The following sections explain the types of expansion in detail.
Immediately before execution, each command is saved in the history list, the size of which is controlled by the HISTSIZE parameter. The one most recent command is always retained in any case. Each saved command in the history list is called a history event and is assigned a number, beginning with 1 (one) when the shell starts up. The history number that you may see in your prompt (see EXPANSION OF PROMPT SEQUENCES in zshmisc(1)) is the number that is to be assigned to the next command.
The first character is followed by an optional event designator (see the section `Event Designators') and then an optional word designator (the section `Word Designators'); if neither of these designators is present, no history expansion occurs.
Input lines containing history expansions are echoed after being expanded, but before any other expansions take place and before the command is executed. It is this expanded form that is recorded as the history event for later references.
History expansions do not nest.
By default, a history reference with no event designator refers to the same event as any preceding history reference on that command line; if it is the only history reference in a command, it refers to the previous command. However, if the option CSH_JUNKIE_HISTORY is set, then every history reference with no event specification always refers to the previous command.
For example, `!' is the event designator for the previous command, so `!!:1' always refers to the first word of the previous command, and `!!$' always refers to the last word of the previous command. With CSH_JUNKIE_HISTORY set, then `!:1' and `!$' function in the same manner as `!!:1' and `!!$', respectively. Conversely, if CSH_JUNKIE_HISTORY is unset, then `!:1' and `!$' refer to the first and last words, respectively, of the same event referenced by the nearest other history reference preceding them on the current command line, or to the previous command if there is no preceding reference.
The character sequence `^foo^bar' (where `^' is actually the second character of the histchars parameter) repeats the last command, replacing the string foo with bar. More precisely, the sequence `^foo^bar^' is synonymous with `!!:s^foo^bar^', hence other modifiers (see the section `Modifiers') may follow the final `^'. In particular, `^foo^bar^:G' performs a global substitution.
If the shell encounters the character sequence `!"' in the input, the history mechanism is temporarily disabled until the current list (see zshmisc(1)) is fully parsed. The `!"' is removed from the input, and any subsequent `!' characters have no special significance.
A less convenient but more comprehensible form of command history support is provided by the fc builtin.
Note that a `%' word designator works only when used in one of `!%', `!:%' or `!?str?:%', and only when used after a !? expansion (possibly in an earlier command). Anything else results in an error, although the error may not be the most obvious one.
This transformation is agnostic about what is in the filesystem, i.e. is on the logical, not the physical directory. It takes place in the same manner as when changing directories when neither of the options CHASE_DOTS or CHASE_LINKS is set. For example, `/before/here/../after' is always transformed to `/before/after', regardless of whether `/before/here' exists or what kind of object (dir, file, symlink, etc.) it is.
Note: on systems that do not have a realpath(3) library function, symbolic links are not resolved, so on those systems `a' and `A' are equivalent.
Note: foo:A and realpath(foo) are different on some inputs. For realpath(foo) semantics, see the `P` modifier.
Unlike realpath(3), non-existent trailing components are permitted and preserved.
The forms `gs/l/r' and `s/l/r/:G' perform global substitution, i.e. substitute every occurrence of r for l. Note that the g or :G must appear in exactly the position shown.
See further notes on this form of substitution below.
The s/l/r/ substitution works as follows. By default the left-hand side of substitutions are not patterns, but character strings. Any character can be used as the delimiter in place of `/'. A backslash quotes the delimiter character. The character `&', in the right-hand-side r, is replaced by the text from the left-hand-side l. The `&' can be quoted with a backslash. A null l uses the previous string either from the previous l or from the contextual scan string s from `!?s'. You can omit the rightmost delimiter if a newline immediately follows r; the rightmost `?' in a context scan can similarly be omitted. Note the same record of the last l and r is maintained across all forms of expansion.
Note that if a `&' is used within glob qualifiers an extra backslash is needed as a & is a special character in this case.
Also note that the order of expansions affects the interpretation of l and r. When used in a history expansion, which occurs before any other expansions, l and r are treated as literal strings (except as explained for HIST_SUBST_PATTERN below). When used in parameter expansion, the replacement of r into the parameter's value is done first, and then any additional process, parameter, command, arithmetic, or brace references are applied, which may evaluate those substitutions and expansions more than once if l appears more than once in the starting value. When used in a glob qualifier, any substitutions or expansions are performed once at the time the qualifier is parsed, even before the `:s' expression itself is divided into l and r sides.
If the option HIST_SUBST_PATTERN is set, l is treated as a pattern of the usual form described in the section FILENAME GENERATION below. This can be used in all the places where modifiers are available; note, however, that in globbing qualifiers parameter substitution has already taken place, so parameters in the replacement string should be quoted to ensure they are replaced at the correct time. Note also that complicated patterns used in globbing qualifiers may need the extended glob qualifier notation (#q:s/.../.../) in order for the shell to recognize the expression as a glob qualifier. Further, note that bad patterns in the substitution are not subject to the NO_BAD_PATTERN option so will cause an error.
When HIST_SUBST_PATTERN is set, l may start with a # to indicate that the pattern must match at the start of the string to be substituted, and a % may appear at the start or after an # to indicate that the pattern must match at the end of the string to be substituted. The % or # may be quoted with two backslashes.
For example, the following piece of filename generation code with the EXTENDED_GLOB option:
print -r -- *.c(#q:s/#%(#b)s(*).c/'S${match[1]}.C'/)
takes the expansion of *.c and applies the glob qualifiers in the (#q...) expression, which consists of a substitution modifier anchored to the start and end of each word (#%). This turns on backreferences ((#b)), so that the parenthesised subexpression is available in the replacement string as ${match[1]}. The replacement string is quoted so that the parameter is not substituted before the start of filename generation.
The following f, F, w and W modifiers work only with parameter expansion and filename generation. They are listed here to provide a single point of reference for all modifiers.
Note that `<<(list)' is not a special syntax; it is equivalent to `< <(list)', redirecting standard input from the result of process substitution. Hence all the following documentation applies. The second form (with the space) is recommended for clarity.
In the case of the < or > forms, the shell runs the commands in list as a subprocess of the job executing the shell command line. If the system supports the /dev/fd mechanism, the command argument is the name of the device file corresponding to a file descriptor; otherwise, if the system supports named pipes (FIFOs), the command argument will be a named pipe. If the form with > is selected then writing on this special file will provide input for list. If < is used, then the file passed as an argument will be connected to the output of the list process. For example,
paste <(cut -f1 file1) <(cut -f3 file2) | tee >(process1) >(process2) >/dev/null
cuts fields 1 and 3 from the files file1 and file2 respectively, pastes the results together, and sends it to the processes process1 and process2.
If =(...) is used instead of <(...), then the file passed as an argument will be the name of a temporary file containing the output of the list process. This may be used instead of the < form for a program that expects to lseek (see lseek(2)) on the input file.
There is an optimisation for substitutions of the form =(<<<arg), where arg is a single-word argument to the here-string redirection <<<. This form produces a file name containing the value of arg after any substitutions have been performed. This is handled entirely within the current shell. This is effectively the reverse of the special form $(<arg) which treats arg as a file name and replaces it with the file's contents.
The = form is useful as both the /dev/fd and the named pipe implementation of <(...) have drawbacks. In the former case, some programmes may automatically close the file descriptor in question before examining the file on the command line, particularly if this is necessary for security reasons such as when the programme is running setuid. In the second case, if the programme does not actually open the file, the subshell attempting to read from or write to the pipe will (in a typical implementation, different operating systems may have different behaviour) block for ever and have to be killed explicitly. In both cases, the shell actually supplies the information using a pipe, so that programmes that expect to lseek (see lseek(2)) on the file will not work.
Also note that the previous example can be more compactly and efficiently written (provided the MULTIOS option is set) as:
paste <(cut -f1 file1) <(cut -f3 file2) \ > >(process1) > >(process2)
The shell uses pipes instead of FIFOs to implement the latter two process substitutions in the above example.
There is an additional problem with >(process); when this is attached to an external command, the parent shell does not wait for process to finish and hence an immediately following command cannot rely on the results being complete. The problem and solution are the same as described in the section MULTIOS in zshmisc(1). Hence in a simplified version of the example above:
paste <(cut -f1 file1) <(cut -f3 file2) > >(process)
(note that no MULTIOS are involved), process will be run asynchronously as far as the parent shell is concerned. The workaround is:
{ paste <(cut -f1 file1) <(cut -f3 file2) } > >(process)
The extra processes here are spawned from the parent shell which will wait for their completion.
Another problem arises any time a job with a substitution that requires a temporary file is disowned by the shell, including the case where `&!' or `&|' appears at the end of a command containing a substitution. In that case the temporary file will not be cleaned up as the shell no longer has any memory of the job. A workaround is to use a subshell, for example,
(mycmd =(myoutput)) &!
as the forked subshell will wait for the command to finish then remove the temporary file.
A general workaround to ensure a process substitution endures for an appropriate length of time is to pass it as a parameter to an anonymous shell function (a piece of shell code that is run immediately with function scope). For example, this code:
() {
print File $1:
cat $1
} =(print This be the verse)
outputs something resembling the following
File /tmp/zsh6nU0kS: This be the verse
The temporary file created by the process substitution will be deleted when the function exits.
Note in particular the fact that words of unquoted parameters are not automatically split on whitespace unless the option SH_WORD_SPLIT is set; see references to this option below for more details. This is an important difference from other shells. However, as in other shells, null words are elided from unquoted parameters' expansions.
With default options, after the assignments:
array=("first word" "" "third word")
scalar="only word"
then $array substitutes two words, `first word' and `third word', and $scalar substitutes a single word `only word'. Note that second element of array was elided. Scalar parameters can be elided too if their value is null (empty). To avoid elision, use quoting as follows: "$scalar" for scalars and "${array[@]}" or "${(@)array}" for arrays. (The last two forms are equivalent.)
Parameter expansions can involve flags, as in `${(@kv)aliases}', and other operators, such as `${PREFIX:-"/usr/local"}'. Parameter expansions can also be nested. These topics will be introduced below. The full rules are complicated and are noted at the end.
In the expansions discussed below that require a pattern, the form of the pattern is the same as that used for filename generation; see the section `Filename Generation'. Note that these patterns, along with the replacement text of any substitutions, are themselves subject to parameter expansion, command substitution, and arithmetic expansion. In addition to the following operations, the colon modifiers described in the section `Modifiers' in the section `History Expansion' can be applied: for example, ${i:s/foo/bar/} performs string substitution on the expansion of parameter $i.
In the following descriptions, `word' refers to a single word substituted on the command line, not necessarily a space delimited word.
If name is an array parameter, and the KSH_ARRAYS option is not set, then the value of each element of name is substituted, one element per word. Otherwise, the expansion results in one word only; with KSH_ARRAYS, this is the first element of an array. No field splitting is done on the result unless the SH_WORD_SPLIT option is set. See also the flags = and s:string:.
In any of the above expressions that test a variable and substitute an alternate word, note that you can use standard shell quoting in the word value to selectively override the splitting done by the SH_WORD_SPLIT option and the = flag, but not splitting by the s:string: flag.
In the following expressions, when name is an array and the substitution is not quoted, or if the `(@)' flag or the name[@] syntax is used, matching and replacement is performed on each array element separately.
a=(1 2 3 4); b=(a b); print ${a:^b}
will output `1 a 2 b'. For `:^^', then the input is repeated until all of the longer array has been used up and the above will output `1 a 2 b 3 a 4 b'.
Either or both inputs may be a scalar, they will be treated as an array of length 1 with the scalar as the only element. If either array is empty, the other array is output with no extra elements inserted.
Currently the following code will output `a b' and `1' as two separate elements, which can be unexpected. The second print provides a workaround which should continue to work if this is changed.
a=(a b); b=(1 2); print -l "${a:^b}"; print -l "${${a:^b}}"
If offset is non-negative, then if the variable name is a scalar substitute the contents starting offset characters from the first character of the string, and if name is an array substitute elements starting offset elements from the first element. If length is given, substitute that many characters or elements, otherwise the entire rest of the scalar or array.
A positive offset is always treated as the offset of a character or element in name from the first character or element of the array (this is different from native zsh subscript notation). Hence 0 refers to the first character or element regardless of the setting of the option KSH_ARRAYS.
A negative offset counts backwards from the end of the scalar or array, so that -1 corresponds to the last character or element, and so on.
When positive, length counts from the offset position toward the end of the scalar or array. When negative, length counts back from the end. If this results in a position smaller than offset, a diagnostic is printed and nothing is substituted.
The option MULTIBYTE is obeyed, i.e. the offset and length count multibyte characters where appropriate.
offset and length undergo the same set of shell substitutions as for scalar assignment; in addition, they are then subject to arithmetic evaluation. Hence, for example
print ${foo:3}
print ${foo: 1 + 2}
print ${foo:$(( 1 + 2))}
print ${foo:$(echo 1 + 2)}
all have the same effect, extracting the string starting at the fourth character of $foo if the substitution would otherwise return a scalar, or the array starting at the fourth element if $foo would return an array. Note that with the option KSH_ARRAYS $foo always returns a scalar (regardless of the use of the offset syntax) and a form such as ${foo[*]:3} is required to extract elements of an array named foo.
If offset is negative, the - may not appear immediately after the : as this indicates the ${name:-word} form of substitution. Instead, a space may be inserted before the -. Furthermore, neither offset nor length may begin with an alphabetic character or & as these are used to indicate history-style modifiers. To substitute a value from a variable, the recommended approach is to precede it with a $ as this signifies the intention (parameter substitution can easily be rendered unreadable); however, as arithmetic substitution is performed, the expression ${var: offs} does work, retrieving the offset from $offs.
For further compatibility with other shells there is a special case for array offset 0. This usually accesses the first element of the array. However, if the substitution refers to the positional parameter array, e.g. $@ or $*, then offset 0 instead refers to $0, offset 1 refers to $1, and so on. In other words, the positional parameter array is effectively extended by prepending $0. Hence ${*:0:1} substitutes $0 and ${*:1:1} substitutes $1.
The pattern may begin with a `#', in which case the pattern must match at the start of the string, or `%', in which case it must match at the end of the string, or `#%' in which case the pattern must match the entire string. The repl may be an empty string, in which case the final `/' may also be omitted. To quote the final `/' in other cases it should be preceded by a single backslash; this is not necessary if the `/' occurs inside a substituted parameter. Note also that the `#', `%' and `#% are not active if they occur inside a substituted parameter, even at the start.
If, after quoting rules apply, ${name} expands to an array, the replacements act on each element individually. Note also the effect of the I and S parameter expansion flags below; however, the flags M, R, B, E and N are not useful.
For example,
foo="twinkle twinkle little star" sub="t*e" rep="spy"
print ${foo//${~sub}/$rep}
print ${(S)foo//${~sub}/$rep}
Here, the `~' ensures that the text of $sub is treated as a pattern rather than a plain string. In the first case, the longest match for t*e is substituted and the result is `spy star', while in the second case, the shortest matches are taken and the result is `spy spy lispy star'.
If the option POSIX_IDENTIFIERS is not set, and spec is a simple name, then the braces are optional; this is true even for special parameters so e.g. $#- and $#* take the length of the string $- and the array $* respectively. If POSIX_IDENTIFIERS is set, then braces are required for the # to be treated in this fashion.
Internally, each such expansion is converted into the equivalent list for brace expansion. E.g., ${^var} becomes {$var[1],$var[2],...}, and is processed as described in the section `Brace Expansion' below: note, however, the expansion happens immediately, with any explicit brace expansion happening later. If word splitting is also in effect the $var[N] may themselves be split into different list elements.
Note that splitting is applied to word in the assignment forms of spec before the assignment to name is performed. This affects the result of array assignments with the A flag.
In nested substitutions, note that the effect of the ~ applies to the result of the current level of substitution. A surrounding pattern operation on the result may cancel it. Hence, for example, if the parameter foo is set to *, ${~foo//\*/*.c} is substituted by the pattern *.c, which may be expanded by filename generation, but ${${~foo}//\*/*.c} substitutes to the string *.c, which will not be further expanded.
If a ${...} type parameter expression or a $(...) type command substitution is used in place of name above, it is expanded first and the result is used as if it were the value of name. Thus it is possible to perform nested operations: ${${foo#head}%tail} substitutes the value of $foo with both `head' and `tail' deleted. The form with $(...) is often useful in combination with the flags described next; see the examples below. Each name or nested ${...} in a parameter expansion may also be followed by a subscript expression as described in Array Parameters in zshparam(1).
Note that double quotes may appear around nested expressions, in which case only the part inside is treated as quoted; for example, ${(f)"$(foo)"} quotes the result of $(foo), but the flag `(f)' (see below) is applied using the rules for unquoted expansions. Note further that quotes are themselves nested in this context; for example, in "${(@f)"$(foo)"}", there are two sets of quotes, one surrounding the whole expression, the other (redundant) surrounding the $(foo) as before.
If the MULTIBYTE option is set and the number is greater than 127 (i.e. not an ASCII character) it is treated as a Unicode character.
This assigns an array parameter with `${...=...}', `${...:=...}' or `${...::=...}'. If this flag is repeated (as in `AA'), assigns an associative array parameter. Assignment is made before sorting or padding; if field splitting is active, the word part is split before assignment. The name part may be a subscripted range for ordinary arrays; when assigning an associative array, the word part must be converted to an array, for example by using `${(AA)=name=...}' to activate field splitting.
Surrounding context such as additional nesting or use of the value in a scalar assignment may cause the array to be joined back into a single string again.
Quoting using one of the q family of flags does not work for this purpose since quotes are not stripped from non-pattern characters by GLOB_SUBST. In other words,
pattern=${(q)str}
[[ $str = ${~pattern} ]]
works if $str is `a*b' but not if it is `a b', whereas
pattern=${(b)str}
[[ $str = ${~pattern} ]]
is always true for any possible value of $str.
If used with a nested parameter or command substitution, the result of that will be taken as a parameter name in the same way. For example, if you have `foo=bar' and `bar=baz', the strings ${(P)foo}, ${(P)${foo}}, and ${(P)$(echo bar)} will be expanded to `baz'.
Likewise, if the reference is itself nested, the expression with the flag is treated as if it were directly replaced by the parameter name. It is an error if this nested substitution produces an array with more than one word. For example, if `name=assoc' where the parameter assoc is an associative array, then `${${(P)name}[elt]}' refers to the element of the associative subscripted `elt'.
If this flag is given twice, the resulting words are quoted in single quotes and if it is given three times, the words are quoted in double quotes; in these forms no special handling of unprintable or invalid characters is attempted. If the flag is given four times, the words are quoted in single quotes preceded by a $. Note that in all three of these forms quoting is done unconditionally, even if this does not change the way the resulting string would be interpreted by the shell.
If a q- is given (only a single q may appear), a minimal form of single quoting is used that only quotes the string if needed to protect special characters. Typically this form gives the most readable output.
If a q+ is given, an extended form of minimal quoting is used that causes unprintable characters to be rendered using $'...'. This quoting is similar to that used by the output of values by the typeset family of commands.
Note that this is done very late, even later than the `(s)' flag. So to access single words in the result use nested expansions as in `${${(z)foo}[2]}'. Likewise, to remove the quotes in the resulting words use `${(Q)${(z)foo}}'.
The following flags (except p) are followed by one or more arguments as shown. Any character, or the matching pairs `(...)', `{...}', `[...]', or `<...>', may be used in place of a colon as delimiters, but note that when a flag takes more than one argument, a matched pair of delimiters must surround each argument.
Alternatively, with this option string arguments may be in the form $var in which case the value of the variable is substituted. Note this form is strict; the string argument does not undergo general parameter expansion.
For example,
sep=:
val=a:b:c
print ${(ps.$sep.)val}
splits the variable on a :.
[[ "?" = ${(~j.|.)array} ]]
treats `|' as a pattern and succeeds if and only if $array contains the string `?' as an element. The ~ may be repeated to toggle the behaviour; its effect only lasts to the end of the parenthesised group.
The arguments :string1: and :string2: are optional; neither, the first, or both may be given. Note that the same pairs of delimiters must be used for each of the three arguments. The space to the left will be filled with string1 (concatenated as often as needed) or spaces if string1 is not given. If both string1 and string2 are given, string2 is inserted once directly to the left of each word, truncated if necessary, before string1 is used to produce any remaining padding.
If either of string1 or string2 is present but empty, i.e. there are two delimiters together at that point, the first character of $IFS is used instead.
If the MULTIBYTE option is in effect, the flag m may also be given, in which case widths will be used for the calculation of padding; otherwise individual multibyte characters are treated as occupying one unit of width.
If the MULTIBYTE option is not in effect, each byte in the string is treated as occupying one unit of width.
Control characters are always assumed to be one unit wide; this allows the mechanism to be used for generating repetitions of control characters.
If the m is repeated, the character either counts zero (if it has zero width), else one. For printable character strings this has the effect of counting the number of glyphs (visibly separate characters), except for the case where combining characters themselves have non-zero width (true in certain alphabets).
Left and right padding may be used together. In this case the strategy is to apply left padding to the first half width of each of the resulting words, and right padding to the second half. If the string to be padded has odd width the extra padding is applied on the left.
For historical reasons, the usual behaviour that empty array elements are retained inside double quotes is disabled for arrays generated by splitting; hence the following:
line="one::three"
print -l "${(s.:.)line}"
produces two lines of output for one and three and elides the empty field. To override this behaviour, supply the `(@)' flag as well, i.e. "${(@s.:.)line}".
The following flags are meaningful with the ${...#...} or ${...%...} forms. The S and I flags may also be used with the ${.../...} forms.
% str="aXbXc"
% echo ${(S)str#X*}
abXc
% echo ${(S)str##X*}
a
%
With % or %%, search for the match that starts closest to the end of the string:
% str="aXbXc"
% echo ${(S)str%X*}
aXbc
% echo ${(S)str%%X*}
aXb
%
(Note that % and %% don't search for the match that ends closest to the end of the string, as one might expect.)
With substitution via ${.../...} or ${...//...}, specifies non-greedy matching, i.e. that the shortest instead of the longest match should be replaced:
% str="abab"
% echo ${str/*b/_}
_
% echo ${(S)str/*b/_}
_ab
%
The exprth match is counted such that there is either one or zero matches from each starting position in the string, although for global substitution matches overlapping previous replacements are ignored. With the ${...%...} and ${...%%...} forms, the starting position for the match moves backwards from the end as the index increases, while with the other forms it moves forward from the start.
Hence with the string
which switch is the right switch for Ipswich?
Here is a summary of the rules for substitution; this assumes that braces are present around the substitution, i.e. ${...}. Some particular examples are given below. Note that the Zsh Development Group accepts no responsibility for any brain damage which may occur during the reading of the following rules.
Note that, unless the `(P)' flag is present, the flags and any subscripts apply directly to the value of the nested substitution; for example, the expansion ${${foo}} behaves exactly the same as ${foo}. When the `(P)' flag is present in a nested substitution, the other substitution rules are applied to the value before it is interpreted as a name, so ${${(P)foo}} may differ from ${(P)foo}.
At each nested level of substitution, the substituted words undergo all forms of single-word substitution (i.e. not filename generation), including command substitution, arithmetic expansion and filename expansion (i.e. leading ~ and =). Thus, for example, ${${:-=cat}:h} expands to the directory where the cat program resides. (Explanation: the internal substitution has no parameter but a default value =cat, which is expanded by filename expansion to a full path; the outer substitution then applies the modifier :h and takes the directory part of the path.)
At the outermost level of substitution, the `(P)' flag (rule 4.) ignores these transformations and uses the unmodified value of the parameter as the name to be replaced. This is usually the desired behavior because padding may make the value syntactically illegal as a parameter name, but if capitalization changes are desired, use the ${${(P)foo}} form (rule 25.).
If the value so far names a parameter that has internal flags (rule 2.), those internal flags are applied to the new value after replacement.
If no `(s)', `(f)' or `=' was given, but the word is not quoted and the option SH_WORD_SPLIT is set, the word is split on occurrences of any of the characters in $IFS. Note this step, too, takes place at all levels of a nested substitution.
If a single word is not required, this rule is skipped.
Strictly speaking, the removal happens later as the same happens with other forms of substitution; the point to note here is simply that it occurs after any of the above parameter operations.
The following illustrates the rules for nested parameter expansions. Suppose that $foo contains the array (bar baz):
As an example of the rules for word splitting and joining, suppose $foo contains the array `(ax1 bx1)'. Then
The substitution `$(cat foo)' may be replaced by the faster `$(<foo)'. In this case foo undergoes single word shell expansions (parameter expansion, command substitution and arithmetic expansion), but not filename generation.
If the option GLOB_SUBST is set, the result of any unquoted command substitution, including the special form just mentioned, is eligible for filename generation.
An expression of the form `{n1..n2}', where n1 and n2 are integers, is expanded to every number between n1 and n2 inclusive. If either number begins with a zero, all the resulting numbers will be padded with leading zeroes to that minimum width, but for negative numbers the - character is also included in the width. If the numbers are in decreasing order the resulting sequence will also be in decreasing order.
An expression of the form `{n1..n2..n3}', where n1, n2, and n3 are integers, is expanded as above, but only every n3th number starting from n1 is output. If n3 is negative the numbers are output in reverse order, this is slightly different from simply swapping n1 and n2 in the case that the step n3 doesn't evenly divide the range. Zero padding can be specified in any of the three numbers, specifying it in the third can be useful to pad for example `{-99..100..01}' which is not possible to specify by putting a 0 on either of the first two numbers (i.e. pad to two characters).
An expression of the form `{c1..c2}', where c1 and c2 are single characters (which may be multibyte characters), is expanded to every character in the range from c1 to c2 in whatever character sequence is used internally. For characters with code points below 128 this is US ASCII (this is the only case most users will need). If any intervening character is not printable, appropriate quotation is used to render it printable. If the character sequence is reversed, the output is in reverse order, e.g. `{d..a}' is substituted as `d c b a'.
If a brace expression matches none of the above forms, it is left unchanged, unless the option BRACE_CCL (an abbreviation for `brace character class') is set. In that case, it is expanded to a list of the individual characters between the braces sorted into the order of the characters in the ASCII character set (multibyte characters are not currently handled). The syntax is similar to a [...] expression in filename generation: `-' is treated specially to denote a range of characters, but `^' or `!' as the first character is treated normally. For example, `{abcdef0-9}' expands to 16 words 0 1 2 3 4 5 6 7 8 9 a b c d e f.
Note that brace expansion is not part of filename generation (globbing); an expression such as */{foo,bar} is split into two separate words */foo and */bar before filename generation takes place. In particular, note that this is liable to produce a `no match' error if either of the two expressions does not match; this is to be contrasted with */(foo|bar), which is treated as a single pattern but otherwise has similar effects.
To combine brace expansion with array expansion, see the ${^spec} form described in the section Parameter Expansion above.
A `~' by itself is replaced by the value of $HOME. A `~' followed by a `+' or a `-' is replaced by current or previous working directory, respectively.
A `~' followed by a number is replaced by the directory at that position in the directory stack. `~0' is equivalent to `~+', and `~1' is the top of the stack. `~+' followed by a number is replaced by the directory at that position in the directory stack. `~+0' is equivalent to `~+', and `~+1' is the top of the stack. `~-' followed by a number is replaced by the directory that many positions from the bottom of the stack. `~-0' is the bottom of the stack. The PUSHD_MINUS option exchanges the effects of `~+' and `~-' where they are followed by a number.
If the function zsh_directory_name exists, or the shell variable zsh_directory_name_functions exists and contains an array of function names, then the functions are used to implement dynamic directory naming. The functions are tried in order until one returns status zero, so it is important that functions test whether they can handle the case in question and return an appropriate status.
A `~' followed by a string namstr in unquoted square brackets is treated specially as a dynamic directory name. Note that the first unquoted closing square bracket always terminates namstr. The shell function is passed two arguments: the string n (for name) and namstr. It should either set the array reply to a single element which is the directory corresponding to the name and return status zero (executing an assignment as the last statement is usually sufficient), or it should return status non-zero. In the former case the element of reply is used as the directory; in the latter case the substitution is deemed to have failed. If all functions fail and the option NOMATCH is set, an error results.
The functions defined as above are also used to see if a directory can be turned into a name, for example when printing the directory stack or when expanding %~ in prompts. In this case each function is passed two arguments: the string d (for directory) and the candidate for dynamic naming. The function should either return non-zero status, if the directory cannot be named by the function, or it should set the array reply to consist of two elements: the first is the dynamic name for the directory (as would appear within `~[...]'), and the second is the prefix length of the directory to be replaced. For example, if the trial directory is /home/myname/src/zsh and the dynamic name for /home/myname/src (which has 16 characters) is s, then the function sets
reply=(s 16)
The directory name so returned is compared with possible static names for parts of the directory path, as described below; it is used if the prefix length matched (16 in the example) is longer than that matched by any static name.
It is not a requirement that a function implements both n and d calls; for example, it might be appropriate for certain dynamic forms of expansion not to be contracted to names. In that case any call with the first argument d should cause a non-zero status to be returned.
The completion system calls `zsh_directory_name c' followed by equivalent calls to elements of the array zsh_directory_name_functions, if it exists, in order to complete dynamic names for directories. The code for this should be as for any other completion function as described in zshcompsys(1).
As a working example, here is a function that expands any dynamic names beginning with the string p: to directories below /home/pws/perforce. In this simple case a static name for the directory would be just as effective.
zsh_directory_name() {
emulate -L zsh
setopt extendedglob
local -a match mbegin mend
if [[ $1 = d ]]; then
# turn the directory into a name
if [[ $2 = (#b)(/home/pws/perforce/)([^/]##)* ]]; then
typeset -ga reply
reply=(p:$match[2] $(( ${#match[1]} + ${#match[2]} )) )
else
return 1
fi
elif [[ $1 = n ]]; then
# turn the name into a directory
[[ $2 != (#b)p:(?*) ]] && return 1
typeset -ga reply
reply=(/home/pws/perforce/$match[1])
elif [[ $1 = c ]]; then
# complete names
local expl
local -a dirs
dirs=(/home/pws/perforce/*(/:t))
dirs=(p:${^dirs})
_wanted dynamic-dirs expl 'dynamic directory' compadd -S\] -a dirs
return
else
return 1
fi
return 0
}
It is also possible to define directory names using the -d option to the hash builtin.
When the shell prints a path (e.g. when expanding %~ in prompts or when printing the directory stack), the path is checked to see if it has a named directory as its prefix. If so, then the prefix portion is replaced with a `~' followed by the name of the directory. The shorter of the two ways of referring to the directory is used, i.e. either the directory name or the full path; the name is used if they are the same length. The parameters $PWD and $OLDPWD are never abbreviated in this fashion.
If a word begins with an unquoted `=' and the EQUALS option is set, the remainder of the word is taken as the name of a command. If a command exists by that name, the word is replaced by the full pathname of the command.
Filename expansion is performed on the right hand side of a parameter assignment, including those appearing after commands of the typeset family. In this case, the right hand side will be treated as a colon-separated list in the manner of the PATH parameter, so that a `~' or an `=' following a `:' is eligible for expansion. All such behaviour can be disabled by quoting the `~', the `=', or the whole expression (but not simply the colon); the EQUALS option is also respected.
If the option MAGIC_EQUAL_SUBST is set, any unquoted shell argument in the form `identifier=expression' becomes eligible for file expansion as described in the previous paragraph. Quoting the first `=' also inhibits this.
The word is replaced with a list of sorted filenames that match the pattern. If no matching pattern is found, the shell gives an error message, unless the NULL_GLOB option is set, in which case the word is deleted; or unless the NOMATCH option is unset, in which case the word is left unchanged.
In filename generation, the character `/' must be matched explicitly; also, a `.' must be matched explicitly at the beginning of a pattern or after a `/', unless the GLOB_DOTS option is set. No filename generation pattern matches the files `.' or `..'. In other instances of pattern matching, the `/' and `.' are not treated specially.
Another set of named classes is handled internally by the shell and is not sensitive to the locale:
Note that the square brackets are additional to those enclosing the whole set of characters, so to test for a single alphanumeric character you need `[[:alnum:]]'. Named character sets can be used alongside other types, e.g. `[[:alpha:]0-9]'.
Be careful when using other wildcards adjacent to patterns of this form; for example, <0-9>* will actually match any number whatsoever at the start of the string, since the `<0-9>' will match the first digit, and the `*' will match any others. This is a trap for the unwary, but is in fact an inevitable consequence of the rule that the longest possible match always succeeds. Expressions such as `<0-9>[^[:digit:]]*' can be used instead.
Note that grouping cannot extend over multiple directories: it is an error to have a `/' within a group (this only applies for patterns used in filename generation). There is one exception: a group of the form (pat/)# appearing as a complete path segment can match a sequence of directories. For example, foo/(a*/)#bar matches foo/bar, foo/any/bar, foo/any/anyother/bar, and so on.
For example,
foo="a_string_with_a_message"
if [[ $foo = (a|an)_(#b)(*) ]]; then
print ${foo[$mbegin[1],$mend[1]]}
fi
prints `string_with_a_message'. Note that the first set of parentheses is before the (#b) and does not create a backreference.
Backreferences work with all forms of pattern matching other than filename generation, but note that when performing matches on an entire array, such as ${array#pattern}, or a global substitution, such as ${param//pat/repl}, only the data for the last match remains available. In the case of global replacements this may still be useful. See the example for the m flag below.
The numbering of backreferences strictly follows the order of the opening parentheses from left to right in the pattern string, although sets of parentheses may be nested. There are special rules for parentheses followed by `#' or `##'. Only the last match of the parenthesis is remembered: for example, in `[[ abab = (#b)([ab])# ]]', only the final `b' is stored in match[1]. Thus extra parentheses may be necessary to match the complete segment: for example, use `X((ab|cd)#)Y' to match a whole string of either `ab' or `cd' between `X' and `Y', using the value of $match[1] rather than $match[2].
If the match fails none of the parameters is altered, so in some cases it may be necessary to initialise them beforehand. If some of the backreferences fail to match -- which happens if they are in an alternate branch which fails to match, or if they are followed by # and matched zero times -- then the matched string is set to the empty string, and the start and end indices are set to -1.
Pattern matching with backreferences is slightly slower than without.
For example,
arr=(veldt jynx grimps waqf zho buck)
print ${arr//(#m)[aeiou]/${(U)MATCH}}
forces all the matches (i.e. all vowels) into uppercase, printing `vEldt jynx grImps wAqf zhO bUck'.
Unlike backreferences, there is no speed penalty for using match references, other than the extra substitutions required for the replacement strings in cases such as the example shown.
Another use is in parameter substitution; for example `${array/(#s)A*Z(#e)}' will remove only elements of an array which match the complete pattern `A*Z'. There are other ways of performing many operations of this type, however the combination of the substitution operations `/' and `//' with the `(#s)' and `(#e)' flags provides a single simple and memorable method.
Note that assertions of the form `(^(#s))' also work, i.e. match anywhere except at the start of the string, although this actually means `anything except a zero-length portion at the start of the string'; you need to use `(""~(#s))' to match a zero-length portion of the string not at the start.
For example, the test string fooxx can be matched by the pattern (#i)FOOXX, but not by (#l)FOOXX, (#i)FOO(#I)XX or ((#i)FOOX)X. The string (#ia2)readme specifies case-insensitive matching of readme with up to two errors.
When using the ksh syntax for grouping both KSH_GLOB and EXTENDED_GLOB must be set and the left parenthesis should be preceded by @. Note also that the flags do not affect letters inside [...] groups, in other words (#i)[a-z] still matches only lowercase letters. Finally, note that when examining whole paths case-insensitively every directory must be searched for all files which match, so that a pattern of the form (#i)/foo/bar/... is potentially slow.
Thus, the pattern (#a3)abcd matches dcba, with the errors occurring by using the first rule twice and the second once, grouping the string as [d][cb][a] and [a][bc][d].
Non-literal parts of the pattern must match exactly, including characters in character ranges: hence (#a1)??? matches strings of length four, by applying rule 4 to an empty part of the pattern, but not strings of length two, since all the ? must match. Other characters which must match exactly are initial dots in filenames (unless the GLOB_DOTS option is set), and all slashes in filenames, so that a/bc is two errors from ab/c (the slash cannot be transposed with another character). Similarly, errors are counted separately for non-contiguous strings in the pattern, so that (ab|cd)ef is two errors from aebf.
When using exclusion via the ~ operator, approximate matching is treated entirely separately for the excluded part and must be activated separately. Thus, (#a1)README~READ_ME matches READ.ME but not READ_ME, as the trailing READ_ME is matched without approximation. However, (#a1)README~(#a1)READ_ME does not match any pattern of the form READ?ME as all such forms are now excluded.
Apart from exclusions, there is only one overall error count; however, the maximum errors allowed may be altered locally, and this can be delimited by grouping. For example, (#a1)cat((#a0)dog)fox allows one error in total, which may not occur in the dog section, and the pattern (#a1)cat(#a0)dog(#a1)fox is equivalent. Note that the point at which an error is first found is the crucial one for establishing whether to use approximation; for example, (#a1)abc(#a0)xyz will not match abcdxyz, because the error occurs at the `x', where approximation is turned off.
Entire path segments may be matched approximately, so that `(#a1)/foo/d/is/available/at/the/bar' allows one error in any path segment. This is much less efficient than without the (#a1), however, since every directory in the path must be scanned for a possible approximate match. It is best to place the (#a1) after any path segments which are known to be correct.
As a shorthand, `**/' is equivalent to `(*/)#'; note that this therefore matches files in the current directory as well as subdirectories. Thus:
ls -ld -- (*/)#bar
or
ls -ld -- **/bar
does a recursive directory search for files named `bar' (potentially including the file `bar' in the current directory). This form does not follow symbolic links; the alternative form `***/' does, but is otherwise identical. Neither of these can be combined with other forms of globbing within the same path segment; in that case, the `*' operators revert to their usual effect.
Even shorter forms are available when the option GLOB_STAR_SHORT is set. In that case if no / immediately follows a ** or *** they are treated as if both a / plus a further * are present. Hence:
setopt GLOBSTARSHORT ls -ld -- **.c
is equivalent to
ls -ld -- **/*.c
If the option BARE_GLOB_QUAL is set, then a trailing set of parentheses containing no `|' or `(' characters (or `~' if it is special) is taken as a set of glob qualifiers. A glob subexpression that would normally be taken as glob qualifiers, for example `(^x)', can be forced to be treated as part of the glob pattern by doubling the parentheses, in this case producing `((^x))'.
If the option EXTENDED_GLOB is set, a different syntax for glob qualifiers is available, namely `(#qx)' where x is any of the same glob qualifiers used in the other format. The qualifiers must still appear at the end of the pattern. However, with this syntax multiple glob qualifiers may be chained together. They are treated as a logical AND of the individual sets of flags. Also, as the syntax is unambiguous, the expression will be treated as glob qualifiers just as long any parentheses contained within it are balanced; appearance of `|', `(' or `~' does not negate the effect. Note that qualifiers will be recognised in this form even if a bare glob qualifier exists at the end of the pattern, for example `*(#q*)(.)' will recognise executable regular files if both options are set; however, mixed syntax should probably be avoided for the sake of clarity. Note that within conditions using the `[[' form the presence of a parenthesised expression (#q...) at the end of a string indicates that globbing should be performed; the expression may include glob qualifiers, but it is also valid if it is simply (#q). This does not apply to the right hand side of pattern match operators as the syntax already has special significance.
A qualifier may be any one of the following:
If the qualifier `f' is followed by any other character anything up to the next matching character (`[', `{', and `<' match `]', `}', and `>' respectively, any other character matches itself) is taken as a list of comma-separated sub-specs. Each sub-spec may be either an octal number as described above or a list of any of the characters `u', `g', `o', and `a', followed by a `=', a `+', or a `-', followed by a list of any of the characters `r', `w', `x', `s', and `t', or an octal digit. The first list of characters specify which access rights are to be checked. If a `u' is given, those for the owner of the file are used, if a `g' is given, those of the group are checked, a `o' means to test those of other users, and the `a' says to test all three groups. The `=', `+', and `-' again says how the modes are to be checked and have the same meaning as described for the first form above. The second list of characters finally says which access rights are to be expected: `r' for read access, `w' for write access, `x' for the right to execute the file (or to search a directory), `s' for the setuid and setgid bits, and `t' for the sticky bit.
Thus, `*(f70?)' gives the files for which the owner has read, write, and execute permission, and for which other group members have no rights, independent of the permissions for other users. The pattern `*(f-100)' gives all files for which the owner does not have execute permission, and `*(f:gu+w,o-rx:)' gives the files for which the owner and the other members of the group have at least write permission, and for which other users don't have read or execute permission.
In the first form, the first character after the `e' will be used as a separator and anything up to the next matching separator will be taken as the string; `[', `{', and `<' match `]', `}', and `>', respectively, while any other character matches itself. Note that expansions must be quoted in the string to prevent them from being expanded before globbing is done. string is then executed as shell code. The string globqual is appended to the array zsh_eval_context the duration of execution.
During the execution of string the filename currently being tested is available in the parameter REPLY; the parameter may be altered to a string to be inserted into the list instead of the original filename. In addition, the parameter reply may be set to an array or a string, which overrides the value of REPLY. If set to an array, the latter is inserted into the command line word by word.
For example, suppose a directory contains a single file `lonely'. Then the expression `*(e:'reply=(${REPLY}{1,2})':)' will cause the words `lonely1' and `lonely2' to be inserted into the command line. Note the quoting of string.
The form +cmd has the same effect, but no delimiters appear around cmd. Instead, cmd is taken as the longest sequence of characters following the + that are alphanumeric or underscore. Typically cmd will be the name of a shell function that contains the appropriate test. For example,
nt() { [[ $REPLY -nt $NTREF ]] }
NTREF=reffile
ls -ld -- *(+nt)
lists all files in the directory that have been modified more recently than reffile.
Any fractional part of the difference between the access time and the current part in the appropriate units is ignored in the comparison. For instance, `echo *(ah-5)' would echo files accessed within the last five hours, while `echo *(ah+5)' would echo files accessed at least six hours ago, as times strictly between five and six hours are treated as five hours.
If this flag is directly followed by a size specifier `k' (`K'), `m' (`M'), or `p' (`P') (e.g. `Lk-50') the check is performed with kilobytes, megabytes, or blocks (of 512 bytes) instead. (On some systems additional specifiers are available for gigabytes, `g' or `G', and terabytes, `t' or `T'.) If a size specifier is used a file is regarded as "exactly" the size if the file size rounded up to the next unit is equal to the test size. Hence `*(Lm1)' matches files from 1 byte up to 1 Megabyte inclusive. Note also that the set of files "less than" the test size only includes files that would not match the equality test; hence `*(Lm-1)' only matches files of zero size.
Implies oN when no oc qualifier is used.
The default sorting is n (by name) unless the Y glob qualifier is used, in which case it is N (unsorted).
oe and o+ are special cases; they are each followed by shell code, delimited as for the e glob qualifier and the + glob qualifier respectively (see above). The code is executed for each matched file with the parameter REPLY set to the name of the file on entry and globsort appended to zsh_eval_context. The code should modify the parameter REPLY in some fashion. On return, the value of the parameter is used instead of the file name as the string on which to sort. Unlike other sort operators, oe and o+ may be repeated, but note that the maximum number of sort operators of any kind that may appear in any glob expression is 12.
A typical use for this is to prepend an option before all occurrences of a file name; for example, the pattern `*(P:-f:)' produces the command line arguments `-f file1 -f file2 ...'
If the modifier ^ is active, then string will be appended instead of prepended. Prepending and appending is done independently so both can be used on the same glob expression; for example by writing `*(P:foo:^P:bar:^P:baz:)' which produces the command line arguments `foo baz file1 bar ...'
More than one of these lists can be combined, separated by commas. The whole list matches if at least one of the sublists matches (they are `or'ed, the qualifiers in the sublists are `and'ed). Some qualifiers, however, affect all matches generated, independent of the sublist in which they are given. These are the qualifiers `M', `T', `N', `D', `n', `o', `O' and the subscripts given in brackets (`[...]').
If a `:' appears in a qualifier list, the remainder of the expression in parenthesis is interpreted as a modifier (see the section `Modifiers' in the section `History Expansion'). Each modifier must be introduced by a separate `:'. Note also that the result after modification does not have to be an existing file. The name of any existing file can be followed by a modifier of the form `(:...)' even if no actual filename generation is performed, although note that the presence of the parentheses causes the entire expression to be subjected to any global pattern matching options such as NULL_GLOB. Thus:
ls -ld -- *(-/)
lists all directories and symbolic links that point to directories, and
ls -ld -- *(-@)
lists all broken symbolic links, and
ls -ld -- *(%W)
lists all world-writable device files in the current directory, and
ls -ld -- *(W,X)
lists all files in the current directory that are world-writable or world-executable, and
print -rC1 /tmp/foo*(u0^@:t)
outputs the basename of all root-owned files beginning with the string `foo' in /tmp, ignoring symlinks, and
ls -ld -- *.*~(lex|parse).[ch](^D^l1)
lists all files having a link count of one whose names contain a dot (but not those starting with a dot, since GLOB_DOTS is explicitly switched off) except for lex.c, lex.h, parse.c and parse.h.
print -rC1 b*.pro(#q:s/pro/shmo/)(#q.:s/builtin/shmiltin/)
demonstrates how colon modifiers and other qualifiers may be chained together. The ordinary qualifier `.' is applied first, then the colon modifiers in order from left to right. So if EXTENDED_GLOB is set and the base pattern matches the regular file builtin.pro, the shell will print `shmiltin.shmo'.