Word expansion is substitution of part of a word with another particular string. There are seven types of word expansions:

These types of expansions are performed in the order specified above.

Tilde expansion, parameter expansion, command substitution, and arithmetic expansion are called the four expansions.

Tilde expansion

In tilde expansion, parts of words that start with a tilde (~) are substituted with particular pathnames. The part of each word that gets substituted is from the beginning of the word, which is a tilde, up to (but not including) the first slash (/) in the word. If the word does not contain a slash, the whole word is substituted. If any character in the substituted part is quoted, tilde expansion is not performed on the word.

The results of expansion are determined by the format of the substituted part:

~

A single tilde is substituted with the value of the HOME variable.

~username

A tilde followed by a user name is substituted with the pathname of the user’s home directory.

~+

~+ is substituted with the value of the PWD variable.

~-

~- is substituted with the value of the OLDPWD variable.

~+n
~-n

where n is a non-negative integer. This type of tilde expansion yields the pathname of a directory of which ~+n or ~-n is the index in the directory stack.

When tilde expansion is performed on the value of a variable assignment that occurs during execution of a simple command, the value is considered as a colon-separated list of words and those words are each subject to tilde expansion. For example, the variable assignment

VAR=~/a:~/b:~/c

is equivalent to

VAR=/home/foo/a:/home/foo/b:/home/foo/c

if the value of HOME variable is /home/foo.

The POSIX standard does not prescribe how the shell should behave when it encounters an error during tilde expansion (e.g., when the HOME variable is not defined). Yash silently ignores any errors during tilde expansion; the part of the word that would be substituted is left intact.

In the POSIXly-correct mode, tilde expansion supports the formats of ~ and ~username only.

Parameter expansion

Parameter expansion expands to the value of a parameter.

The syntax of typical, simple parameter expansion is ${parameter}, which expands to the value of the parameter whose name is parameter. You can omit the braces (e.g., $parameter) if

  • parameter is a special parameter,

  • parameter is a positional parameter whose index is a one-digit integer, or

  • parameter is a variable and the parameter expansion is not followed by a character that can be used as part of a variable name.

    For example, ${path}-name is equivalent to $path-name, but ${path}name and $pathname are different.

If parameter is none of a special parameter, positional parameter, and variable, it is a syntax error. (Some shells other than yash may treat such a case as an expansion error.)

If the unset option is disabled and the parameter is an undefined variable, it is an expansion error. If the unset option is enabled, an undefined variable expands to the empty string.

More complex syntax of parameter expansion allows modifying the value of a parameter.

Parameter expansion

${ prefix parameter index modifier }

The spaces in the syntax definition above are for readability only and must be omitted. You can omit prefix, index, and/or modifier.

Prefix

The prefix, if any, must be a hash sign (#). If a parameter expansion has the prefix, the result of expansion is the number of characters in the value this expansion would be expanded to without the prefix.

Parameter name

The parameter name (parameter) must be either

The parameter expansion is expanded to the value of the parameter. If parameter is an array variable, the values of the array are field-split like the @ special parameter unless the index [*] is specified.

If parameter is another expansion, it is called a nested expansion. Nested expansion cannot be used in the POSIXly-correct mode. The braces ({ }) of a nested parameter expansion cannot be omitted.

Index

An index allows extracting part of the parameter value (or some of array values).

Index

[word1]

[word1,word2]

where word1 and word2 are parsed in the same manner as normal tokens except that they are always delimited by , or ] and can contain whitespace characters.

If there is an index in a parameter expansion, it is interpreted as follows:

  1. Words word1 and word2 are subjected to parameter expansion, command substitution, and arithmetic expansion.

  2. If there is no word2 and if word1 expands to one of *, @, and #, then that is the interpretation of index and the next step is not taken.

  3. The results of the previous steps (the expanded word1 and word2) are interpreted and evaluated as an arithmetic expression in the same manner as in arithmetic expansion. The resulting integers are the interpretation of index. If the results are not integers, it is an expansion error. If there is no word2, it is assumed that word2 is equal to word1.

If parameter is an array variable, the index specifies the part of the array. If parameter is either the * or @ special parameter, the index specifies the index range of positional parameters. In other cases, the index specifies the index range of a substring of the parameter value that is being expanded. In all cases, the specified range of the array values, positional parameters, or parameter value remains in the results of the expansion and other values are dropped.

If the interpretation of index is one or two integers, the following rules apply:

  • If the interpreted index value is negative, it wraps around. For example, the index value of -1 corresponds to the last value/character.

  • It is not an error when the index value is out of range. Existing values/characters within the range are just selected.

  • If the interpretation of either word1 or word2 is 0, the range is assumed empty and the expansion results in nothing.

If the interpretation of index is one of *, @, and #, it is treated as follows:

*

If parameter is an array, all the array values are field-split or concatenated in the same manner as the * special parameter. If parameter is the * or @ special parameter, the positional parameters are likewise field-split or concatenated. In other cases, the interpretation of index is treated as if the interpretation is the two integers 1 and -1.

@

The interpretation of index is treated as if the interpretation is the two integers 1 and -1.

#

The interpretation of the # index is special in that it does not simply specify a range. Instead, the expanded values are substituted with the count.

If parameter is an array, the result of this parameter expansion will be the number of values in the array being expanded. If parameter is the * or @ special parameter, the result will be the number of current positional parameters. Otherwise, the result will be the number of characters in the value that is being expanded.

If a parameter expansion does not contain an index, it is assumed to be [@]. In the POSIXly-correct mode, index cannot be specified.

Example 1. Expansion of a normal variable

The following commands will print the string ABC:

var='123ABC789'
echo "${var[4,6]}"
Example 2. Expansion of positional parameters

The following commands will print the string 2 3 4:

set 1 2 3 4 5
echo "${*[2,-2]}"
Example 3. Expansion of an array

The following commands will print the string 2 3 4:

array=(1 2 3 4 5)
echo "${array[2,-2]}"

Modifier

You can modify the value to be expanded by using modifiers:

-word

If the parameter name (parameter) is an undefined variable, the parameter expansion is expanded to word. It is not treated as an error if the unset option is disabled.

+word

If the parameter name (parameter) is an existing variable, the parameter expansion is expanded to word. It is not treated as an error if the unset option is disabled.

=word

If the parameter name (parameter) is an undefined variable, word is assigned to the variable and the parameter expansion is expanded to word. It is not treated as an error if the unset option is disabled.

?word

If the parameter name (parameter) is an undefined variable, word is printed as an error message to the standard error. If word is empty, the default error message is printed instead.

:-word
:+word
:=word
:?word

These are similar to the four types of modifiers above. The only difference is that, if parameter exists and has an empty value, it is also treated as an undefined variable.

#word

The shell performs pattern matching against the value that is being expanded, using word as a pattern. If word matches the beginning of the value, the matching part is removed from the value and the other part remains as expansion results. The shortest matching is used if more than one matching is possible.

##word

This is similar to #word above. The only difference is that the longest matching is used if more than one matching is possible.

%word

This is similar to #word above. The only difference is that matching is tried at the end of the value rather than at the beginning: if word matches the end of the value, the matching part is removed from the value and the other part remains as expansion results.

%%word

This is similar to %word above. The only difference is that the longest matching is used if more than one matching is possible.

/word1/word2

The shell performs pattern matching against the value that is being expanded, using word1 as a pattern. If word1 matches any part of the value, the matching part is replaced with word2 and the whole value after the replacement remains as expansion results. If word1 matches more than one part of the value, only the first part is replaced. The shortest matching is replaced if more than one matching is possible for the same starting point in the value.

This modifier cannot be used in the POSIXly-correct mode.

/#word1/word2

This is similar to /word1/word2 above. The only difference is that word1 matches only at the beginning of the value being expanded.

/%word1/word2

This is similar to /word1/word2 above. The only difference is that word1 matches only at the end of the value being expanded.

//word1/word2

This is similar to /word1/word2 above. The only difference is that all matched parts are replaced if word1 matches more than one part of the value.

:/word1/word2

This is similar to /word1/word2 above. The only difference is that the value is replaced only when word1 matches the whole value.

In all types of modifiers above, words are subjected to the four expansions when (and only when) they are used.

If parameter is an array variable or the @ or * special parameter, modifiers affect each value of the array or all positional parameters.

Command substitution

Command substitution expands to output of commands specified.

Command substitution

$(commands)

`commands`

When command substitution is evaluated, commands are executed by a subshell with output pipelined to the shell. When the commands finished, command substitution is substituted with the output of the commands. Any trailing newline characters in the output are ignored.

When command substitution of the form $(commands) is parsed, the commands are parsed carefully so that complex commands such as nested command substitution are parsed correctly. If commands start with (, you should put a space before commands so that the whole command substitution is not confused with arithmetic expansion. If the shell is in the POSIXly-correctly mode, the commands are parsed each time the command substitution is expanded; otherwise, commands are parsed only when the command substitution is parsed.

If command substitution is of the form `commands`, the commands are not parsed when the command substitution is parsed; the commands are parsed each time the command substitution is expanded. The end of commands is detected by the first backquote character (`) after the beginning of commands that is not quoted by a backslash. Backquotes that are part of commands (typically used for nested command substitution) must be quoted by backslashes. In commands, backslashes are treated as quotes only when preceding a dollar ($), backquote, newline, or another backslash. Additionally, if the command substitution occurs inside double quotes, double quotes in commands must be quoted with a backslash. Those backslashes are removed before commands are parsed.

Arithmetic expansion

Arithmetic expansion evaluates an arithmetic expression and expands to the value of the expression.

Arithmetic expansion

$((expression))

When arithmetic expansion is expanded, the expression is subject to parameter expansion, command substitution, and (nested) arithmetic expansion. The expression is parsed in (almost) same manner as an expression of the C programming language.

Yash allows an expression to be either an integer (of the long type in C) or a floating-point number (of the double type in C). An operation on integers yields an integer and an operation involving a floating-point number yields a floating-point number. In the POSIXly-correct mode, you can use integers only.

The following operators are available (in the order of precedence):

  1. ( )

  2. ++ -- (postfix operators)

  3. ++ -- + - ~ ! (prefix operators)

  4. * / %

  5. + - (binary operators)

  6. << >>

  7. < <= > >=

  8. == !=

  9. &

  10. ^

  11. |

  12. &&

  13. ||

  14. ? :

  15. = *= /= %= += -= <<= >>= &= ^= |=

The ++ and -- operators cannot be used in the POSIXly-correct mode.

An atomic expression can be one of an integer literal, a floating-point number literal, and a variable. Literals are parsed in the same manner as in C. An octal integer literal starts with 0, and hexadecimal with 0x. A floating-point number literal may have an exponent (i.e. 1.23e+6). A variable with a non-numeric value will result in an error when parsed as a number. An unset variable is treated as a value of zero.

In the POSIXly-correct mode, variables are always parsed as numbers. Otherwise, variables are parsed only when they are used as numbers in computation. Unparsed variables are left intact.

set +o posixly-correct
foo=bar
echo $((0 ? foo : foo)) # prints "bar"
echo $((foo + 0))       # error

Brace expansion

Brace expansion expands to several split words with preceding and succeeding portions duplicated to each split words. Brace expansion is expanded only when the brace-expand option is enabled.

Comma-separated brace expansion

{word1,word2,…,wordn}

Range brace expansion

{start..end}

{start..end..delta}

Comma-separated brace expansion is expanded to each comma-separated word. For example, a{1,2,3}b is expanded to the three words a1b, a2b, and a3b.

Range brace expansion is expanded to integers in the range defined by start and end. The difference between each integer can be defined by delta. If start is larger than end, the results will be in descending order. When ..delta is omitted, it defaults to 1 or -1. For example, a{1..3}b is expanded to the three words a1b, a2b, and a3b; and a{1..7..2}b to the four words a1b, a3b, a5b, and a7b.

Multiple brace expansions can be used in one word. Brace expansions can also be nested. You can quote braces and/or commas to prevent them from being treated as brace expansion.

Any errors in brace expansion are silently ignored.

Field splitting

In field splitting, words are split at predefined separators.

Field splitting can occur only within parts of words that resulted from parameter expansion, command substitution, and arithmetic expansion that are not between double-quotation marks. Expansion results of the @ special parameter are exceptionally split even between double-quotation marks.

Separators used in field splitting are defined by the value of the IFS variable. If the variable does not exist, the value is assumed to be the three characters of space, tab, and newline.

Characters included in the value of the IFS variable are called IFS characters. IFS characters that are any of space, tab, and newline are called IFS whitespace and other IFS characters are called IFS non-whitespace.

Field splitting is performed as follows:

  1. The shell searches words for split points. A split point is one or more adjacent IFS characters within the word portions where field splitting can occur. The following steps are taken for each split point found.

  2. If the split point includes one or more IFS non-whitespaces, all the IFS whitespaces in the split point are ignored and the word is split at each IFS non-whitespace in the split point.

  3. If the split point includes no IFS non-whitespaces, the word is split at the split point unless it is at the beginning or end of the word.

  4. The split points are removed from the results.

Finally, the last word is removed from the results if:

Note
Words are not split at all when the value of the IFS variable is empty.

Pathname expansion

Pathname expansion performs pattern matching and expands to pathnames matched by the pattern.

A word subjected to pathname expansion is treated as a pattern. If one or more pathnames are found that are matched by the pattern, the pathnames become the results of the pathname expansion.

Pathname expansion is not performed when the glob option is disabled.

The shell searches readable directories for matching pathnames. Unreadable directories are silently ignored.

The following options affect the behavior of pathname expansion:

null-glob

This option affects the result of pathname expansion when no matching pathnames are found. If enabled, the result is no word. If disabled, the result is the original pattern word.

case-glob

This option specifies case-sensitivity in matching. If enabled, pattern matching is done case-sensitively.

dot-glob

This option affects matching of filenames that start with a period (.). If disabled, a period at the beginning of a filename does not match wildcard patterns (? and *) or bracket expressions. If enabled, there is no such special treatment of periods.

mark-dirs

If enabled, each resulting pathname that is a directory name is suffixed by a slash (/).

extended-glob

This option enables the extension.

Any errors in pathname expansion are silently ignored. If the word is an invalid pattern, it just becomes the result. The results depend on the null-glob option when no matching pathnames are found.

Pattern matching is done for each filename (or pathname component) of pathnames. The shell skips matching for literal patterns that contain no wildcards or bracket expressions. As a result, the patterns /*/foo and /*/fo[o] may yield different expansion results when the case-glob option is disabled; for example, the pattern /*/fo[o] matches the pathname /bar/FOO but the pattern /*/foo does not because matching is skipped for foo.

Extension in pathname expansion

The following patterns can be used when the extended-glob option is enabled.

**

The directory is searched recursively and the pattern matches any number of directory filenames (each separated by a slash). Any directory whose name begins with a period is excluded from search. For example, the pattern dir/**/file can match the pathnames dir/file, dir/foo/file, dir/a/b/c/file, etc.

This pattern is not effective when appearing at the end of the whole pattern (i.e. foo/bar/**).

.**

This pattern is like **, but all directories are searched including ones with a name starting with a period.

***

This pattern is like **, but if a symbolic link to a directory is found during recursive search, the directory is searched recursively as well.

.***

This pattern is like ***, but all directories are searched including ones with a name starting with a period.