOP_CODE, we know that we have finished
decompression.
If the basic idea behind the compression and decompression algorithms do not immediately make sense to you,
MAX_CODE
or if you desire a more visual representation of how the algorithms work, make sure to attend section and get help
early! Things will not get easier as time goes on.
3 Your Task
Your task is to implement two programs called encode and decode which perform LZ78 compression and decompression,
respectively. The requirements for your programs are as follows:
- encode can compress any file, text or binary.
- decode can decompress any file, text or binary, that was compressed with encode.
- Both operate on both little and big endian systems. Interoperability is required.
- Both use variable bit-length codes.
- Both perform read and writes in efficient blocks of 4KB.
- Both encode and decode must interoperate with the provided binaries – not just your code.
- Specifics
You will need to implement some new ADTs for this assignment: an ADT for tries and an ADT for words. In addition
to these new ADTS, you will need to be concerned about variable-length codes, I/O, and endianness.
4.1 Tries
The most costly part of compression is checking for existing prefixes, or words. You could utilize a hash table, or
just an array to store words, but that wouldn’t be optimal, as many of the words you need to store are prefixes of
other words. Instead you choose to utilize a trie.
A trie1
is an efficient information re-trie-val data structure, commonly known as a prefix tree. Each node in a trie
represents a symbol, or a character, and contains n child nodes, where n is the size of the alphabet you are using.
In most cases, the alphabet used is the set of ASCII characters, so n = 256. You will use a trie during compression
to store words.
1Edward Fredkin, “Trie memory.” Communications of the ACM 3, no. 9 (1960): 490–499.
© 2021 Darrell Long 3
Above is an example of a trie containing the following words: “She”, “sel l s”, “sea”, “shel l s”, “by”, “the”, “sea”,
“shor e”. Searching for a word in the trie means stepping down for each symbol in the word, starting from the root.
Stepping down the trie is simply checking if the current node we have traversed to has a child node representing
the symbol we are looking for, and setting the current node to be the child node if it does exist. Thus, to find “sea”,
we would start from the trie’s root and step down to ‘s’, then ‘e’, then ‘a’. If any symbol is missing, or the end of the
trie is reached without fully matching a word, while stepping through the trie, then the word is not in the trie. You
must follow the specification shown below when implementing your trie ADT.
4.1.1 TrieNode - struct TrieNode {
- TrieNode * children [ ALPHABET ];
- uint16_t code ;
- };
The TrieNode struct will have the two fields shown above. Each trie node has an array of 256 pointers to trie
nodes as children, one for each ASCII character. It should be easy to see how this simplifies searching for the next
character in a word in the trie. The code field stores the 16-bit code for the word that ends with the trie node
containing the code. This means that the code for some word “abc” would be contained in the trie node for ‘c’.
Note that there isn’t a field that indicates what character a trie node represents. This is because the trie node’s
index in its parent’s array of child nodes indicates what character it represents. The trie_step() function will be
repeatedly called to check if a word exists in the trie. A word only exists if the trie node returned by the last step
corresponding to the last character in the word isn’t NULL.
4.1.2 TrieNode *trie_node_create(uint16_t code)
Constructor for a TrieNode. The node’s code is set to code. Make sure each of the children node pointers are NULL.
4.1.3 void trie_node_delete(TrieNode *n)
Destructor for a TrieNode. Note that only a single pointer is passed here. The destructors you have written in
the past have taken double pointers in order to NULL the pointer by dereferencing it. For this assignment, a single
pointer is much more manageable and simplifies the code you have to otherwise write.
© 2021 Darrell Long 4
4.1.4 TrieNode *trie_create(void)
Initializes a trie: a root TrieNode with the code EMPTY_CODE. Returns the root, a TrieNode *, if successful, NULL
otherwise.
4.1.5 void trie_reset(TrieNode *root)
Resets a trie to just the root TrieNode. Since we are working with finite codes, eventually we will arrive at the
end of the available codes (MAX_CODE). At that point, we must reset the trie by deleting its children so that we can
continue compressing/decompressing the file. Make sure that each of the root’s children nodes are NULL.
4.1.6 void trie_delete(TrieNode *n)
Deletes a sub-trie starting from the trie rooted at node n. This will require recursive calls on each of n’s children.
Make sure to set the pointer to the children nodes to NULL after you free them with trie_node_delete().
4.1.7 TrieNode trie_step(TrieNode n, uint8_t sym)
Returns a pointer to the child node reprsenting the symbol sym. If the symbol doesn’t exist, NULL is returned.
4.2 Word Tables
Although compression can be performed using a trie, decompression still needs to use a look-up table for quick
code to word translation. This look-up table will be defined as a new struct called a WordTable. Since we can
only have 216 −1 codes, one of which is reserved as a stop code, we can use a fixed word table size of UINT16_MAX,
where UINT16_MAX is a macro defined in inttypes.h as the maximum value of a unsigned 16-bit integer. Hint:
this has the exact same value as MAX_CODE, which we defined earlier.
Why aren’t we using hash tables to store words? Because there is a fixed number of codes. Each index of this word
table will be a new struct called a Word. You will store words in byte arrays, or arrays of uint8_t. This is because
strings in C are null-terminated, and problems with compression occur if a binary file is being compressed and
contains null characters that are placed into strings. Since we need to know how long a word is, a Word will also
have a field for storing the length of the byte array, since we can’t use string.h functions like strlen() on byte
arrays. You must use the following specification for the new Word ADT.
4.2.1 Word - struct Word {
- uint8_t * syms ;
- uint32_t len;
- };
A Word holds an array of symbols, syms, stored as bytes in an array. The length of the array storing the symbols a
Word represents is stored in len.
© 2021 Darrell Long 5
4.2.2 WordTable - typedef Word * WordTable
To make things easier to reason about, we can define an array of Words as a WordTable.
4.2.3 Word word_create(uint8_t syms, uint32_t len)
Constructor for a word where sysms is the array of symbols a Word represents. The length of the array of symbols
is given by len. This function returns a Word * if successful or NULL otherwise.
4.2.4 Word word_append_sym(Word w, uint8_t sym)
Constructs a new Word from the specified Word, w, appended with a symbol, sym. The Word specified to append to
may be empty. If the above is the case, the new Word should contain only the symbol. Returns the new Word which
represents the result of appending.
4.2.5 void word_delete(Word *w)
Destructor for a Word, w. Like with trie_node_create() in §4.1.3, a single pointer is used here to reduce the
complexity of memory management, thus reducing the chances of having memory-related errors.
4.2.6 WordTable *wt_create(void)
Creates a new WordTable, which is an array of Words. A WordTable has a pre-defined size of MAX_CODE, which
has the value UINT16_MAX - 1. This is because codes are 16-bit integers. A WordTable is initialized with a single
Word at index EMPTY_CODE. This Word represents the empty word, a string of length of zero.
4.2.7 void wt_reset(WordTable *wt)
Resets a WordTable, wt, to contain just the empty Word. Make sure all the other words in the table are NULL.
4.3 Codes
It is a requirement that your programs, encode and decode, be able to read and write pairs containing variable
bit-length codes. As an example, assume we are going to write the pair (13, ‘a’). What is the minimum number
of bits needed to represent 13? We can easily calculate the minimum number of bits needed to represent any
integer x ≥ 1 using the formula blog2
(x)c+1. The only edge case is with 0, whose bit-length is 1. Thus, we calculate
that the minimum number of bits needed to represent 13, or the bit-length of 13, to be 4. That being said, in
practice, the bit-length of the code in the pair you’re going output wouldn’t be the bit-length of the code itself,
but rather the bit-length of the next available code assignable, as described in the earlier compression example.
The reason for this is because decompression must know at all times the bit-length of the code it is going to read.
Because the dictionary constructed by decompression contains exactly the words and codes contained by the trie
constructed by compression, it is evident that, by using the bit-length of the next available code, that compression
and decompression will agree on the variable bit-lengths of codes. Note that only codes have variable bit-lengths:
symbols in a pair are always 8 bits.
As an example, assume we are going to output the pair (13, ‘a’), and the next available code assignable in our
dictionary is 64. We will need to convert this pair to binary, starting with the code. As stated earlier, the bit-length
of the code is the bit-length of the next available code. We calculate that the bit-length of 64 is 7. We start from
the least significant bit of our code, 13, and construct the code portion of the pair’s binary representation: LSB
© 2021 Darrell Long 6 - MSB. Note that the LSB is on the left and the MSB is on the right. Zeroes are padded because the bitlength
of 64 is 7, while the bit-length of 13 is 4, which is why there are three padded zeroes. In the same fashion, we
start from the least significant bit of our symbol, ‘a’, and add the symbol portion of the pair’s binary representation
to the previous code portion, which yields LSB 101100010000110 MSB, since the ASCII value of ‘a’ is 97.
Now assume that we are reading in the binary LSB 101100010000110 MSB and converting that back into a pair.
We know that the next available code assignable by compression and decompression are the same at every step.
Thus, we know that compression must have output the pair’s code with the bit-length of 64, which is 7. We go
through the first 7 bits of the binary: 1, 0, 1, 1, 0, 0, and 0, summing up the bits as we go, simulating how positional
numbering systems work. This means we perform 1 · 2 - +0 · 2
- +1 · 2
- +1 · 2
- +0 · 2
- +0 · 2
- +0 · 2
- = 13, exactly the
code output by compression. We do the same for the remaining 8 bits to reconstruct the symbol.
code.h ifndef CODE_H
define CODE_H
3
include < inttypes .h >
5
define STOP_CODE 0 // Signals end of decoding / decoding .
define EMPTY_CODE 1 // Code denoting the empty Word .
define START_CODE 2 // Starting code of new Words .
define MAX_CODE UINT16_MAX // Maximum code .
10
endif
4.4 I/O
It is also a requirement that your programs, encode and decode, perform efficient I/O. Reads and writes will be
done 4KB, or a block, at a time, which implicitly requires that you buffer I/O. Buffering is the act of storing data into
a buffer, which you can think of as an array of bytes. You will be implementing an I/O module for this assignment,
and its API is explained in the following sections.
4.4.1 FileHeader- struct FileHeader {
- uint32_t magic ;
- uint16_t protection ;
- };
This is the struct definition for the file header, which contains the magic number for your program and the
protection bit mask for the original file. The file header is the first thing that appears in a compressed file. The
magic number field, magic, serves as a unique identifier for files compressed by encode. decode should only be
able to decompress files which have the correct magic number. This magic number is 0xBAADBAAC.
Before writing the file header to the compressed file using write_header(), you must swap the endianness of the
fields if necessary since interoperability is required. If your program is run on a system using big endian, the fields
© 2021 Darrell Long 7
must be swapped to little endian, since little endian is canonical. A module specifically for handling endianness
will be provided.
4.4.2 int read_bytes(int infile, uint8_t *buf, int to_read)
This will be a useful helper function to perform reads. As you may know, the read() syscall does not always guarantee
that it will read all the bytes specified. For example, a call could be issued to read a block of bytes, but it
might only read half a block. So, we write a wrapper function to loop calls to read() until we have either read all
the bytes that were specified (to_read), or there are no more bytes to read. The number of bytes that were read
are returned. You should use this function whenever you need to perform a read.
4.4.3 int write_bytes(int outfile, uint8_t *buf, int to_write)
This function is very much the same as read_bytes(), except that it is for looping calls to write(). As you may
imagine, write() isn’t guaranteed to write out all the specified bytes (to_write), and so we loop until we have
either written out all the bytes specified, or no bytes were written. The number of bytes written out is returned.
You should use this function whenever you need to perform a write.
4.4.4 void read_header(int infile, FileHeader *header)
This reads in sizeof(FileHeader) bytes from the input file. These bytes are read into the supplied header. Endianness
is swapped if byte order isn’t little endian. Along with reading the header, it must verify the magic number.
4.4.5 void write_header(int outfile, FileHeader *header)
Writes sizeof(FileHeader) bytes to the output file. These bytes are from the supplied header. Endianness is
swapped if byte order isn’t little endian.
4.4.6 bool read_sym(int infile, uint8_t *sym)
An index keeps track of the currently read symbol in the buffer. Once all symbols are processed , another block is
read. If less than a block is read , the end of the buffer is updated. Returns true if there are symbols to be read , false
otherwise.
4.4.7 void write_pair(int outfile, uint16_t code, uint8_t sym, int bitlen)
“Writes” a pair to outfile. In reality, the pair is buffered. A pair is comprised of a code and a symbol. The bits of
the code are buffered first, starting from the LSB. The bits of the symbol are buffered next, also starting from the
LSB. The code buffered has a bit-length of bitlen. The buffer is written out whenever it is filled.
4.4.8 void flush_pairs(int outfile)
Writes out any remaining pairs of symbols and codes to the output file.
4.4.9 bool read_pair(int infile, uint16_t code, uint8_t sym, int bitlen)
“Reads” a pair (code and symbol) from the input file. The “read” code is placed in the pointer to code (e.g. *code
= val) The “read” symbol is placed in the pointer to sym (e.g. *sym = val). In reality, a block of pairs is read into a
buffer. An index keeps track of the current bit in the buffer. Once all bits have been processed, another block is
read. The first bitlen bits are the code, starting from the LSB. The last 8 bits of the pair are the symbol, starting
from the LSB. Returns true if there are pairs left to read in the buffer, else false. There are pairs left to read if the
read code is not STOP_CODE.
© 2021 Darrell Long 8
4.4.10 void write_word(int outfile, Word *w)
“Writes” a pair to the output file. Each symbol of the Word is placed into a buffer. The buffer is written out when it
is filled.
4.4.11 void flush_words(int outfile)
Writes out any remaining symbols in the buffer to the outfile.
Note that the output file in which you write to must have the same protection bits as the original file. Like in
assignment 4, you will make use of fstat() and fchmod().
All reads and writes in this program must be done using the system calls read() and write(), which means that
you must use the system calls open() and close() to get your file descriptors. As stated earlier, all reads and
writes must be performed in efficient blocks of 4KB. You will want to use two static 4KB uint8_t arrays to serve as
buffers: one to store binary pairs and the other to store characters. Each of these buffers should have an index, or
a variable, to keep track of the current byte or bit that has been processed. - Program Options
Your encode program must support the following getopt() options:
• -v : Print compression statistics to stderr.
• -i : Specify input to compress (stdin by default)
• -o - Compression
The following steps for compression will refer to the input file descriptor to compress as infile and the compressed
output file descriptor as outfile. - Open infile with open(). If an error occurs, print a helpful message and exit with a status code indicating
that an error occurred. infile should be stdin if an input file wasn’t specified.
© 2021 Darrell Long 9 - The first thing in outfile must be the file header, as defined in the file io.h. The magic number in the
header must be 0xBAADBAAC. The file size and the protection bit mask you will obtain using fstat(). See
the man page on it for details. - Open outfile using open(). The permissions for outfile should match the protection bits as set in your
file header. Any errors with opening outfile should be handled like with infile. outfile should be
stdout if an output file wasn’t specified. - Write the filled out file header to outfile using write_header(). This means writing out the struct itself
to the file, as described in the comment block of the function. - Create a trie. The trie initially has no children and consists solely of the root. The code stored by this root trie
node should be EMPTY_CODE to denote the empty word. You will need to make a copy of the root node and
use the copy to step through the trie to check for existing prefixes. This root node copy will be referred to as
curr_node. The reason a copy is needed is that you will eventually need to reset whatever trie node you’ve
stepped to back to the top of the trie, so using a copy lets you use the root node as a base to return to. - You will need a monotonic counter to keep track of the next available code. This counter should start at
START_CODE, as defined in the supplied code.h file. The counter should be a uint16_t since the codes
used are unsigned 16-bit integers. This will be referred to as next_code. - You will also need two variables to keep track of the previous trie node and previously read symbol. We will
refer to these as prev_node and prev_sym, respectively. - Use read_sym() in a loop to read in all the symbols from infile. Your loop should break when read_sym()
returns false. For each symbol read in, call it curr_sym, perform the following:
(a) Set next_node to be trie_step(curr_node, curr_sym), stepping down from the current node to
the currently read symbol.
(b) If next_node is not NULL, that means we have seen the current prefix. Set prev_node to be curr_node
and then curr_node to be next_node.
(c) Else, since next_node is NULL, we know we have not encountered the current prefix. We write the pair
(curr_node->code, curr_sym), where the bit-length of the written code is the bit-length of next_code.
We now add the current prefix to the trie. Let curr_node->children[curr_sym] be a new trie node
whose code is next_code. Reset curr_node to point at the root of the trie and increment the value of
next_code.
(d) Check if next_code is equal to MAX_CODE. If it is, use trie_reset() to reset the trie to just having the
root node. This reset is necessary since we have a finite number of codes.
(e) Update prev_sym to be curr_sym. - After processing all the characters in infile, check if curr_node points to the root trie node. If it does not,
it means we were still matching a prefix. Write the pair (prev_node->code, prev_sym). The bit-length of the
code written should be the bit-length of next_code. Make sure to increment next_code and that it stays
within the limit of MAX_CODE. Hint: use the modulo operator. - Write the pair (STOP_CODE, 0) to signal the end of compressed output. Again, the bit-length of code written
should be the bit-length of next_code. - Make sure to use flush_pairs() to flush any unwritten, buffered pairs. Remember, calls to write_pair()
end up buffering them under the hood. So, we have to remember to flush the contents of our buffer. - Use close() to close infile and outfile.
© 2021 Darrell Long 10 - Decompression
The following steps for decompression will refer to the input file to decompress as infile and the uncompressed
output file as outfile. - Open infile with open(). If an error occurs, print a helpful message and exit with a status code indicating
that an error occurred. infile should be stdin if an input file wasn’t specified. - Read in the file header with read_header(), which also verifies the magic number. If the magic number is
verified then decompression is good to go and you now have a header which contains the original protection
bit mask. - Open outfile using open(). The permissions for outfile should match the protection bits as set in
your file header that you just read. Any errors with opening outfile should be handled like with infile.
outfile should be stdout if an output file wasn’t specified. - Create a new word table with wt_create() and make sure each of its entries are set to NULL. Initialize the
table to have just the empty word, a word of length 0, at the index EMPTY_CODE. We will refer to this table as
table. - You will need two uint16_t to keep track of the current code and next code. These will be referred to as
curr_code and next_code, respectively. next_code should be initialized as START_CODE and functions
exactly the same as the monotonic counter used during compression, which was also called next_code. - Use read_pair() in a loop to read all the pairs from infile. We will refer to the code and symbol from each
read pair as curr_code and curr_sym, respectively. The bit-length of the code to read is the bit-length of
next_code. The loop breaks when the code read is STOP_CODE. For each read pair, perform the following:
(a) As seen in the decompression example, we will need to append the read symbol with the word denoted
by the read code and add the result to table at the index next_code. The word denoted by the
read code is stored in table[curr_code]. We will append table[curr_code] and curr_sym using
word_append_sym().
(b) Write the word that we just constructed and added to the table with write_word(). This word should
have been stored in table[next_code].
(c) Increment next_code and check if it equals MAX_CODE. If it has, reset the table using wt_reset() and
set next_code to be START_CODE. This mimics the resetting of the trie during compression. - Flush any buffered words using flush_words(). Like with write_pair(), write_word() buffers words
under the hood, so we have to remember to flush the contents of our buffer. - Close infile and outfile with close().
- Deliverables
You will need to turn in: - Makefile:
• CFLAGS=-Wall -Wextra -Werror -Wpedantic must be included.
• CC=clang must be specified.
• make clean must remove all files that are compiler generated.
• make encode should build your encode program.
• make decode should build your decode program.
© 2021 Darrell Long 11
• make should build both encode and decode, as should make all.
• Your programs should have no memory leaks. - Your programs must have the following source and header files:
• encode.c : contains the main() function for the encode program.
• decode.c : contains the main() function for the decode program.
• trie.c: the source file for the Trie ADT.
• trie.h: the header file for the Trie ADT. You must not modify this file.
• word.c: the source file for the Word ADT.
• word.h: the header file for the Word ADT. You must not modify this file.
• io.c: the source file for the I/O module.
• io.h: the header file for the I/O module. You must not modify this file.
• endian.h: the header file for the endianness module. You must not modify this file.
• code.h: the header file containing macros for reserved codes. You must not modify this file. - You may have other source and header files, but do not try to be overly clever.
- Running scan-build yields no bugs. If there are any false positives, make sure to explain why they are false
positives in your README.md. - README.md: This must be in Markdown. This should describe how to use your program and Makefile. This
also contains any explanations for complaints generated by scan-build. - DESIGN.pdf: This must be a PDF. The design document should describe your design for your program with
enough detail that a sufficiently knowledgeable programmer would be able to replicate your implementation.
This does not mean copying your entire program in verbatim. You should instead describe how your
program works with supporting pseudo-code. - Working encode and decode programs, along with test files, and other source files will be supplied in the
resources repository. - Submission
To submit your assignment, refer back to assignment0 for the steps on how to submit your assignment through
git. Remember: add, commit, and push!
Your assignment is turned in only after you have pushed. If you forget to push, you have not turned in your assignment
and you will get a zero. “I forgot to push” is not a valid excuse. It is highly recommended to commit and push
your changes often.
© 2021 Darrell Long 12
A LZ78 Algorithm Pseudocode
A.1 Compression
COMPRESS(infile, outfile) - root = TRIE_CREATE()
- curr_node = root
- prev_node = NULL
- curr_sym = 0
- prev_sym = 0
- next_code = START_CODE
- while READ_SYM(infile, &curr_sym) is TRUE
- next_node = TRIE_STEP(curr_node, curr_sym)
- if next_node is not NULL
- prev_node = curr_node
- curr_node = next_node
- else
- WRITE_PAIR(outfile, curr_node.code, curr_sym, BIT-LENGTH(next_code))
- curr_node.children[curr_sym] = TRIE_NODE_CREATE(next_code)
- curr_node = root
- next_code = next_code + 1
- if next_code is MAX_CODE
- TRIE_RESET(root)
- curr_node = root
- next_code = START_CODE
- prev_sym = curr_sym
- if curr_node is not root
- WRITE_PAIR(outfile, prev_node.code, prev_sym, BIT-LENGTH(next_code))
- next_code = (next_code +1) % MAX_CODE
- WRITE_PAIR(outfile, STOP_CODE, 0, BIT-LENGTH(next_code))
- FLUSH_PAIRS(outfile)
A.2 Decompression
DECOMPRESS(infile, outfile) - table = WT_CREATE()
- curr_sym = 0
- curr_code = 0
- next_code = START_CODE
- while READ_PAIR(infile, &curr_code, &curr_sym, BIT-LENGTH(next_code)) is TRUE
- table[next_code] = WORD_APPEND_SYM(table[curr_code], curr_sym)
- WRITE_WORD(outfile, table[next_code])
- next_code = next_code + 1
- if next_code is MAX_CODE
- WT_RESET(table)
- next_code = START_CODE
- FLUSH_WORDS(outfile)
© 2021 Darrell Long 13