[Chapter 6] 6.4 Pattern-matching Examples

6.4 Pattern-matching Examples

Unless you are already familiar with regular expressions, the discussion of special characters above probably looks forbiddingly complex. A few more examples should make things clearer. In the examples that follow, a square () is used to mark a blank space; it is not a special character.

Let's work through how you might use some special characters in a replacement. Suppose that you have a long file and that you want to substitute the word child with the word children throughout that file. You first save the edited buffer with :w, then try the global replacement:

:%s/child/children/g

When you continue editing, you notice occurrences of words such as childrenish. You have unintentionally matched the word childish. Returning to the last saved buffer with :e!, you now try:

:%s/child/children/g

(Note that there is a space after child.) But this command misses the occurrences child., child,, child: and so on. After some thought, you remember that brackets allow you to specify one character from among a list, so you come upon the solution:

:%s/child[,.;:!?]/children[,.;:!?]/g

This searches for child followed by either a space (indicated by ) or any one of the punctuation characters ,.;:!?. You expect to replace this with children followed by the corresponding space or punctuation mark, but you've ended up with a bunch of punctuation marks after every occurrence of children. You need to save the space and punctuation marks inside a $ and $. Then you can "replay" them with a \1. Here's the next attempt:

:%s/child\([,.;:!?]\)/children\1/g

When the search matches a character inside the $ and $, the \1 on the right-hand side restores the same character. The syntax may seem awfully complicated, but this command sequence can save you a lot of work! Any time you spend learning regular expression syntax will be repaid a thousandfold!

The command is still not perfect, though. You've noticed that occurrences of Fairchild have been changed, so you need a way to match child when it isn't part of another word.

As it turns out, vi (but not all other programs that use regular expressions) has a special syntax for saying "only if the pattern is a complete word." The character sequence \< requires the pattern to match at the beginning of a word, whereas \> requires the pattern to match at the end of a word. Using both will restrict the match to a whole word. So, in the task given above, \<child\> will find all instances of the word child, whether followed by punctuation or spaces. Here's the substitution command you should use:

:%s/\<child\>/children/g

6.4.1 Search for General Class of Words

Suppose you have subroutine names beginning with the prefixes: mgi, mgr and mga.
mgibox routine, mgrbox routine, mgabox routine,

If you want to save the prefixes but want to change the name box to square, either of the following replacement commands will do the trick. The first example illustrates how $ and $ can be used to save whatever pattern was actually matched. The second example shows how you can search for one pattern but change another.

:g/mg\([ira]\)box/s//mg\1square/g

mgisquare routine, mgrsquare routine, mgasquare routine,
The global replacement keeps track of whether an i, r or a is saved. In that way, box is changed to square only when box is part of the routine's name.

:g/mg[ira]box/s/box/square/g

mgisquare routine, mgrsquare routine, mgasquare routine,
Has the same effect as the previous command, but it is a little less safe since it could change other instances of box on the same line, not just those within the routine names.

6.4.2 Block Move by Patterns

You can also move blocks of text delimited by patterns. For example, assume you have a 150-page reference manual. All references pages are organized into three paragraphs with the same three headings: SYNTAX, DESCRIPTION, and PARAMETERS. A sample of one reference page follows:

.Rh 0 "Get status of named file" "STAT"
.Rh "SYNTAX"
.nf
integer*4 stat, retval
integer*4 status(11)
character*123 filename
...
retval = stat (filename, status)
.fi
.Rh "DESCRIPTION"
Writes the fields of a system data structure into the
status array.  
These fields contain (among other
things) information about the file's location, access
privileges, owner, and time of last modification.
.Rh "PARAMETERS"
.IP "\fBfilename\fR" 15n
A character string variable or constant containing
the UNIX pathname for the file whose status you want
to retrieve.  
You can give the ...

Suppose that it is decided to move DESCRIPTION above the SYNTAX paragraph. With pattern matching, you can move blocks of text on all 150 pages with one command!

:g /SYNTAX/,/DESCRIPTION/-1 mo /PARAMETERS/-1

This command operates on the block of text between the line containing the word SYNTAX and the line just before the word DESCRIPTION (/DESCRIPTION/-1). The block is moved (using mo) to the line just before PARAMETERS (/PARAMETERS/-1). Note that ex can place text only below the line specified. To tell ex to place text above a line, you first have to move up a line with -1, and then place your text below. In a case like this, one command saves literally hours of work. (This is a real-life example - we once used a pattern match like this to rearrange a reference manual containing hundreds of pages.)

Block definition by patterns can be used equally well with other ex commands. For example, if you wanted to delete all DESCRIPTION paragraphs in the reference chapter, you could enter:

:g/DESCRIPTION/,/PARAMETERS/-1d

This very powerful kind of change is implicit in ex's line addressing syntax, but it is not readily apparent even to experienced users. For this reason, whenever you are faced with a complex, repetitive editing task, take the time to analyze the problem and find out if you can apply pattern-matching tools to get the job done.

6.4.3 More Examples

Since the best way to learn pattern matching is by example, here is a list of pattern-matching examples, with explanations. Study the syntax carefully, so that you understand the principles at work. You should then be able to adapt these examples to your own situation.

Put troff italicization codes around the word RETURN:
```
:%s/RETURN/\\fIRETURN\\fP/g
```
Notice that two backslashes (\\) are needed in the replacement, because the backslash in the troff italicization code will be interpreted as a special character. (\fI alone would be interpreted as fI; you must type \\fI to get \fI.)
Modify a list of pathnames in a file:
```
:%s/\/usr\/tim/\/usr\/linda/g
```
A slash (used as a delimiter in the global replacement sequence) must be escaped with a backslash when it is part of the pattern or replacement; use \/ to get /. An alternate way to achieve this same effect is to use a different character as the pattern delimiter. For example, you could make the above replacement using colons as delimiters. Thus:
```
:%s:/usr/tim:/usr/linda:g
```
Change all periods to semicolons in lines 1 to 10:
```
:1,10s/\./;/g
```
A dot has special meaning in regular expression syntax and must be escaped with a backslash (\.).
Change all occurrences of the word help (or Help) to HELP:
```
:%s/[Hh]elp/HELP/g
```
or:
```
:%s/[Hh]elp/\U&/g
```
The \U changes the pattern that follows to all uppercase. The pattern that follows is the repeated search pattern, which is either help or Help.
Replace one or more spaces with a single space:
```
:%s/*//g
```
Make sure you understand how the asterisk works as a special character. An asterisk following any character (or following any regular expression that matches a single character, such as . or [a-z]) matches zero or more instances of that character. Therefore, you must specify two spaces followed by an asterisk to match one or more spaces (one space, plus zero or more spaces).
Replace one or more spaces following a colon with two spaces:
```
:%s/:*/:/g
```
Replace one or more spaces following a period or a colon with two spaces:
```
:%s/$[:.]$*/\1/g
```
Either of the two characters within brackets can be matched. This character is saved into a hold buffer, using $ and $, and restored on the right-hand side by the \1. Note that within brackets a special character such as a dot does not need to be escaped.
Standardize various uses of a word or heading:
```
:%s/^Note[:s]*/Notes:/g
```
The brackets enclose three characters: a space, a colon, and the letter s. Therefore, the pattern Note[s:] will match Note, Notes or Note:. An asterisk is added to the pattern so that it also matches Note (with zero spaces after it) and Notes: (the already correct spelling). Without the asterisk, Note would be missed entirely and Notes: would be incorrectly changed to Notes::.
Delete all blank lines:
```
:g/^$/d
```
What you are actually matching here is the beginning of the line (^) followed by the end of the line ($), with nothing in between.
Delete all blank lines, plus any lines that contain only white space:
```
:g/^[tab]*$/d
```
(In the line above, a tab is shown as tab.) A line may appear to be blank but may in fact contain spaces or tabs. The previous example will not delete such a line. This example, like the one above it, searches for the beginning and end of the line. But instead of having nothing in between, the pattern tries to find any number of spaces or tabs. If no spaces or tabs are matched, the line is blank. To delete lines that contain white space but that aren't blank, you would have to match lines with at least one space or tab:
```
:g/^[tab][tab]*$/d
```
Delete all leading spaces on a line:
```
:%s/^*$.*$/\1/
```
Use ^* to search for one or more spaces at the beginning of a line; then use $.*$ to save the rest of the line into the first hold buffer. Restore the line without spaces, using \1.
Delete all spaces at the end of a line:
```
:%s/$.*$*$/\1/
```
Use $.*$ to save all the text on the line, but only up until one or more spaces at the end of the line. Restore the saved text without the spaces.
The substitutions in this example and the previous one will happen only once on any given line, so the g option doesn't need to follow the replacement string.
Insert a > at the start of every line in a file:
```
:%s/^/>/
```
What we're really doing here is "replacing" the start of the line with > Of course, the start of the line (being a logical construct, not an actual character) isn't really replaced!
This command is useful when replying to mail or USENET news postings. Frequently, it is desirable to include part of the original message in your reply. By convention, the inclusion is distinguished from your reply by setting off the included text with a right angle bracket and a couple of spaces at the start of the line. This can be done easily as shown above. (Typically, only part of the original message will be included. Unneeded text can be deleted either before or after the above replacement.) Advanced mail systems do this automatically. However, if you're using a primitive mail program, you may need to do it manually.
Add a period to the end of the next six lines:
```
:.,+5s/$/./
```
The line address indicates the current line plus five lines. The $ indicates the end of line. As in the previous example, the $ is a logical construct. You aren't really replacing the end of the line.
Reverse the order of all hyphen-separated items in a list:
```
:%s/$.*$-$.*$/\2-\1/
```
Use $.*$ to save text on the line into the first hold buffer, but only until you find -. Then use $.*$ to save the rest of the line into the second hold buffer. Restore the saved portions of the line, reversing the order of the two hold buffers. The effect of this command on several items is shown below.
```
more - display files
```
becomes:
```
display files - more
```
and:
```
lp - print files
```
becomes:
```
print files - lp
```
Change every word in a file to uppercase:
```
:%s/.*/\U&/
```
or:
```
:%s/./\U&/g
```
The \U flag at the start of the replacement string tells vi to change the replacement to uppercase. The & character replays the search pattern as the replacement. These two commands are equivalent; however, the first form is considerably faster, since it results in only one substitution per line (.* matches the entire line, once per line), whereas the second form results in repeated substitutions on each line (. matches only a single character, with the replacement repeated on account of the trailing g).
Reverse the order of lines in a file: [1]
[1] From the article by Walter Zintz in UNIX World, May 1990.
```
:g/.*/mo0
```
The search pattern matches all lines (a line contains zero or more characters). Each line is moved, one by one, to the top of the file (that is, moved after imaginary line 0). As each matched line is placed at the top, it pushes the previously moved lines down, one by one, until the last line is on top. Since all lines have a beginning, the same result can be achieved more succinctly:
```
:g/^/mo0
```
In a database, on all lines not marked Paid in full, append the phrase Overdue:
```
:g!/Paidinfull/s/$/Overdue/
```
or the equivalent:
```
:v/Paidinfull/s/$/Overdue/
```
To affect all lines except those matching your pattern, add a ! to the g command, or simply use the v command.
For any line that doesn't begin with a number, move the line to the end of the file:
```
:g!/^[1-9]/m$
```
or:
```
:g/^[^1-9]/m$
```
As the first character within brackets, a caret negates the sense, so the two commands have the same effect. The first one says, "Don't match lines that begin with a number," and the second one says, "Match lines that don't begin with a number."
Change manually numbered section heads (e.g., 1.1, 1.2, etc.) to a troff macro (e.g., .Ah for an A-level heading):
```
:%s/[1-9]\.[1-9]/.Ah/
```
The search string matches a digit other than zero, followed by a period, followed by another nonzero digit. Notice that the period doesn't need to be escaped in the replacement (though a \ would have no effect, either). The command above won't find chapter numbers containing two or more digits. To do so, modify the command like this:
```
:%s/[1-9][0-9]*\.[1-9]/.Ah/
```
Now it will match chapters 10 to 99 (digits 1 to 9, followed by a digit), 100 to 999 (digits 1 to 9, followed by two digits), etc. The command still finds chapters 1 to 9 (digits 1 to 9, followed by no digit).
Remove numbering from section headings in a document. You want to change the sample lines:
```
2.1 Introduction
10.3.8 New Functions
```
into the lines:
```
Introduction
New Functions
```
Here's the command to do this:
```
:%s/^[1-9][0-9]*\.[1-9][1-9.]*//
```
The search pattern resembles the one in the previous example, but now the numbers vary in length. At a minimum, the headings contain number, period, number, so you start with the search pattern from the previous example:
```
[1-9][0-9]*\.[1-9]
```
But in this example, the heading may continue with any number of digits or periods:
```
[1-9.]*
```
Change the word Fortran to the phrase FORTRAN (acronym of FORmula TRANslation):
```
:%s/$For$$tran$/\U\1\2\E(acronymof\U\1\Emula\U\2\Eslation)/g
```
First, since we notice that the words FORmula and TRANslation use portions of the original word, we decide to save the search pattern in two pieces: $For$ and $tran$. The first time we restore it, we use both pieces together, converting all characters to uppercase: \U\1\2. Next, we undo the uppercase with \E; otherwise the remaining replacement text would all be uppercase. The replacement continues with actual typed words, then we restore the first hold buffer. This buffer still contains For, so again we convert to uppercase first: \U\1. Immediately after, we lowercase the rest of the word: \Emula. Finally, we restore the second hold buffer. This contains tran, so we precede the "replay" with uppercase, follow it with lowercase, and type out the rest of the word: \U\2\Eslation).

6.4.4 A Final Look at Pattern Matching

We conclude this chapter by presenting sample tasks that involve complex pattern-matching concepts. Rather than solve the problems right away, we'll work toward the solutions step by step.

6.4.4.1 Deleting an Unknown Block of Text

Suppose you have a few lines with this general form:

the best of times; the worst of times:  moving
The coolest of times; the worst of times:  moving

The lines that you're concerned with always end with moving, but you never know what the first two words might be. You want to change any line that ends with moving to read:

The greatest of times; the worst of times:  moving

Since the changes must occur on certain lines, you need to specify a context-sensitive global replacement. Using :g/moving$/ will match lines that end with moving. Next, you realize that your search pattern could be any number of any character, so the metacharacters .* come to mind. But these will match the whole line unless you somehow restrict the match. Here's your first attempt:

:g/moving$/s/.*of/Thegreatestof/

This search string, you decide, will match from the beginning of the line to the first of. Since you needed to specify the word of to restrict the search, you simply repeat it in the replacement. Here's the resulting line:

The greatest of times:  moving

Something went wrong. The replacement gobbled the line up to the second of instead of the first. Here's why. When given a choice, the action of "match any number of any character" will match as much text as possible. In this case, since the word of appears twice, your search string finds:

the best of times; the worst of

rather than:

the best of

Your search pattern needs to be more restrictive:

:g/moving$/s/.*of times;/The greatest of times;/

Now the .* will match all characters up to the instance of the phrase of times;. Since there's only one instance, it has to be the first.

There are cases, though, when it is inconvenient, or even incorrect, to use the .* metacharacters. For example, you might find yourself typing many words to restrict your search pattern, or you might be unable to restrict the pattern by specific words (if the text in the lines varies widely). The next section presents such a case.

6.4.4.2 Switching Items in a Database

Suppose you want to switch the order of all last names and first names in a database. The lines look like this:

Name: Feld, Ray; Areas: PC, UNIX; Phone: 123-4567
Name: Joy, Susan S.; Areas: Graphics; Phone: 999-3333

The name of each field ends with a colon, and each field is separated by a semicolon. Using the top line as an example, you want to change Feld, Ray to Ray Feld. We'll present some commands that look promising but don't work. After each command, we show you the line the way it looked before the change and after the change.

:%s/: \(.*\), \(.*\);/: \2 \1;/

Name: Feld, Ray; Areas: PC, UNIX; Phone: 123-4567	Before
Name: UNIX Feld, Ray; Areas: PC; Phone: 123-4567	After

We've highlighted the contents of the first hold buffer in bold and the contents of the second hold buffer in italic. Note that the first hold buffer contains more than you want. Since it was not sufficiently restricted by the pattern that follows it, the hold buffer was able to save up to the second comma. Now you try to restrict the contents of the first hold buffer:

:%s/: \(....\), \(.*\);/: \2 \1;/

Name: Feld, Ray; Areas: PC, UNIX; Phone: 123-4567	Before
Name: Ray; Areas: PC, UNIX Feld; Phone: 123-4567	After

Here you've managed to save the last name in the first hold buffer, but now the second hold buffer will save anything up to the last semicolon on the line. Now you restrict the second hold buffer, too:

:%s/: \(....\), \(...\);/: \2 \1;/

Name: Feld, Ray; Areas: PC, UNIX; Phone: 123-4567	Before
Name: Ray Feld; Areas: PC, UNIX; Phone: 123-4567	After

This gives you what you want, but only in the specific case of a four-letter last name and a three-letter first name. (The previous attempt included the same mistake.) Why not just return to the first attempt, but this time be more selective about the end of the search pattern?

:%s/: \(.*\), \(.*\); Area/: \2 \1;/

Name: Feld, Ray; Areas: PC, UNIX; Phone: 123-4567	Before
Name: Ray Feld; Areas: PC, UNIX; Phone: 123-4567	After

This works, but we'll continue the discussion by introducing an additional concern. Suppose that the Area field isn't always present or isn't always the second field. The above command won't work on such lines.

We introduce this problem to make a point. Whenever you rethink a pattern match, it's usually better to work toward refining the variables (the metacharacters), rather than using specific text to restrict pattterns. The more variables you use in your patterns, the more powerful your commands will be.

In the current example, think again about the patterns you want to switch. Each word starts with an uppercase letter and is followed by any number of lowercase letters, so you can match the names like this:

[A-Z][a-z]*

Ok, but a last name might also have more than one uppercase letter (McFly, for example), so you'd want to search for this possibility in the second and succeeding letters:

[A-Z][A-Za-z]*

It doesn't hurt to use this for the first name, too (you never know when McGeorge Bundy will turn up). Your command now becomes:

:%s/: \([A-Z][A-Za-z]*\), \([A-Z][A-Za-z]*\);/: \2 \1;/

Quite forbidding, isn't it? It still doesn't cover the case of a name like Joy, Susan S. Since the first-name field might include a middle initial, you need to add a space and a period within the second pair of brackets. But enough is enough. Sometimes, specifying exactly what you want is more difficult than specifying what you don't want. In your sample database, the last names end with a comma, so a last-name field can be thought of as a string of characters that are not commas:

[^,]*

This pattern matches characters up until the first comma. Similarly, the first-name field is a string of characters that are not semicolons:

[^;]*

Putting these more efficient patterns back into your previous command, you get this:

:%s/: \([^,]*\), \([^;]*\);/: \2 \1;/

The same command could also be entered as a context-sensitive replacement. If all lines begin with Name, you can say:

:g/^Name/s/: \([^,]*\), \([^;]*\);/: \2 \1;/

You can also add an asterisk after the first space, in order to match a colon that has extra spaces (or no spaces) after it:

:g/^Name/s/: *\([^,]*\), \([^;]*\);/: \2 \1;/

6.4.4.3 Using :g to Repeat a Command

As we've usually seen the :g command used, it selects lines that are typically then edited by subsequent commands on the same line - for example, we select lines with g, and then make substitutions on them, or select them and delete them:

:g/mg[ira]box/s/box/square/g
:g/^$/d

However, in his two-part tutorial in UNIX World, [2] Walter Zintz makes an interesting point about the g command. This command selects lines - but the associated editing commands need not actually affect the lines that are selected.

[2] Part 1, "vi Tips for Power Users," appears in the April 1990 issue of UNIX World. Part 2, "Using vi to Automate Complex Edits," appears in the May 1990 issue. The examples presented are from Part 2.

Instead, he demonstrates a technique by which you can repeat ex commands some arbitrary number of times. For example, suppose you want to place ten copies of lines 12 through 17 of your file at the end of your current file. You could type:

:1,10g/^/ 12,17t$

This is a very unexpected use of g, but it works! The g command selects line 1, executes the specified t command, then goes on to line 2, to execute the next copy command. When line 10 is reached, ex will have made ten copies.

6.4.4.4 Collecting Lines

Here's another advanced g example, again building on suggestions provided in Zintz's article. Suppose you're editing a document that consists of several parts. Part 2 of this file is shown below, using ellipses to show omitted text and displaying line numbers for reference.

301  Part 2
302  Capability Reference
303  .LP
304  Chapter 7
305  Introduction to the Capabilities
306  This and the next three chapters ...

400  ... and a complete index at the end.
401  .LP
402  Chapter 8
403  Screen Dimensions
404  Before you can do anything useful
405  on the screen, you need to know ...

555  .LP
556  Chapter 9
557  Editing the Screen
558  This chapter discusses ...

821  .LP
822  Part 3:
823  Advanced Features
824  .LP
825  Chapter 10

The chapter numbers appear on one line, their titles appear on the line below, and the chapter text (highlighted for emphasis) begins on the line below that. The first thing you'd like to do is copy the beginning line of each chapter, sending it to an already existing file called begin.

Here's the command that does this:

:g /^Chapter/ .+2w >> begin

You must be at the top of your file before issuing this command. First you search for Chapter at the start of a line, but then you want to run the command on the beginning line of each chapter - the second line below Chapter. Because a line beginning with Chapter is now selected as the current line, the line address .+2 will indicate the second line below it. The equivalent line addresses +2 or ++ work as well. You want to write these lines to an existing file named begin, so you issue the w command with the append operator >>.

Suppose you want to send the beginnings of chapters that are only within Part 2. You need to restrict the lines selected by g, so you change your command to this:

:/^Part 2/,/^Part 3/g /^Chapter/ .+2w >> begin

Here, the g command selects the lines that begin with Chapter, but it searches only that portion of the file from a line starting with Part 2 through a line starting with Part 3. If you issue the above command, the last lines of the file begin will read as follows:

This and the next three chapters ...
Before you can do anything useful
This chapter discusses ...

These are the lines that begin Chapters 7, 8, and 9.

In addition to the lines you've just sent, you'd like to copy chapter titles to the end of the document, in preparation for making a table of contents. You can use the vertical bar to tack a second command after your first command, like so:

:/^Part 2/,/^Part 3/g /^Chapter/ .+2w >> begin | +t$

Remember that with any subsequent command, line addresses are relative to the previous command. The first command has marked lines (within Part 2) that start with Chapter, and the chapter titles appear on a line below such lines. Therefore, to access chapter titles in the second command, the line address is + (or the equivalents +1 or .+1). Then use t$ to copy the chapter titles to the end of the file.

As these examples illustrate, thought and experimentation may lead you to some unusual editing solutions. Don't be afraid to try things! Just be sure to back up your file first.

	mgisquare routine, mgrsquare routine, mgasquare routine,
	The global replacement keeps track of whether an i, r or a is saved. In that way, box is changed to square only when box is part of the routine's name.

	mgisquare routine, mgrsquare routine, mgasquare routine,
	Has the same effect as the previous command, but it is a little less safe since it could change other instances of box on the same line, not just those within the routine names.