Vim Regular Expressions

From trapsink.com
Jump to: navigation, search


Basic Vi Cheat Sheet | Advanced Vi Cheat Sheet

Overview

This page has been culled from an old geocities site no longer available. Original text © 2000-2001 Oleg Raisky (volontir at yahoo dot com). Much editing has taken place but the spirit of the original remains.


Search & Replace

The most common task is to make replacements in text following certain rules. An excerpt from VIM help:

:range s[ubstitute]/pattern/string/cgiI
For each line in the range replace a match of the pattern with the string where:
c Confirm each substitution
g Replace all occurrences in the line (without g - only first).
i Ignore case for the pattern.
I Don't ignore case for the pattern.

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


Range of Operation, Line Addressing and Marks

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.

Specifier Description
number an absolute line number
. the current line
$ the last line in the file
% the whole file. The same as 1,$
't position of mark "t"
/pattern[/] the next line where text "pattern" matches.
 ?pattern[?] the previous line where text "pattern" matches
\/ the next line where the previously used search pattern matches
\? the previous line where the previously used search pattern matches
\& the next line where the previously used substitute pattern matches

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


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).
:/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.


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.


"Escaped" characters or metacharacters

So far our pattern strings were constructed from normal or literal text characters. The power of regex 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.

# Matching # Matching
. any character except new line
\s whitespace character \S non-whitespace character
\d digit \D non-digit
\x hex digit \X non-hex digit
\o octal digit \O non-octal digit
\h head of word character (a,b,c...z,A,B,C...Z and _) \H non-head of word character
\p printable character \P like \p, but excluding digits
\w word character \W non-word character
\a alphabetic character \A non-alphabetic character
\l lowercase character \L non-lowercase character
\u uppercase character \U non-uppercase character

So, to match a date like 09/01/2000 you can use (assuming you don't use "/" as a separator in the S&R):

\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.


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:

Quantifier Description
* matches 0 or more of the preceding characters, ranges or metacharacters .* matches everything including empty line
\+ matches 1 or more of the preceding characters...
\= matches 0 or 1 more of the preceding characters...
\{n,m} matches from n to m of the preceding characters...
\{n} matches exactly n times of the preceding characters...
\{,m} matches at most m (from 0 to m) of the preceding characters...
\{n,} matches at least n of of the preceding characters...
where n and m are positive integers (>0)

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:

Quantifier Description
\{-} matches 0 or more of the preceding atom, as few as possible
\{-n,m} matches 1 or more of the preceding characters...
\{-n,} matches at lease or more of the preceding characters...
\{-,m} matches 1 or more of the preceding characters...
colspan="2" where n and m are positive integers (>0)

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.


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 a 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.


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:

s:\(\w\+\)\(\s\+\)\(\w\+\):\3\2\1:

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:

# Meaning # Meaning
& the whole matched pattern \L the following characters are made lowercase
\0 the whole matched pattern \U the following characters are made uppercase
\1 the matched pattern in the first pair of \(\) \E end of \U and \L
\2 the matched pattern in the second pair of \(\) \e end of \U and \L
... ... \r split line in two at this point
\9 the matched pattern in the ninth pair of \(\) \l next character made lowercase
~ the previous substitute string \u next character made uppercase

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.


Alternations

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

<codeSubject:\>\(Date:\</code>

will parse various mail headings and their contents into \1 and \2, respectively. The thing to remember about VIM alternation that it is not greedy. It won't search for the longest possible match, it will use the first that matched. That means that the order of the items in the alternation is important!


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:

Precedence Regexp Description
1 \( \) grouping
2 \=,\+,*,\{n} etc. quantifiers
3 abc\t\.\w sequence of characters/ metacharacters, not containing quantifiers or grouping operators
4 alternation


Global search and execution

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

:range g[lobal][!]/pattern/cmd
Execute the Ex command cmd (default ":p") on the lines within [range] where pattern matches. If pattern is preceded with a ! - only where match does not occur.

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


Examples

Some examples of :global usage:

:g/^$/ d

- delete all empty lines in a file

:g/^$/,/./-j

- reduce multiple blank lines to a single blank

: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.


Tips and Techniques

Tip 1: frequently you need to do S&R in a text which contains UNIX file paths - text strings with slashes ("/") inside. Because S&R command uses slashes for pattern/replacement separation you have to escape every slash in your pattern, i.e. use "\/" for every "/" in your pattern:

s/\/dir1\/dir2\/dir3\/file/dir4\/dir5\/file2/g

To avoid this so-called "backslashitis" you can use different separators in S&R (I prefer ":")

s:/dir1/dir2/dir3/file:/dir4/dir5/file2:g

Tip 2: You may find these mappings useful (put them in your .vimrc file)

noremap ;; :%s:::g<Left><Left><Left>
noremap ;' :%s:::cg<Left><Left><Left><Left>

These mappings save you some keystrokes and put you where you start typing your search pattern. After typing it you move to the replacement part , type it and hit return. The second version adds confirmation flag.

Tip 3: Quick mapping to put \(\) in your pattern string

cmap ;\ \(\)<Left><Left>


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>


Working with Tables

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

Asia America Africa Europe
Africa Europe Europe Africa
Europe Asia Europe Europe

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

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

To swap the first and the last columns:

:%s:\(\w\+\)\(.*\s\+\)\(\w\+\)$:\3\2\1:
Europe America Africa Asia
Africa Europe Europe Africa
Europe Asia Europe Europe


References

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