Home > Uncategorized, unix > Unix tip #3: Introduction to Find, Grep, Sed

Unix tip #3: Introduction to Find, Grep, Sed

I’ve written a few times before about Unix command line tools and how learning them can make you a more efficient programmer.  Today I’m going to introduce a few essential tools in the Unix toolkit.  While programming, one often notes future improvements or tasks with the use of a TODO comment.  For instance, if you have a dummy implementation of a method, you might comment that you need to fill in the actual implementation later:

 public int randomValue() {
 // TODO: hook up the actual random number generator
 return 0;

The problem is that these TODOs more often than not get ignored, especially if you have to search through the code yourself to try to find all of the remaining tasks.  Fortunately, certain Programs (NetBeans and TextMate for two examples) can find instances of keywords indicating a task, extract the comments, and present them to you in a nice table view.

I’m going to step through the use of a few Unix tools that can be tied together to extract the data and create a similar view.  In particular I will illustrate the use of find, grep, sed,  and pipes.

The general steps I’ll be presenting are:

Step Tools used
1. Find all Java files find
2. Find each TODO item grep
3. Extract filename, line number, task sed
4. Format results of step 3 as an HTML table find/grep/sed/shell script


Finding instances of text with grep

In order to extract all of the TODO items from within our java files, we need a way of searching for matching text. grep is the tool to do that. Grep takes as input a list of files to search and a pattern to try to match against; it will then emit a set of lines matching the pattern.

For instance, to search for TODO or any version of that string (todo, ToDO), in all the .java files in the current directory, you would execute the following:

grep -i TODO *.java
Telephone.java:    // TODO: Document
Telephone.java:     // TODO: throw exception if precondition is violated

Note that the line numbers are omitted. If we want them, we use the -n command

grep -i -n TODO *.java
Telephone.java:20:    // TODO: Document
Telephone.java:29:     // TODO: throw exception if precondition is violated

If all we want to do is get a rough estimate as to how many documented TODOs we have, we can pipe the result of this argument into the wc utility, which counts bytes, characters, or lines. We want the number of lines.

grep -i -n TODO *.java | wc -l

This works fine with a single directory of files, but it will not handle nested directories. For instance, if my directory structure looks like the following:

|-- BalancedTernary.java
`-- Telephone.java

0 directories, 2 files

All of these files will be searched when grep is run. But if I introduce new files in subdirectories:

mkdir Subdir
echo "//TODO: Create this file" > Subdir/Test.java

|-- BalancedTernary.java
|-- Subdir
|   `-- Test.java
`-- Telephone.java

1 directory, 3 files

The new Test.java will not be searched. In order make grep search through all of the subdirectories (i.e., recursively), you can combine grep with another extremely useful Unix utility, find. Before moving on to find, I want to stress that grep is extremely useful and vital to anyone using a Unix based machine. See grep tutorials for many good examples of how to use grep.

Finding files with find

The find command is extremely useful. The man page describes find as

find – search for files in a directory hierarchy

There are a lot of arguments you can use, but to get started, the basic syntax is

find [<starting location>] -name <name pattern>

If the starting location is not provided, it is assumed to be in the current directory (. in Unix terms). In all the examples that follow I will explicitly list the starting directory.

For instance, if we want to find all the files that end with the extension “.java” in the current working directory, we could run the following:

find . -name "*.java"

Note that we must enclose the pattern in quotes in this example in order to prevent the shell from trying to expand the * wildcard. If we don’t, the shell will convert the asterisk into a space delimited set of all the files/directories in the current folder, which will lead to an error

 find . -name *.java # expands to find . -name BalancedTernary.java Telephone.java
find: Telephone.java: unknown option

Just as we can use the wc command to count the number of times a phrase appears in a file, we can use it to count the number of files matching a given pattern. That is because find outputs each matching file path to a separate line. Thus if we wanted to count the number of java files in all folders rooted in the current folder, we could do

find . -name "*.java" | wc -l

While I have only presented the -name flag, there are numerous other flags as well, such as whether the candidate file is a file or directory (-type f or -type d respectively), whether the match is smaller, the same, or bigger than a given size (-size +100M == bigger than 100 megabytes), or when the file was last modified (find -newer ordinary_file would only accept files that have a modification time newer than that of ordinary_file). A A great article for gaining more expertise is Mommy I found it! – 15 practical unix find commands.

Combining find with other commands

find becomes even more powerful when combined with the -exec option, which allows you to execute arbitrary commands for each file that matches the pattern. The syntax for doing that looks like

find [<starting location>] -name <name pattern> -exec <commands> {} \;

where the file path will be substituted for the {} characters. For instance, if we want to count the number of lines in each Java file, we could run

find . -name "*.java" -exec wc -l {} \;
      23 ./BalancedTernary.java
       1 ./Subdir/Test.java
      88 ./Telephone.java

This has precisely the same effect as if we explicitly executed the wc -l command ourselves:

wc -l ./BalancedTernary.java wc -l ./Subdir/Test.java wc -l ./Telephone.java

As another example, we could backup all of the Java files in the directory by copying them and appending the suffix .bk to each

find . -name "*.java" -exec cp {} {}.bk \;
Nick@Macintosh-3 ~/Desktop/Programming/Java/example$ ls
BalancedTernary.java    Subdir                  Telephone.java.bk
BalancedTernary.java.bk Telephone.java

To undo this, we could remove all of the files ending in .bk:

find . -name “*.bk” -exec rm {} \;

Combining find and grep

Since I started the article talking about grep, it’s only natural that you can combine grep with find, and it often pays to do so.

For instance, by combining the earlier grep command to find all TODO items with the find command to find all java files, we suddenly have a command which will traverse an arbitrarily nested directory structure and search all the files we are interested in.

find . -name "*.java" -exec grep -i -n TODO {}  \;
1://todo: Create this file
20:    // todo: Document
29:     // todo: throw exception if precondition is violated

Note that we no longer have the filename prepended to the output; if we want it back we can add the -H flag.

find . -name "*.java" -exec grep -Hin TODO {} \;
./Subdir/Test.java:1://todo: Create this file
./Telephone.java:20:    // todo: Document
./Telephone.java:29:     // todo: throw exception if precondition is violated

In this last snippet I have combined the individual -H, -i and -n flags together into the shorter -Hin; this works identically as listing them separately. (Not all Unix commands work this way; check the man page if you’re unsure).

An alternate exec terminator: Performance considerations

I said earlier that the basic syntax for combining find with other commands is

find [<starting location>] -name <name pattern> -exec <commands> {} \;

The ; terminates the exec clause, but because it can be interpreted as text, it has to be backslash escaped. While researching this article I found a Unix/Linux “find” Command Tutorial that introduced me to an alternative syntax for terminating the -exec clause of the find command. By replacing the semicolon with a + sign, files are grouped together in batches and sent to the given command rather than executed one at a time. Let me illustrate:

# Executes the 'echo' command on each file individually
find . -exec echo {} \;

# Executes the 'echo' command on bundled groups of files
find . -exec echo {} +
. ./BalancedTernary.java ./Subdir ./Subdir/Test.java ./table.html ./Telephone.java ./test.a

This technique of grouping the files together can have a profound performance boost when used with commands that can handle space terminated arguments. For instance:

time find /Applications/ -name "*.java" -exec grep -i TODO {} \;
real    1m36.458s
user    0m3.912s
sys 0m10.933s

time find /Applications/ -name "*.java" -exec grep -i TODO {} +
real    0m39.060s
user    0m3.660s
sys 0m6.571s

# An alternate way of executing grep on batches of files at once #
time find /Applications/ -name "*.java" -print0 | xargs -0 grep -i "TODO"
real    0m50.486s
user    0m4.230s
sys 0m7.924s

By replacing the semicolon with the plus sign, I gained almost a 2.5x speed increase. Again, this will only work with commands that correctly handle whitespace separated arguments; the previous example with copy would fail miserably, because cp expects a single src/destination pair

# Will not work!
find . -name "*.java" -exec cp {} {}.bk +

Converting results of find/grep into table form – Intro to sed, cut, and basename

In the last section, I showed how to combine find and grep. The output of the command will look something like this:

find . -name "*.java" -exec grep -Hin TODO {} +
./Subdir/Test.java:1://todo: Create this file
./Telephone.java:20:    // todo: Document
./Telephone.java:29:     // todo: throw exception if precondition is violated

The output has the path to the file, followed by a semicolon, followed by the matching line in the input file that had the TODO in it. Let’s mimic the output of the TODO list in TextMate, which simply displayed a two column table with File name and line number followed by the extracted comment. While we could use any programming language to do this text manipulation (Python springs to mind), I’m going to use a combination of sed and shell scripts to illustrate a few more powerful command line tools.

Recall that the output of our script so far looks like the following:

./Telephone.java:20: // todo: Document

In other words each line is in the form

relative/path/to/File:lineNumber:todo text

The colons delimiting the text allow us to split the constituent parts very easily. The command to do that is cut. With cut you specify the delimiter on which to split the text, and then which numbered fields you want (where fields are numbered 1 .. n)

As an example, here is code to extract the path (the first column of text):

find . -name "*.java" -exec grep -Hin TODO {} + | cut -d ":" -f 1

This gives us the path, one per line. If we want to convert the relative path into just the name of the file, like the TextMate example does, we want to strip out all of the leading directories, leaving just the file name. While we could code up a regular expression to perform the substitution, I prefer to avoid doing more work than I need to. Instead I’ll use the basename command, which does that for us.

find . -name "*.java" -exec grep -Hin TODO {} + | basename `cut -d ":" -f 1`

The line number, the second column of text, is just as easy to extract.

find . -name “*.java” -exec grep -Hin TODO {} + | cut -d “:” -f 2 1 20 29

The fact that the line of text extracted by grep could contain the colon character (and often will; I always write my TODOs as TODO: do x) means we have to be a bit smarter about how we use cut. If we assume that the text is just in the third column, we will lose the text if there are colons.

# Only taking the third column
echo "./Telephone.java:20:    // todo: Document" | cut -d ":" -f 3    // todo
# Taking all columns after and including the third column
echo "./Telephone.java:20:    // todo: Document" | cut -d ":" -f 3-
    // todo: Document

While this works, it’s not the neatest output. In particular we want to get rid of the leading white space; otherwise it will mess up the formatting in the HTML table. Performing text substitution is the job of the sed tool. sed stands for stream editor and it is capable of doing extremely heavy duty find and replace tasks. I don’t pretend to be an expert with sed and this article won’t make you one either, but hopefully I can at least illustrate its usefulness. For a more in depth tutorial, see Sed – An Introduction and Tutorial.

A common use case for sed, as I mentioned, is to replace text. The general pattern is

sed ‘s/regexpToReplace/textToReplaceItWith/[g]’

The s can be read as “substitute”, and the optional g stands for global. If you omit it, it will only replace the first instance of the regular expression match that it finds. The g makes it search for all matches in the text.

Thus to remove leading white space, we can use the expression sed ‘s/^[ <tab>]*//g’

where the ^ character indicates that it must match the start of the line, and the text within brackets are the characters that will be matched by the regular expression. The * means to match zero or more instances. In other words, this line says “match the start of the string and all spaces and tabs you can until reaching other text, and replace it with nothing”.

The above command is not strictly correct. We need to indicate to sed that we want to replace the tab character. Unlike many Unix utilities, sed does not allow you to use the character sequence \t to indicate the tab character. Instead you need a literal tab at that place in the command. The problem with doing this is that your shell might swallow the tab before it gets to the sed command. In bash, the default shell environment on the Mac, the tab key is interpreted as a command to auto complete what is being typed. If you press the tab key twice, the shell will print out all the possible autocompletions.

For instance,

lp           lpc          lpmove       lppasswd     lpr          lprsetup.sh
lpadmin      lpinfo       lpoptions    lpq          lprm         lpstat

Here I started typing lp, hit tab twice, and the shell produced a list of all the commands it knew about (technically, that are on the PATH environment variable). So we need a way to smuggle the tab key into the sed command, without triggering the shell’s autocompletion. The way to do this is with the “verbatim” command sequence, which instructs the shell not to interpret certain commands and instead to pass them treat them verbatim, as text.

To enter this temporary verbatim mode, you press Ctrl V (sometimes indicated as ^V online) followed by the key combination you want treated as text. Thus the real sed command to remove leading white space is sed ‘s/^[ ]*//’

$ sed 's/^[    ]*//'
           tabs and spaces
tabs and spaces

The above snippet illustrates that sed reads from standard input by default and thus can be used interactively to test the replacements you have specified. Again, in the above text it looks like I have a string of spaces, but it’s really <space><ctrl v><tab> within the brackets. From here on out I will put a \t to indicate a tab but you should realize that you need to do the ctrl v tab sequence I just described instead.

(Aside: I have read online that some versions of sed actually do support the \t character sequence to indicate tabs, but the default sed shipping with Mac OSX does not.)

sed – combine multiple commands into one

If you have series of text replacements you want to do using sed, you can either pipe the chain of transformations you want to do from one sed invocation to another, or you can use the -e flag to chain them together.

echo "hello world" | sed 's/hello/goodbye/' | sed 's/world/frank/'
goodbye frank
echo "hello world" | sed -e 's/hello/goodbye/' -e 's/world/frank/'goodbye frank

Note that you need the -e immediately after the first sed pattern as well; I naively tried to do

echo "hello world" | sed 's/hello/goodbye/' -e 's/world/frank/'sed: -e: No such file or directory
sed: s/world/frank/: No such file or directory

Integrating sed with find and grep

Combining all of the above sed goodness with the previous code we have

find . -name "*.java" -exec grep -Hin TODO {} + | cut -d ":" -f 3- | sed 's/^[ \t]*//'
//todo: Create this file
// todo: Document
// todo: throw exception if precondition is violated

I don’t want the todo text in the comments, as it would be redundant. As such I will remove the double slashes followed by any white space followed by todo, followed by an optional colon, followed by any space.

find . -name "*.java" -exec grep -Hin TODO {} + | cut -d ":" -f 3- | sed -e 's/^[ \t]*//' -e 's/[\/*]*[ \t]*//' -e 's/TODO/todo/' -e 's/todo[:]*[ \t]*//'
 Create this file
 throw exception if precondition is violated

This can be read as

s/^[ \t]*//         remove leading whitespace
s/[\/*]*            remove any number of forward slashes (/) or stars (*), which indicate the start of a comment
[ \t]*              remove whitespace
s/TODO/todo         convert uppercase TODO string into lower case
todo                remove the literal string 'todo'
[:]*                remove any colons that exist
[ \t]*              remove whitespace

We now have all the pieces we need to create our script.

Putting it all together

I’m going to show the script in its entirety without a huge amount of explanation. This post is more about the use of find/grep/sed than it is about shell scripting. I don’t claim to be an expert at writing shell scripts, so I wouldn’t be surprised if there’s a better way to do some of the following. It is not perfect; as the comments indicate, it wouldn’t handle text like ToDo correctly in the sed command. More importantly, there are some false positives in the lines it returns: things like toDouble match, because it contains the string ‘todo’. I’ll leave such improvements to the reader; if you do have any suggestions for the script, please add them to the comments below.


# From http://www.linuxweblog.com/bash-argument-numbers-check
if [ $# -gt $EXPECTED_ARGS ]
  echo "Usage: ./extract [starting_directory]" >&2
  exit $E_BADARGS

# By default, start in the current working directory, but if they provide
# an argument, use that instead.
if [ $# -eq $EXPECTED_ARGS ]

# Start creating the HTML document
echo "<html><head></head><body>"
echo "<table border=1>"
echo "<tr><td>Location</td><td>Comment</td></tr>"

# The output of the find command will look like
# ./Telephone.java:20:    // todo: Document

find $startingDir -name "*.java" -exec grep -Hin todo {} + |
# Allows the script to read in piped in arguments
while read data; do

    # The location of the file is the first argument
    fileLoc=`echo "$data" | cut -d ":" -f 1`
    fileName=`basename $fileLoc`

    # the line number is the second
    lineNumber=`echo "$data" | cut -d ":" -f 2`

    # all arguments after the second colon are the comment.  Eliminate the TODO
    # text with a simple find and replace.
    # Note: only handles todo and TODO, would need some more logic to handle other cases
    comment=`echo "$data" | cut -d ":" -f 3- | sed -e 's/^[     ]*//' -e 's/[\/*]*[     ]*//' -e 's/TODO/todo/' -e 's/todo[:]*[     ]*//'`
    echo "<tr>"
    echo "  <td><a href="$fileLoc">$fileName ($lineNumber)</a></td>"
    echo "  <td>$comment</td>"
    echo "</tr>"

# Finish off the HTML document
echo "</table>"
echo "</body></html>"

exit 0

If you save this script as a .sh file, you will need to make it executable before you can run it. From the terminal:

chmod +x extract.sh
# Extract all the TODO comments in the Applications folder, and save it as an html table
# Redirect the printed HTML to an HTML document
./extract.sh /Applications > table.html

The source code for the script is available on github. Running the script in my /Applications directory leads to the following HTML table:

Location Comment
Aquamacs (629) return ((ObjectReference)val).toString(); //
Aquamacs (633) return val.toString(); // not correct in all cases
Cycling (11) support joint operations on more than one channel.
Cycling (27) what about objects with more than one input?
Cycling (36) improve feedback math — fixed point, like jit.wake?
Cycling (277) theta shift?
Cycling (349) double closest[] = new double[] {a[0].toDouble(), a[1].toDouble(), a[2].toDouble()};
Cycling (351) double farthest[] = new double[] {a[0].toDouble(), a[1].toDouble(), a[2].toDouble()};
Cycling (5) describe the class
Cycling (22) implement with a Vector to improve performance
Cycling (8) abort a thread if an incoming message arrives before completion
Cycling (8) have the search happen in a separate thread
Cycling (9) possible to separate the errors that results from not
Cycling (191) implement automatic replacement of shader name in prototype file
PGraphicsOpenGL.java (738) make this more efficient and just update a sub-part
PGraphicsOpenGL.java (1165) P3D overrides box to turn on triangle culling, but that’s a waste
PGraphicsOpenGL.java (1180) P3D overrides sphere to turn on triangle culling, but that’s a waste
PGraphicsOpenGL.java (1508) Should instead override textPlacedImpl() because createGlyphVector
PGraphicsOpenGL.java (2207) this expects a fourth arg that will be set to 1
PGraphicsOpenGL.java (2847) not optimized properly, creates multiple temporary buffers
PGraphicsOpenGL.java (2858) is this possible without intbuffer?
PGraphicsOpenGL.java (2870) remove the implementation above and use setImpl instead,
PGraphicsOpenGL.java (2978) – extremely slow and not optimized.
PGraphicsOpenGL.java (738) make this more efficient and just update a sub-part
PGraphicsOpenGL.java (1165) P3D overrides box to turn on triangle culling, but that’s a waste
PGraphicsOpenGL.java (1180) P3D overrides sphere to turn on triangle culling, but that’s a waste
PGraphicsOpenGL.java (1508) Should instead override textPlacedImpl() because createGlyphVector
PGraphicsOpenGL.java (2207) this expects a fourth arg that will be set to 1
PGraphicsOpenGL.java (2847) not optimized properly, creates multiple temporary buffers
PGraphicsOpenGL.java (2858) is this possible without intbuffer?
PGraphicsOpenGL.java (2870) remove the implementation above and use setImpl instead,
PGraphicsOpenGL.java (2978) – extremely slow and not optimized.

The complete result can be found as another github gist.

Quick note: You have to be careful about what you echo in the shell. In an early version, I forgot to surround the text ($data) with quotes. This led to a problem when there were asterisks in the text, since the shell expanded the star into a list of all the files in the directory (aka file globbing). This is a relatively harmless problem; had the line had something like rm * instead, it would have been devastating. So make sure you surround your output text in quotes!

$ echo *
ApplicationTODO.html BlogPost.mkdown Find text.mkdown PGraphicsOpenGL.java TabTodo.java Test.html TodoTest.java appTable.html extract.sh tab tab.txt table body.html table.awk table.html table1.html test.java
$ echo "*"


I have introduced the find command and how it can be used to locate files or directories on disk with certain properties (name, last modified date, etc). I then showed how grep can be used to search the contents of a file or stream of content for matching regular expressions. Next I showed you how to combine find with arbitrary Unix commands, including grep with the -exec option. Finally I tied all these concepts together by creating a simple script which searches through all of the java files in a directory for those lines that have TODO in them, and creates an HTML table summarizing the location of each of these tasks, alongside the TODO item text.

Categories: Uncategorized, unix Tags: , , , , ,
  1. October 3, 2010 at 7:31 pm

    Possibly the best read I have read all day =D

    Fondest regards

  2. September 25, 2013 at 1:56 am

    Howdy, i read your blog occasionally and i own a similar one and i
    was just curious if you get a lot of spam remarks?
    If so how do you reduce it, any plugin or anything you can suggest?
    I get so much lately it’s driving me mad so any assistance
    is very much appreciated.

  1. September 8, 2010 at 12:13 am
  2. October 5, 2010 at 3:59 pm
  3. October 11, 2010 at 3:20 pm
  4. October 8, 2015 at 3:59 pm

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: