Contents

Introduction
Credits
Substitute Command
2.1 Search & Replace
2.2 Line Ranges & Addressing
Pattern Description
3.1 Anchors
3.2 "Escaped" characters or metacharacters
3.3 Quantifiers, Greedy and Non-Greedy
3.4 Character ranges
3.5 Grouping and Backreferences
3.6 Alternations
3.7 Operator Precedence
Global Command
4.1 Global search and execution
4.2 Examples
Examples
5.1 Tips & Techniques
5.2 Creating Outline
5.3 Working with Tables
Other Regexp Flavors
Links

I started this tutorial for one simple reason - I like regular expressions. Nothing compares to the satisfaction from a well-crafted regexp which does exactly what you wanted it to do :-). And yes, I have a life too. I hope it's passable as a foreword. Feel free to send me your comments, corrections and suggestions concerning this tutorial.

Speaking more seriously, regular expressions (or regexps for short) are tools used to manipulate text and data. They don't exist as a standalone product but usually are a part of some program/utility. The most well known example is UNIX grep, a program to search files for lines that match certain pattern. The search pattern is described in terms of regular expressions. You can think of regexps as a specialized pattern language. Regexps are quite useful and can greatly reduce amount of time required to do some tedious text editing.

Note: Regexp terminology is largely borrowed from Jeffrey Friedl "Mastering Regular Expressions".

Credits

Many thanks to everybody (especially Benji Fisher & Zdenek Sekera) who sent their corrections and suggestions.

2.1 Search & Replace

So, what can you do with regular expressions? The most common task is to make replacements in a text according to certain rules. For this tutorial you need to know VIM search and replace command (S&R) :substitute. Here is an excerpt from VIM help:

Part of the command word enclosed in the "[" & "]" can be omitted.

2.2 Range of Operation, Line Addressing and Marks

Before I begin with a pattern description let's talk about line addresses in Vim. Some Vim commands can accept a line range in front of them. By specifying the line range you restrict the command execution to this particular part of text only. Line range consists of one or more line specifiers, separated with a comma or semicolon. You can also mark your current position in the text typing ml , where "l" can be any letter, and use it later defining the line address.

If no line range is specified the command will operate on the current line only.

Here are a few examples:

10,20

- from 10 to 20 line.

Each may be followed (several times) by "+" or "-" and an optional number. This number is added or subtracted from the preceding line number. If the number is omitted, 1 is used.

/Section 1/+,/Section 2/-

- all lines between Section 1 and Section 2, non-inclusively, i.e. the lines containing Section 1 and Section 2 will not be affected.

The /pattern/ and ?pattern? may be followed by another address separated by a semicolon. A semicolon between two search patterns tells Vim to find the location of the first pattern, then start searching from that location for the second pattern.

/Section 1/;/Subsection/+,/Subsection/-

- first find Section 1, then the first line with Subsection, step one line down (beginning of the range) and find the next line with Subsection, step one line up (end of the range).

The next example shows how you can reuse you search pattern:

:/Section/+ y

- this will search for the Section line and yank (copy) one line after into the memory.

:// normal p

- and that will search for the next Section line and put (paste) the saved text on the next line.

3.1 Anchors

Suppose you want to replace all occurrences of vi with VIM. This can be easily done with

s/vi/VIM/g

If you've tried this example then you, no doubt, noticed that VIM replaced all occurrences of vi even if it's a part of the word (e.g. navigator). If we want to be more specific and replace only whole words vi then we need to correct our pattern. We may rewrite it by putting spaces around vi:

s: vi : VIM :g

But it will still miss vi followed by the punctuation or at the end of the line/file. The right way is to put special word boundary symbols "\<" and "\>" around vi.

s:\<vi\>:VIM:g

The beginning and the end of the line have their own special anchors - "^" and "$", respectively. So, for all vi only at the start of the line:

s:^vi\>:VIM:

To match the lines where vi is the only word:

s:^vi$:VIM:

Now suppose you want to replace not only all vi but also Vi and VI. There are several ways to do this:

• probably the simplest way is to put "i" - ignore case in a pattern %s:vi:VIM:gi
• define a class of characters. This is a sequence of characters enclosed by square brackets "[" and "]". It matches any character from this set. So :%s:[Vv]i:VIM: will match vi and Vi. More on character ranges in the following section.

3.2 "Escaped" characters or metacharacters

So far our pattern strings were constructed from normal or literal text characters. The power of regexps is in the use of metacharacters. These are types of characters which have special meaning inside the search pattern. With a few exceptions these metacharacters are distinguished by a "magic" backslash in front of them. The table below lists some common VIM metacharacters.

So, to match a date like 09/01/2000 you can use

\d\d/\d\d/\d\d\d\d

To match 6 letter word starting with a capital letter

\u\w\w\w\w\w

Obviously, it is not very convenient to write \w for any character in the pattern - what if you don't know how many letters in your word? This can be helped by introducing so-called quantifiers.

3.3 Quantifiers, Greedy and Non-Greedy

Using quantifiers you can set how many times certain part of you pattern should repeat by putting the following after your pattern:

Now it's much easier to define a pattern that matches a word of any length \u\w\+.

These quantifiers are greedy - that is your pattern will try to match as much text as possible. Sometimes it presents a problem. Let's consider a typical example - define a pattern to match delimited text, i.e. text enclosed in quotes, brackets, etc. Since we don't know what kind of text is inside the quotes we'll use

/".*"/

But this pattern will match everything between the first " and the last " in the following line:

this file is normally "$VIM/.gvimrc". You can check this with ":version".

This problem can be resolved by using non-greedy quantifiers:

Let's use \{-} in place of * in our pattern. So, now ".\{-}" will match the first quoted text:

this file is normally "$VIM/gvimrc". You can check this with ":version".

.\{-} pattern is not without surprises. Look what will happen to the following text after we apply:

:s:.\{-}:_:g

Before:

n and m are decimal numbers between

After:

_n_ _a_n_d_ _m_ _a_r_e_ _d_e_c_i_m_a_l_ _n_u_m_b_e_r_s_ _b_e_t_w_e_e_n_

"As few as possible" applied here means zero character replacements. However match does occur between characters! To explain this behavior I quote Bram himself:

Matching zero characters is still a match. Thus it will replace zero characters with a "_". And then go on to the next position, where it will match again.

It's true that using "\{-}" is mostly useless. It works this way to be consistent with "*", which also matches zero characters. There are more useless ones: "x\{-1,}" always matches one x. You could just use "x". More useful is something like "x\{70}". The others are just consistent behavior: ..., "x\{-3,}", "x\{-2,}", "x\{-1,}.

- Bram

But what if we want to match only the second occurrence of quoted text? Or we want to replace only a part of the quoted text keeping the rest untouched? We will need grouping and backreferences. But before let's talk more about character ranges.

3.4 Character ranges

Typical character ranges:

[012345] will match any of the numbers inside the brackets. The same range can be written as [0-5], where dash indicates a range of characters in ASCII order. Likewise, we can define the range for all lowercase letters: [a-z], for all letters: [a-zA-Z], letters and digits: [0-9a-zA-Z] etc. Depending on your system locale you can define range which will include characters like à, Ö, ß and other non ASCII characters.

Note that the range represents just one character in the search pattern, that is [0123] and 0123 are not the same. Likewise the order (with a few exceptions) is not important: [3210] and [0123] are the same character ranges, while 0123 and 3210 are two different patterns. Watch what happens when we apply

s:[65]:Dig:g

to the following text:

Before:

High 65 to 70. Southeast wind around 10

After:

High DigDig to 70. Southeast wind around 10

and now:

s:65:Dig:g

Before:

High 65 to 70. Southeast wind around 10

After:

High Dig to 70. Southeast wind around 10

Sometimes it's easier to define the characters you don't want to match. This is done by putting a negation sign "^" (caret) as a first character of the range

/[^A-Z]/

- will match any character except capital letters. We can now rewrite our pattern for quoted text using

/"[^"]\+"/

Note: inside the [ ] all metacharacters behave like ordinary characters. If you want to include "-" (dash) in your range put it first

/[-0-9]/

- will match all digits and -. "^" will lose its special meaning if it's not the first character in the range.

Now, let's have some real life example. Suppose you want to run a grammar check on your file and find all places where new sentence does not start with a capital letter. The pattern that will catch this:

\.\s\+[a-z]

- a period followed by one or more blanks and a lowercase word. We know how to find an error, now let's see how we can correct it. To do this we need some ways to remember our matched pattern and recall it later. That is exactly what backreferences are for.

3.5 Grouping and Backreferences

You can group parts of the pattern expression enclosing them with "\(" and "\)" and refer to them inside the replacement pattern by their special number \1, \2 ... \9. Typical example is swapping first two words of the line:

where \1 holds the first word, \2 - any number of spaces or tabs in between and \3 - the second word. How to decide what number holds what pair of \(\) ? - count opening "\(" from the left.

Replacement Part of :substitute

Replacement part of the S&R has its own special characters which we are going to use to fix grammar:

Now the full S&R to correct non-capital words at the beginning of the sentences looks like

s:\([.!?]\)\s\+\([a-z]\):\1  \u\2:g

We have corrected our grammar and as an extra job we replaced variable number of spaces between punctuation and the first letter of the next sentence with exactly two spaces.

3.6 Alternations

Using "\|" you can combine several expressions into one which matches any of its components. The first one matched will be used.

\(Date:\|Subject:\|From:\)\(\s.*\)

3.7 Regexp Operator Precedence

As in arithmetic expressions, regular expressions are executed in a certain order of precedence. Here the table of precedence, from highest to lowest:

4.1 Global search and execution

I want to introduce another quite useful and powerful Vim command which we're going to use later

The global commands work by first scanning through the [range] of of the lines and marking each line where a match occurs. In a second scan the [cmd] is executed for each marked line with its line number prepended. If a line is changed or deleted its mark disappears. The default for the [range] is the whole file.

Note: Ex commands are all commands you are entering on the Vim command line like :s[ubstitute], :co[py] , :d[elete], :w[rite] etc. Non-Ex commands (normal mode commands) can be also executed via

:norm[al]non-ex command

mechanism.

4.2 Examples

Some examples of :global usage:

:g/^$/ d

- delete all empty lines in a file

:10,20g/^/ mo 10

- reverse the order of the lines starting from the line 10 up to the line 20.

Here is a modified example from Walter Zintz vi tutorial:

:'a,'b g/^Error/ . w >> errors.txt

- in the text block marked by 'a and 'b find all the lines starting with Error and copy (append) them to "errors.txt" file. Note: . (current line address) in front of the w is very important, omitting it will cause :write to write the whole file to "errors.txt" for every Error line found.

You can give multiple commands after :global using "|" as a separator. If you want to use "|' in an argument, precede it with "\'. Another example from Zintz tutorial:

:g/^Error:/ copy $ | s /Error/copy of the error/

- will copy all Error line to the end of the file and then make a substitution in the copied line. Without giving the line address :s will operate on the current line, which is the newly copied line.

:g/^Error:/ s /Error/copy of the error/ | copy $

- here the order is reversed: first modify the string then copy to the end.

5.1 Tips and Techniques

A collection of some useful S&R tips:

(1) sent by Antonio Colombo:

"a simple regexp I use quite often to clean up a text: it drops the blanks at the end of the line:"

s:\s*$::

or (to avoid acting on all lines):

s:\s\+$::

5.2 Creating outline

For this example you need to know a bit of HTML. We want to make a table of contents out of h1 and h2 headings, which I will call majors and minors. HTML heading h1 is a text enclosed by <h1> tags as in <h1>Heading</h1>.

(1) First let's make named anchors in all headings, i.e. put <h1><a name="anchor">Heading</a></h1> around all headings. The "anchor" is a unique identifier of this particular place in HTML document. The following S&R does exactly this:

:s:\(<h[12]>\)\(.*\s\+\([-a-zA-Z]\+\)\)\s*\(</h[12]>\):\1<a name="\3">\2</a>\4:

Explanation: the first pair of \(\) saves the opening tag (h1 or h2) to the \1, the second pair saves all heading text before the closing tag, the third pair saves the last word in the heading which we will later use for "anchor" and the last pair saves the closing tag. The replacement is quite obvious - we just reconstruct a new "named" heading using \1-\4 and link tag <a>.

(2) Now let's copy all headings to one place:

:%g/<h[12]>/ t$

This command searches our file for the lines starting with <h1> or <h2> and copies them to the end of the file. Now we have a bunch of lines like:

<h1><a name="anchor1">Heading1></a></h1>
<h2><a name="anchor2">Heading2></a></h2>
<h2><a name="anchor3">Heading3></a></h2>
..........................
<h1><a name="anchorN">HeadingN></a></h1>

First, we want to convert all name=" to href="# in order to link table entries to their respective places in the text:

s:name=":href="#:

Second, we want our h1 entries look different from h2. Let's define CSS classes "majorhead" and "minorhead" and do the following:

g/<h1>/ s:<a:& class="majorhead":
g/<h2>/ s:<a:& class="minorhead":

Now our entries look like:

<h1><a class="majorhead" name="anchor1">Heading1></a></h1>
<h2><a class="minorhead" name="anchor2">Heading2></a></h2>

We no longer need h1 and h2 tags:

s:<h[21]>::

and replace closing tags with breaklines <br>

s:/h[21]:br:

<a class="majorhead" name="anchor1">Heading1></a><br>
<a class="minorhead" name="anchor2">Heading2></a><br>

5.3 Working with Tables

Quite often you have to work with a text organized in tables/columns. Consider, for example, the following text

Suppose we want to change all "Europe" cells in the third column to "Asia":

:%s:\(\(\w\+\s\+\)\{2}\)Europe:\1Asia:

To swap the first and the last columns:

:%s:\(\w\+\)\(.*\s\+\)\(\w\+\)$:\3\2\1:

To be continued...

Read VIM documentation about pattern and searching - type ":help pattern" in VIM normal mode to get this information.

The only book I know of which deals with VIM/vi regular expressions is the same "Learning the vi Editor" by Linda Lamb and Arnold Robbins.

Definitive reference on regular expressions is Jeffrey Friedl's "Mastering Regular Expressions" published by O'Reilly & Associates, but it mostly deals with Perl regular expressions. Note that Perl regexp flavor is a bit different from VIM's. O'Reilly has one of the book chapters available online.

Others sources: practically every Perl book has some information on regular expressions.