path: root/Doc/Zsh/expn.yo

texinode(Expansion)(Parameters)(Prompt Expansion)(Top)
chapter(Expansion)
cindex(expansion)
ifnztexi(sect(Description))
The following types of expansions are performed in the indicated order in
five steps:

startitem()
item(em(History Expansion))(
This is performed only in interactive shells.
)
item(em(Alias Expansion))(
Aliases are expanded immediately before the command line is parsed as
explained
ifzman(under Aliasing in zmanref(zshmisc))\
ifnzman(in noderef(Aliasing))\
.
)
xitem(em(Process Substitution))
xitem(em(Parameter Expansion))
xitem(em(Command Substitution))
xitem(em(Arithmetic Expansion))
item(em(Brace Expansion))(
These five are performed in one step in left-to-right fashion.  After
these expansions, all unquoted occurrences of the characters `tt(\)',
`tt(')' and `tt(")' are removed.
)
item(em(Filename Expansion))(
If the tt(SH_FILE_EXPANSION) option is set, the order of expansion is
modified for compatibility with bf(sh) and bf(ksh).  In that case
em(filename expansion) is performed immediately after em(alias expansion),
preceding the set of five expansions mentioned above.
)
cindex(globbing)
item(em(Filename Generation))(
This expansion, commonly referred to as bf(globbing), is always done last.
)
enditem()

The following sections explain the types of expansion in detail.

startmenu()
menu(History Expansion)
menu(Process Substitution)
menu(Parameter Expansion)
menu(Command Substitution)
menu(Arithmetic Expansion)
menu(Brace Expansion)
menu(Filename Expansion)
menu(Filename Generation)
endmenu()
texinode(History Expansion)(Process Substitution)()(Expansion)
sect(History Expansion)
cindex(history)
cindex(history expansion)
cindex(expansion, history)
History expansion allows you to use words from previous command
lines in the command line you are typing.  This simplifies spelling
corrections and the repetition of complicated commands or arguments.
vindex(HISTSIZE, use of)
Immediately before execution, each command is saved in the history list,
the size of which is controlled by the tt(HISTSIZE) parameter.  The one
most recent command is always retained in any case.  Each saved command in
the history list is called a history em(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
ifzman(EXPANSION OF PROMPT SEQUENCES in zmanref(zshmisc))\
ifnzman(noderef(Prompt Expansion))\
) is the number that is to be assigned to the em(next) command.

startmenu()
menu(Overview)
menu(Event Designators)
menu(Word Designators)
menu(Modifiers)
endmenu()
texinode(Overview)(Event Designators)()(History Expansion)
subsect(Overview)
vindex(histchars, use of)
A history expansion begins with the first character of the tt(histchars)
parameter, which is `tt(!)' by default, and may occur anywhere on the
command line; history expansions do not nest.  The `tt(!)' can be escaped
with `tt(\)' or can be enclosed between a pair of single quotes (tt(''))
to suppress its special meaning.  Double quotes will em(not) work for
this.  Following this history character is an optional event designator
(ifzman(see )noderef(Event Designators)) and then an optional word
designator (noderef(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.

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.
pindex(CSH_JUNKIE_HISTORY, use of)
However, if the option tt(CSH_JUNKIE_HISTORY) is set, then every history
reference with no event specification em(always) refers to the previous
command.

For example, `tt(!)' is the event designator for the previous command, so
`tt(!!:1)' always refers to the first word of the previous command, and
`tt(!!$)' always refers to the last word of the previous command.  With
tt(CSH_JUNKIE_HISTORY) set, then `tt(!:1)' and `tt(!$)' function in the
same manner as `tt(!!:1)' and `tt(!!$)', respectively.  Conversely, if
tt(CSH_JUNKIE_HISTORY) is unset, then `tt(!:1)' and `tt(!$)' 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 `tt(^)var(foo)tt(^)var(bar)' (where `tt(^)' is
actually the second character of the tt(histchars) parameter)
repeats the last command, replacing the string var(foo) with var(bar).
More precisely, the sequence `tt(^)var(foo)tt(^)var(bar)tt(^)' is
synonymous with `tt(!!:s)tt(^)var(foo)tt(^)var(bar)tt(^)', hence other
modifiers (see noderef(Modifiers)) may follow the final `tt(^)'.
In particular, `tt(^)var(foo)tt(^)var(bar)tt(:G)' performs a global
substitution.

If the shell encounters the character sequence `tt(!")'
in the input, the history mechanism is temporarily disabled until
the current list (see
ifzman(zmanref(zshmisc))\
ifnzman(noderef(Shell Grammar))\
) is fully parsed.  The `tt(!")' is removed from the input, and any
subsequent `tt(!)' characters have no special significance.

findex(fc, use of)
A less convenient but more comprehensible form of command history support
is provided by the tt(fc) builtin.
texinode(Event Designators)(Word Designators)(Overview)(History Expansion)
subsect(Event Designators)
cindex(history event designators)
cindex(event designators, history)
An event designator is a reference to a command-line entry in the history
list.  In the list below, remember that the initial tt(`!') in each item
may be changed to another character by setting the tt(histchars)
parameter.

startitem()
item(tt(!))(
Start a history expansion, except when followed by a blank, newline,
`tt(=)' or `tt(LPAR())'.  If followed immediately by a word designator
(ifzman(see )noderef(Word Designators)), this forms a history reference
with no event designator (ifzman(see )noderef(Overview)).
)
item(tt(!!))(
Refer to the previous command.
By itself, this expansion
repeats the previous command.
)
item(tt(!)var(n))(
Refer to command-line var(n).
)
item(tt(!-)var(n))(
Refer to the current command-line minus var(n).
)
item(tt(!)var(str))(
Refer to the most recent command starting with var(str).
)
item(tt(!?)var(str)[tt(?)])(
Refer to the most recent command containing var(str).  The trailing
`tt(?)' is necessary if this reference is to be followed by a modifier or
followed by any text that is not to be considered part of var(str).
)
item(tt(!#))(
Refer to the current command line typed in so far.  The line is
treated as if it were complete up to and including the word before the
one with the `tt(!#)' reference.
)
item(tt(!{)...tt(}))(
Insulate a history reference from adjacent characters (if necessary).
)
enditem()
texinode(Word Designators)(Modifiers)(Event Designators)(History Expansion)
subsect(Word Designators)
cindex(history word designators)
cindex(word designators, history)
A word designator indicates which word or words of a given command line are
to be included in a history reference.  A `tt(:)' usually
separates the event specification from the word designator.
It may be omitted only if the word designator begins with a
`tt(^)', `tt($)', `tt(*)', `tt(-)' or `tt(%)'.
Word designators include:

startsitem()
sitem(tt(0))(The first input word (command).)
sitem(var(n))(The var(n)th argument.)
sitem(tt(^))(The first argument.  That is, tt(1).)
sitem(tt($))(The last argument.)
sitem(tt(%))(The word matched by (the most recent) tt(?)var(str) search.)
sitem(var(x)tt(-)var(y))(A range of words; var(x) defaults to tt(0).)
sitem(tt(*))(All the arguments, or a null value if there are none.)
sitem(var(x)tt(*))(Abbreviates `var(x)tt(-$)'.)
sitem(var(x)tt(-))(Like `var(x)tt(*)' but omitting word tt($).)
endsitem()

Note that a `tt(%)' word designator works only when used in one of
`tt(!%)', `tt(!:%)' or `tt(!?)var(str)tt(?:%)', and only when used after a
tt(!?) expansion (possibly in an earlier command).  Anything else results
in an error, although the error may not be the most obvious one.
texinode(Modifiers)()(Word Designators)(History Expansion)
subsect(Modifiers)
cindex(modifiers)
cindex(colon modifiers)
cindex(history modifiers)
cindex(globbing modifiers)
cindex(parameter modifiers)
After the optional word designator, you can add
a sequence of one or more of the following modifiers,
each preceded by a `tt(:)'.  These modifiers also work on the result
of em(filename generation) and em(parameter expansion), except where
noted.

startitem()
item(tt(a))(
Turn a file name into an absolute path:  prepends the current directory,
if necessary, and resolves any use of `tt(..)' and `tt(.)' in the path.
)
item(tt(A))(
As `tt(a)', but also resolve use of symbolic links where possible.
Note that resolution of `tt(..)' occurs em(before) resolution of symbolic
links.  This call is equivalent to tt(a) unless your system has the
tt(realpath) system call (modern systems do).
)
item(tt(e))(
Remove all but the extension.
)
item(tt(h))(
Remove a trailing pathname component, leaving the head.  This works
like `tt(dirname)'.
)
item(tt(l))(
Convert the words to all lowercase.
)
item(tt(p))(
Print the new command but do not execute it.  Only works with history
expansion.
)
item(tt(q))(
Quote the substituted words, escaping further substitutions.  Works
with history expansion and parameter expansion, though for parameters
it is only useful if the resulting text is to be re-evaluated such as
by tt(eval).
)
item(tt(Q))(
Remove one level of quotes from the substituted words.
)
item(tt(r))(
Remove a filename extension of the form `tt(.)var(xxx)', leaving
the root name.
)
item(tt(s/)var(l)tt(/)var(r)[tt(/)])(
Substitute var(r) for var(l) as described below.
The substitution is done only for the
first string that matches var(l).  For arrays and for filename
generation, this applies to each word of the expanded text.  See
below for further notes on substitutions.

The forms `tt(gs/)var(l)tt(/)var(r)' and `tt(s/)var(l)tt(/)var(r)tt(/:G)'
perform global substitution, i.e. substitute every occurrence of var(r)
for var(l).  Note that the tt(g) or tt(:G) must appear in exactly the
position shown.
)
item(tt(&))(
Repeat the previous tt(s) substitution.  Like tt(s), may be preceded
immediately by a tt(g).  In parameter expansion the tt(&) must appear
inside braces, and in filename generation it must be quoted with a
backslash.
)
item(tt(t))(
Remove all leading pathname components, leaving the tail.  This works
like `tt(basename)'.
)
item(tt(u))(
Convert the words to all uppercase.
)
item(tt(x))(
Like tt(q), but break into words at whitespace.  Does not work with
parameter expansion.
)
enditem()

The tt(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 `tt(/)'.  A
backslash quotes the delimiter character.  The character `tt(&)', in
the right-hand-side var(r), is replaced by the text from the
left-hand-side var(l).  The `tt(&)' can be quoted with a backslash.  A
null var(l) uses the previous string either from the previous var(l)
or from the contextual scan string var(s) from `tt(!?)var(s)'.  You can
omit the rightmost delimiter if a newline immediately follows var(r);
the rightmost `tt(?)' in a context scan can similarly be omitted.
Note the same record of the last var(l) and var(r) is maintained
across all forms of expansion.

If the option tt(HIST_SUBST_PATTERN) is set, var(l) is treated as
a pattern of the usual form described in
ifzman(the section FILENAME GENERATION below)\
ifnzman(noderef(Filename Generation)).  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
tt(LPAR()#q:s/)var(...)tt(/)var(...)tt(/RPAR()) 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 tt(NO_BAD_PATTERN) option so will cause an error.

When tt(HIST_SUBST_PATTERN) is set, var(l) may start with a tt(#)
to indicate that the pattern must match at the start of the string
to be substituted, and a tt(%) may appear at the start or after an tt(#)
to indicate that the pattern must match at the end of the string
to be substituted.  The tt(%) or tt(#) may be quoted with two
backslashes.

For example, the following piece of filename generation code
with the tt(EXTENDED_GLOB) option:

example(print *.c+LPAR()#q:s/#%+LPAR()#b+RPAR()s+LPAR()*+RPAR().c/'S${match[1]}.C'/+RPAR())

takes the expansion of tt(*.c) and applies the glob qualifiers in the
tt(LPAR()#q)var(...)tt(RPAR()) expression, which consists of a substitution
modifier anchored to the start and end of each word (tt(#%)).  This
turns on backreferences (tt(LPAR()#b+RPAR())), so that the parenthesised
subexpression is available in the replacement string as tt(${match[1]}).
The replacement string is quoted so that the parameter is not substituted
before the start of filename generation.

The following tt(f), tt(F), tt(w) and tt(W) modifiers work only with
parameter expansion and filename generation.  They are listed here to
provide a single point of reference for all modifiers.

startitem()
item(tt(f))(
Repeats the immediately (without a colon) following modifier until the
resulting word doesn't change any more.
)
item(tt(F:)var(expr)tt(:))(
Like tt(f), but repeats only var(n) times if the expression
var(expr) evaluates to var(n).  Any character can be used instead of
the `tt(:)'; if `tt(LPAR())', `tt([)', or `tt({)'
is used as the opening delimiter,
the closing delimiter should be 'tt(RPAR())', `tt(])', or `tt(})',
respectively.
)
item(tt(w))(
Makes the immediately following modifier work on each word in the
string.
)
item(tt(W:)var(sep)tt(:))(
Like tt(w) but words are considered to be the parts of the string
that are separated by var(sep). Any character can be used instead of
the `tt(:)'; opening parentheses are handled specially, see above.
)
enditem()
texinode(Process Substitution)(Parameter Expansion)(History Expansion)(Expansion)
sect(Process Substitution)
cindex(process substitution)
cindex(substitution, process)
Each part of a command argument that takes the form
`tt(<LPAR())var(list)tt(RPAR())',
`tt(>LPAR())var(list)tt(RPAR())' or
`tt(=LPAR())var(list)tt(RPAR())'
is subject to process substitution.  The expression may be preceeded
or followed by other strings except that, to prevent clashes with
commonly occurring strings and patterns, the last
form must occur at the start of a command argument, and the forms
are only expanded when first parsing command or assignment arguments.
Process substitutions may be used following redirection operators; in this
case, the substitution must appear with no trailing string.

In the case of the tt(<) or tt(>) forms, the shell runs the commands in
var(list) asynchronously.  If the system supports the tt(/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
tt(>) is selected then writing on this special file will provide input for
var(list).  If tt(<) is used, then the file passed as an argument will
be connected to the output of the var(list) process.  For example,

example(tt(paste <LPAR()cut -f1) var(file1)tt(RPAR() <LPAR()cut -f3) var(file2)tt(RPAR() |
tee >LPAR())var(process1)tt(RPAR() >LPAR())var(process2)tt(RPAR() >/dev/null))

cuts fields 1 and 3 from the files var(file1) and var(file2) respectively,
pastes the results together, and sends it to the processes
var(process1) and var(process2).

If tt(=LPAR())var(...)tt(RPAR()) is used instead of
tt(<LPAR())var(...)tt(RPAR()),
then the file passed as an argument will be the name
of a temporary file containing the output of the var(list)
process.  This may be used instead of the tt(<)
form for a program that expects to lseek (see manref(lseek)(2))
on the input file.

There is an optimisation for substitutions of the form
tt(=LPAR()<<<)var(arg)tt(RPAR()), where var(arg) is a single-word argument
to the here-string redirection tt(<<<).  This form produces a file name
containing the value of var(arg) after any substitutions have been
performed.  This is handled entirely within the current shell.  This is
effectively the reverse of the special form tt($LPAR()<)var(arg)tt(RPAR())
which treats var(arg) as a file name and replaces it with the file's
contents.

The tt(=) form is useful as both the tt(/dev/fd) and the named pipe
implementation of tt(<LPAR())var(...)tt(RPAR()) 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 manref(lseek)(2)) on the file will not work.

Also note that the previous example can be more compactly and
efficiently written (provided the tt(MULTIOS) option is set) as:

example(tt(paste <LPAR()cut -f1) var(file1)tt(RPAR() <LPAR()cut -f3) var(file2)tt(RPAR()) ifzman(\ 
)tt(> >LPAR())var(process1)tt(RPAR() > >LPAR())var(process2)tt(RPAR()))

The shell uses pipes instead of FIFOs to implement the latter
two process substitutions in the above example.

There is an additional problem with tt(>LPAR())var(process)tt(RPAR()); when
this is attached to an external command, the parent shell does not wait
for var(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 em(MULTIOS) in
ifzman(zmanref(zshmisc))\
ifnzman(noderef(Redirection)).  Hence in a simplified
version of the example above:

example(tt(paste <LPAR()cut -f1) var(file1)tt(RPAR() <LPAR()cut -f3) var(file2)tt(RPAR()) tt(> >LPAR())var(process)tt(RPAR()))

(note that no tt(MULTIOS) are involved), var(process) will be run
asynchronously.  The workaround is:

example(tt({ paste <LPAR()cut -f1) var(file1)tt(RPAR() <LPAR()cut -f3) var(file2)tt(RPAR() }) tt(> >LPAR())var(process)tt(RPAR()))

The extra processes here are
spawned from the parent shell which will wait for their completion.

texinode(Parameter Expansion)(Command Substitution)(Process Substitution)(Expansion)
sect(Parameter Expansion)
cindex(parameter expansion)
cindex(expansion, parameter)
The character `tt($)' is used to introduce parameter expansions.
See
ifzman(\
zmanref(zshparam)
)\
ifnzman(\
noderef(Parameters)
)\
for a description of parameters, including arrays, associative arrays,
and subscript notation to access individual array elements.

Note in particular the fact that words of unquoted parameters are not
automatically split on whitespace unless the option tt(SH_WORD_SPLIT) is
set; see references to this option below for more details.  This is an
important difference from other shells.

In the expansions discussed below that require a pattern, the form of
the pattern is the same as that used for filename generation;
see noderef(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
noderef(Modifiers) in noderef(History Expansion) can be
applied:  for example, tt(${i:s/foo/bar/}) performs string
substitution on the expansion of parameter tt($i).

startitem()
item(tt(${)var(name)tt(}))(
The value, if any, of the parameter var(name) is substituted.
The braces are required if the expansion is to be followed by
a letter, digit, or underscore that is not to be interpreted
as part of var(name).  In addition, more complicated forms of substitution
usually require the braces to be present; exceptions, which only apply if
the option tt(KSH_ARRAYS) is not set, are a single subscript or any colon
modifiers appearing after the name, or any of the characters `tt(^)',
`tt(=)', `tt(~)', `tt(#)' or `tt(+)' appearing before the name, all of
which work with or without braces.

If var(name) is an array parameter, and the tt(KSH_ARRAYS) option is not
set, then the value of each
element of var(name) is substituted, one element per word.  Otherwise, the
expansion results in one word only; with tt(KSH_ARRAYS), this is the first
element of an array.  No field splitting is done on the result unless the
tt(SH_WORD_SPLIT) option is set.
See also the flags tt(=) and tt(s:)var(string)tt(:).
)
item(tt(${PLUS())var(name)tt(}))(
If var(name) is the name of a set parameter `tt(1)' is substituted,
otherwise `tt(0)' is substituted.
)
xitem(tt(${)var(name)tt(-)var(word)tt(}))
item(tt(${)var(name)tt(:-)var(word)tt(}))(
If var(name) is set, or in the second form is non-null, then substitute
its value; otherwise substitute var(word).  In the second form var(name)
may be omitted, in which case var(word) is always substituted.
)
xitem(tt(${)var(name)tt(PLUS())var(word)tt(}))
item(tt(${)var(name)tt(:PLUS())var(word)tt(}))(
If var(name) is set, or in the second form is non-null, then substitute
var(word); otherwise substitute nothing.
)
xitem(tt(${)var(name)tt(=)var(word)tt(}))
xitem(tt(${)var(name)tt(:=)var(word)tt(}))
item(tt(${)var(name)tt(::=)var(word)tt(}))(
In the first form, if var(name) is unset then set it to var(word); in the
second form, if var(name) is unset or null then set it to var(word); and
in the third form, unconditionally set var(name) to var(word).  In all
forms, the value of the parameter is then substituted.
)
xitem(tt(${)var(name)tt(?)var(word)tt(}))
item(tt(${)var(name)tt(:?)var(word)tt(}))(
In the first form, if var(name) is set, or in the second form if var(name)
is both set and non-null, then substitute its value; otherwise, print
var(word) and exit from the shell.  Interactive shells instead return to
the prompt.  If var(word) is omitted, then a standard message is printed.
)
enditem()

In any of the above expressions that test a variable and substitute an
alternate var(word), note that you can use standard shell quoting in the
var(word) value to selectively override the splitting done by the
tt(SH_WORD_SPLIT) option and the tt(=) flag, but not splitting by the
tt(s:)var(string)tt(:) flag.

In the following expressions, when var(name) is an array and
the substitution is not quoted, or if the `tt((@))' flag or the
var(name)tt([@]) syntax is used, matching and replacement is
performed on each array element separately.

startitem()
xitem(tt(${)var(name)tt(#)var(pattern)tt(}))
item(tt(${)var(name)tt(##)var(pattern)tt(}))(
If the var(pattern) matches the beginning of the value of
var(name), then substitute the value of var(name) with
the matched portion deleted; otherwise, just
substitute the value of var(name).  In the first
form, the smallest matching pattern is preferred;
in the second form, the largest matching pattern is
preferred.
)
xitem(tt(${)var(name)tt(%)var(pattern)tt(}))
item(tt(${)var(name)tt(%%)var(pattern)tt(}))(
If the var(pattern) matches the end of the value of
var(name), then substitute the value of var(name) with
the matched portion deleted; otherwise, just
substitute the value of var(name).  In the first
form, the smallest matching pattern is preferred;
in the second form, the largest matching pattern is
preferred.
)
item(tt(${)var(name)tt(:#)var(pattern)tt(}))(
If the var(pattern) matches the value of var(name), then substitute
the empty string; otherwise, just substitute the value of var(name).
If var(name) is an array
the matching array elements are removed (use the `tt((M))' flag to
remove the non-matched elements).
)
xitem(tt(${)var(name)tt(/)var(pattern)tt(/)var(repl)tt(}))
item(tt(${)var(name)tt(//)var(pattern)tt(/)var(repl)tt(}))(
Replace the longest possible match of var(pattern) in the expansion of
parameter var(name) by string var(repl).  The first form
replaces just the first occurrence, the second form all occurrences.
Both var(pattern) and var(repl) are subject to double-quoted substitution,
so that expressions like tt(${name/$opat/$npat}) will work, but note the
usual rule that pattern characters in tt($opat) are not treated specially
unless either the option tt(GLOB_SUBST) is set, or tt($opat) is instead
substituted as tt(${~opat}).

The var(pattern) may begin with a `tt(#)', in which case the
var(pattern) must match at the start of the string, or `tt(%)', in
which case it must match at the end of the string, or `tt(#%)' in which
case the var(pattern) must match the entire string.  The var(repl) may
be an empty string, in which case the final `tt(/)' may also be omitted.
To quote the final `tt(/)' in other cases it should be preceded by a
single backslash; this is not necessary if the
`tt(/)' occurs inside a substituted parameter.  Note also that the `tt(#)',
`tt(%)' and `tt(#%) are not active if they occur inside a substituted
parameter, even at the start.

The first `tt(/)' may be preceded by a `tt(:)', in which case the match
will only succeed if it matches the entire word.  Note also the
effect of the tt(I) and tt(S) parameter expansion flags below; however,
the flags tt(M), tt(R), tt(B), tt(E) and tt(N) are not useful.

For example,

example(foo="twinkle twinkle little star" sub="t*e" rep="spy"
print ${foo//${~sub}/$rep}
print ${(S)foo//${~sub}/$rep})

Here, the `tt(~)' ensures that the text of tt($sub) is treated as a
pattern rather than a plain string.  In the first case, the longest
match for tt(t*e) is substituted and the result is `tt(spy star)',
while in the second case, the shortest matches are taken and the
result is `tt(spy spy lispy star)'.
)
item(tt(${#)var(spec)tt(}))(
If var(spec) is one of the above substitutions, substitute
the length in characters of the result instead of
the result itself.  If var(spec) is an array expression,
substitute the number of elements of the result.
Note that `tt(^)', `tt(=)', and `tt(~)', below, must appear
to the left of `tt(#)' when these forms are combined.
)
item(tt(${^)var(spec)tt(}))(
pindex(RC_EXPAND_PARAM, toggle)
cindex(array expansion style, rc)
cindex(rc, array expansion style)
Turn on the tt(RC_EXPAND_PARAM) option for the
evaluation of var(spec); if the `tt(^)' is doubled, turn it off.
When this option is set, array expansions of the form
var(foo)tt(${)var(xx)tt(})var(bar),
where the parameter var(xx)
is set to tt(LPAR())var(a b c)tt(RPAR()), are substituted with
`var(fooabar foobbar foocbar)' instead of the default
`var(fooa b cbar)'.  Note that an empty array will therefore cause
all arguments to be removed.

Internally, each such expansion is converted into the
equivalent list for brace expansion.  E.g., tt(${^var}) becomes
tt({$var[1],$var[2],)...tt(}), and is processed as described in
noderef(Brace Expansion) below.
If word splitting is also in effect the
tt($var[)var(N)tt(]) may themselves be split into different list
elements.
)
item(tt(${=)var(spec)tt(}))(
pindex(SH_WORD_SPLIT, toggle)
cindex(field splitting, sh style, parameter)
cindex(sh, field splitting style, parameter)
Perform word splitting using the rules for tt(SH_WORD_SPLIT) during the
evaluation of var(spec), but regardless of whether the parameter appears in
double quotes; if the `tt(=)' is doubled, turn it off.
vindex(IFS, use of)
This forces parameter expansions to be split into
separate words before substitution, using tt(IFS) as a delimiter.
This is done by default in most other shells.

Note that splitting is applied to var(word) in the assignment forms
of var(spec) em(before) the assignment to var(name) is performed.
This affects the result of array assignments with the tt(A) flag.
)
item(tt(${~)var(spec)tt(}))(
pindex(GLOB_SUBST, toggle)
Turn on the tt(GLOB_SUBST) option for the evaluation of
var(spec); if the `tt(~)' is doubled, turn it off.  When this option is
set, the string resulting from the expansion will be interpreted as a
pattern anywhere that is possible, such as in filename expansion and
filename generation and pattern-matching contexts like the right
hand side of the `tt(=)' and `tt(!=)' operators in conditions.

In nested substitutions, note that the effect of the tt(~) 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 tt(foo) is set to tt(*), tt(${~foo//\*/*.c}) is substituted by
the pattern tt(*.c), which may be expanded by filename generation, but
tt(${${~foo}//\*/*.c}) substitutes to the string tt(*.c), which will not
be further expanded.
)
enditem()

If a tt(${)...tt(}) type parameter expression or a
tt($LPAR())...tt(RPAR()) type command substitution is used in place of
var(name) above, it is expanded first and the result is used as if
it were the value of var(name).  Thus it is
possible to perform nested operations:  tt(${${foo#head}%tail})
substitutes the value of tt($foo) with both `tt(head)' and `tt(tail)'
deleted.  The form with tt($LPAR())...tt(RPAR()) is often useful in
combination with the flags described next; see the examples below.
Each var(name) or nested tt(${)...tt(}) in a parameter expansion may
also be followed by a subscript expression as described in
ifzman(em(Array Parameters) in zmanref(zshparam))\
ifnzman(noderef(Array Parameters)).

Note that double quotes may appear around nested expressions, in which
case only the part inside is treated as quoted; for example,
tt(${(f)"$(foo)"}) quotes the result of tt($(foo)), but the flag `tt((f))'
(see below) is applied using the rules for unquoted expansions.  Note
further that quotes are themselves nested in this context; for example, in
tt("${(@f)"$(foo)"}"), there are two sets of quotes, one surrounding the
whole expression, the other (redundant) surrounding the tt($(foo)) as
before.

subsect(Parameter Expansion Flags)
cindex(parameter expansion flags)
cindex(flags, parameter expansion)
cindex(substitution, parameter, flags)
If the opening brace is directly followed by an opening parenthesis,
the string up to the matching closing parenthesis will be taken as a
list of flags.  In cases where repeating a flag is meaningful, the
repetitions need not be consecutive; for example, `(tt(q%q%q))'
means the same thing as the more readable `(tt(%%qqq))'.  The
following flags are supported:

startitem()
item(tt(#))(
Evaluate the resulting words as numeric expressions and output the
characters corresponding to the resulting integer.  Note that this form is
entirely distinct from use of the tt(#) without parentheses.

If the tt(MULTIBYTE) option is set and the number is greater than 127
(i.e. not an ASCII character) it is treated as a Unicode character.
)
item(tt(%))(
Expand all tt(%) escapes in the resulting words in the same way as in
prompts (see 
ifzman(EXPANSION OF PROMPT SEQUENCES in zmanref(zshmisc))\
ifnzman(noderef(Prompt Expansion))). If this flag is given twice,
full prompt expansion is done on the resulting words, depending on the
setting of the tt(PROMPT_PERCENT), tt(PROMPT_SUBST) and tt(PROMPT_BANG)
options.
)
item(tt(@))(
In double quotes, array elements are put into separate words.
E.g., `tt("${(@)foo}")' is equivalent to `tt("${foo[@]}")' and
`tt("${(@)foo[1,2]}")' is the same as `tt("$foo[1]" "$foo[2]")'.
This is distinct from em(field splitting) by the the tt(f), tt(s)
or tt(z) flags, which still applies within each array element.
)
item(tt(A))(
Create an array parameter with `tt(${)...tt(=)...tt(})',
`tt(${)...tt(:=)...tt(})' or `tt(${)...tt(::=)...tt(})'.
If this flag is repeated (as in `tt(AA)'), create an associative
array parameter.  Assignment is made before sorting or padding.
The var(name) part may be a subscripted range for ordinary
arrays; the var(word) part em(must) be converted to an array, for
example by using `tt(${(AA)=)var(name)tt(=)...tt(})' to activate
field splitting, when creating an associative array.
)
item(tt(a))(
Sort in array index order; when combined with `tt(O)' sort in reverse
array index order.  Note that `tt(a)' is therefore equivalent to the
default but `tt(Oa)' is useful for obtaining an array's elements in reverse
order.
)
item(tt(c))(
With tt(${#)var(name)tt(}), count the total number of characters in an array,
as if the elements were concatenated with spaces between them.
)
item(tt(C))(
Capitalize the resulting words.  `Words' in this case refers to sequences
of alphanumeric characters separated by non-alphanumerics, em(not) to words
that result from field splitting.
)
item(tt(e))(
Perform em(parameter expansion), em(command substitution) and
em(arithmetic expansion) on the result. Such expansions can be
nested but too deep recursion may have unpredictable effects.
)
item(tt(f))(
Split the result of the expansion to lines. This is a shorthand
for `tt(ps:\n:)'.
)
item(tt(F))(
Join the words of arrays together using newline as a separator.
This is a shorthand for `tt(pj:\n:)'.
)
item(tt(i))(
Sort case-insensitively.  May be combined with `tt(n)' or `tt(O)'.
)
item(tt(k))(
If var(name) refers to an associative array, substitute the em(keys)
(element names) rather than the values of the elements.  Used with
subscripts (including ordinary arrays), force indices or keys to be
substituted even if the subscript form refers to values.  However,
this flag may not be combined with subscript ranges.
)
item(tt(L))(
Convert all letters in the result to lower case.
)
item(tt(n))(
Sort decimal integers numerically; if the first differing
characters of two test strings are not digits, sorting
is lexical.   Integers with more initial zeroes
are sorted before those with fewer or none.  Hence the array `tt(foo1 foo02
foo2 foo3 foo20 foo23)' is sorted into the order shown.
May be combined with `tt(i)' or `tt(O)'.
)
item(tt(o))(
Sort the resulting words in ascending order; if this appears on its
own the sorting is lexical and case-sensitive (unless the locale
renders it case-insensitive).  Sorting in ascending order is the
default for other forms of sorting, so this is ignored if combined
with `tt(a)', `tt(i)' or `tt(n)'.
)
item(tt(O))(
Sort the resulting words in descending order; `tt(O)' without `tt(a)',
`tt(i)' or `tt(n)' sorts in reverse lexical order.  May be combined
with `tt(a)', `tt(i)' or `tt(n)' to reverse the order of sorting.
)
item(tt(P))(
This forces the value of the parameter var(name) to be interpreted as a
further parameter name, whose value will be used where appropriate.
Note that flags set with one of the tt(typeset) family of commands
(in particular case transformations) are not applied to the value of
var(name) used in this fashion.

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 `tt(foo=bar)' and `tt(bar=baz)', the strings tt(${(P)foo}),
tt(${(P)${foo}}), and tt(${(P)$(echo bar)}) will be expanded to `tt(baz)'.
)
item(tt(q))(
Quote the resulting words with backslashes; unprintable or invalid
characters are quoted using the tt($'\)var(NNN)tt(') form, with separate
quotes for each octet.  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 tt($).
)
item(tt(Q))(
Remove one level of quotes from the resulting words.
)
item(tt(t))(
Use a string describing the type of the parameter where the value
of the parameter would usually appear. This string consists of keywords
separated by hyphens (`tt(-)'). The first keyword in the string describes
the main type, it can be one of `tt(scalar)', `tt(array)', `tt(integer)',
`tt(float)' or `tt(association)'. The other keywords describe the type in
more detail:

startitem()
item(tt(local))(
for local parameters
)
item(tt(left))(
for left justified parameters
)
item(tt(right_blanks))(
for right justified parameters with leading blanks
)
item(tt(right_zeros))(
for right justified parameters with leading zeros
)
item(tt(lower))(
for parameters whose value is converted to all lower case when it is
expanded
)
item(tt(upper))(
for parameters whose value is converted to all upper case when it is
expanded
)
item(tt(readonly))(
for readonly parameters
)
item(tt(tag))(
for tagged parameters
)
item(tt(export))(
for exported parameters
)
item(tt(unique))(
for arrays which keep only the first occurrence of duplicated values
)
item(tt(hide))(
for parameters with the `hide' flag
)
item(tt(special))(
for special parameters defined by the shell
)
enditem()
)
item(tt(u))(
Expand only the first occurrence of each unique word.
)
item(tt(U))(
Convert all letters in the result to upper case.
)
item(tt(v))(
Used with tt(k), substitute (as two consecutive words) both the key
and the value of each associative array element.  Used with subscripts,
force values to be substituted even if the subscript form refers to
indices or keys.
)
item(tt(V))(
Make any special characters in the resulting words visible.
)
item(tt(w))(
With tt(${#)var(name)tt(}), count words in arrays or strings; the tt(s)
flag may be used to set a word delimiter.
)
item(tt(W))(
Similar to tt(w) with the difference that empty words between
repeated delimiters are also counted.
)
item(tt(X))(
With this flag, parsing errors occurring with the tt(Q), tt(e) and tt(#)
flags or the pattern matching forms such as
`tt(${)var(name)tt(#)var(pattern)tt(})' are reported.  Without the flag,
errors are silently ignored.
)
item(tt(z))(
Split the result of the expansion into words using shell parsing to
find the words, i.e. taking into account any quoting in the value.

Note that this is done very late, as for the `tt((s))' flag. So to
access single words in the result, one has to use nested expansions as 
in `tt(${${(z)foo}[2]})'. Likewise, to remove the quotes in the
resulting words one would do: `tt(${(Q)${(z)foo}})'.
)
item(tt(0))(
Split the result of the expansion on null bytes.  This is a shorthand
for `tt(ps:\0:)'.
)
enditem()

The following flags (except tt(p)) are followed by one or more arguments
as shown.  Any character, or the matching pairs `tt(LPAR())...tt(RPAR())',
`tt({)...tt(})', `tt([)...tt(])', or `tt(<)...tt(>)', 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.

startitem()
item(tt(p))(
Recognize the same escape sequences as the tt(print) builtin
in string arguments to any of the flags described below that
follow this argument.
)
item(tt(~))(
Force string arguments to any of the flags below that follow within
the parentheses to be treated as patterns.  Compare with a tt(~)
outside parentheses, which forces the entire substituted string to
be treated as a pattern.  Hence, for example,
example([[ "?" = ${(~j.|.)array} ]])
with the tt(EXTENDED_GLOB) option set succeeds if and only if tt($array)
contains the string `tt(?)' as an element.  The argument may be
repeated to toggle the behaviour; its effect only lasts to the
end of the parenthesised group.
)
item(tt(j:)var(string)tt(:))(
Join the words of arrays together using var(string) as a separator.
pindex(SH_WORD_SPLIT, use of)
Note that this occurs before field splitting by the tt(s:)var(string)tt(:)
flag or the tt(SH_WORD_SPLIT) option.
)
item(tt(l:)var(expr)tt(::)var(string1)tt(::)var(string2)tt(:))(
Pad the resulting words on the left.  Each word will be truncated if
required and placed in a field var(expr) characters wide.

The arguments tt(:)var(string1)tt(:) and tt(:)var(string2)tt(:) 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 var(string1) (concatenated as
often as needed) or spaces if var(string1) is not given.  If both
var(string1) and var(string2) are given, tt(string2) is inserted once
directly to the left of each word, truncated if necessary, before
var(string1) is used to produce any remaining padding.

If the tt(MULTIBYTE) option is in effect, the flag tt(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 tt(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.
)
item(tt(m))(
Only useful together with one of the flags tt(l) or tt(r) or with the
tt(#) length operator when the tt(MULTIBYTE) option
is in effect.  Use the character width reported by the system in
calculating the how much of the string it occupies or the overall
length of the string.  Most printable characters have a width of one
unit, however certain Asian character sets and certain special effects
use wider characters; combining characters have zero width.
)
item(tt(r:)var(expr)tt(::)var(string1)tt(::)var(string2)tt(:))(
As tt(l), but pad the words on the right and insert var(string2)
immediately to the right of the string to be padded.

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.
)
item(tt(s:)var(string)tt(:))(
Force field splitting at the
separator var(string).  Note that a var(string) of two or more
characters means that all of them must match in sequence; this differs from
the treatment of two or more characters in the tt(IFS) parameter.
See also the tt(=) flag and the tt(SH_WORD_SPLIT) option.

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:

example(line="one::three"
print -l "${(s.:.)line}")

produces two lines of output for tt(one) and tt(three) and elides the
empty field.  To override this behaviour, supply the "(@)" flag as well,
i.e.  tt("${(@s.:.)line}").
)
enditem()

The following flags are meaningful with the tt(${)...tt(#)...tt(}) or
tt(${)...tt(%)...tt(}) forms.  The tt(S) and tt(I) flags may also be
used with the tt(${)...tt(/)...tt(}) forms.

startitem()
item(tt(S))(
Search substrings as well as beginnings or ends; with tt(#) start
from the beginning and with tt(%) start from the end of the string.
With substitution via tt(${)...tt(/)...tt(}) or
tt(${)...tt(//)...tt(}), specifies non-greedy matching, i.e. that the
shortest instead of the longest match should be replaced.
)
item(tt(I:)var(expr)tt(:))(
Search the var(expr)th match (where var(expr) evaluates to a number).
This only applies when searching for substrings, either with the tt(S)
flag, or with tt(${)...tt(/)...tt(}) (only the var(expr)th match is
substituted) or tt(${)...tt(//)...tt(}) (all matches from the
var(expr)th on are substituted).  The default is to take the first match.

The var(expr)th 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 tt(${)...tt(%)...tt(}) and tt(${)...tt(%%)...tt(}) 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
example(which switch is the right switch for Ipswich?)
substitutions of the form
tt(${)LPAR()tt(SI:)var(N)tt(:)RPAR()tt(string#w*ch}) as var(N) increases
from 1 will match and remove `tt(which)', `tt(witch)', `tt(witch)' and
`tt(wich)'; the form using `tt(##)' will match and remove `tt(which switch
is the right switch for Ipswich)', `tt(witch is the right switch for
Ipswich)', `tt(witch for Ipswich)' and `tt(wich)'. The form using `tt(%)'
will remove the same matches as for `tt(#)', but in reverse order, and the
form using `tt(%%)' will remove the same matches as for `tt(##)' in reverse
order.
)
item(tt(B))(
Include the index of the beginning of the match in the result.
)
item(tt(E))(
Include the index of the end of the match in the result.
)
item(tt(M))(
Include the matched portion in the result.
)
item(tt(N))(
Include the length of the match in the result.
)
item(tt(R))(
Include the unmatched portion in the result (the em(R)est).
)
enditem()

subsect(Rules)

Here is a summary of the rules for substitution; this assumes that braces
are present around the substitution, i.e. tt(${...}).  Some particular
examples are given below.  Note that the Zsh Development Group accepts
em(no responsibility) for any brain damage which may occur during the
reading of the following rules.

startitem()
item(tt(1.) em(Nested Substitution))(
If multiple nested tt(${...}) forms are present, substitution is
performed from the inside outwards.  At each level, the substitution takes
account of whether the current value is a scalar or an array, whether the
whole substitution is in double quotes, and what flags are supplied to the
current level of substitution, just as if the nested substitution were the
outermost.  The flags are not propagated up to enclosing
substitutions; the nested substitution will return either a scalar or an
array as determined by the flags, possibly adjusted for quoting.  All the
following steps take place where applicable at all levels of substitution.
Note that, unless the `tt((P))' flag is present, the flags and any subscripts
apply directly to the value of the nested substitution; for example, the
expansion tt(${${foo}}) behaves exactly the same as tt(${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 tt(~) and tt(=)).  Thus, for example, tt(${${:-=cat}:h})
expands to the directory where the tt(cat) program resides.  (Explanation:
the internal substitution has no parameter but a default value tt(=cat),
which is expanded by filename expansion to a full path; the outer
substitution then applies the modifier tt(:h) and takes the directory part
of the path.)
)
item(tt(2.) em(Internal Parameter Flags))(
Any parameter flags set by one of the tt(typeset) family of commands,
in particular the tt(L), tt(R), tt(Z), tt(u) and tt(l) flags for padding
and capitalization, are applied directly to the parameter value.
)
item(tt(3.) em(Parameter Subscripting))(
If the value is a raw parameter reference with a subscript, such as
tt(${)var(var)tt([3]}), the effect of subscripting is applied directly to
the parameter.  Subscripts are evaluated left to right; subsequent
subscripts apply to the scalar or array value yielded by the previous
subscript.  Thus if tt(var) is an array, tt(${var[1][2]}) is the second
character of the first word, but tt(${var[2,4][2]}) is the entire third
word (the second word of the range of words two through four of the
original array).  Any number of subscripts may appear.
)
item(tt(4.) em(Parameter Name Replacement))(
The effect of any tt((P)) flag, which treats the value so far as a
parameter name and replaces it with the corresponding value, is applied.
)
item(tt(5.) em(Double-Quoted Joining))(
If the value after this process is an array, and the substitution
appears in double quotes, and no tt((@)) flag is present at the current
level, the words of the value are joined with the first character of the
parameter tt($IFS), by default a space, between each word (single word
arrays are not modified).  If the tt((j)) flag is present, that is used for
joining instead of tt($IFS).
)
item(tt(6.) em(Nested Subscripting))(
Any remaining subscripts (i.e. of a nested substitution) are evaluated at
this point, based on whether the value is an array or a scalar.  As with
tt(2.), multiple subscripts can appear.  Note that tt(${foo[2,4][2]}) is
thus equivalent to tt(${${foo[2,4]}[2]}) and also to
tt("${${(@)foo[2,4]}[2]}") (the nested substitution returns an array in
both cases), but not to tt("${${foo[2,4]}[2]}") (the nested substitution
returns a scalar because of the quotes).
)
item(tt(7.) em(Modifiers))(
Any modifiers, as specified by a trailing `tt(#)', `tt(%)', `tt(/)'
(possibly doubled) or by a set of modifiers of the form tt(:...) (see
noderef(Modifiers) in noderef(History Expansion)), are applied to the words
of the value at this level.
)
item(tt(8.) em(Forced Joining))(
If the `tt((j))' flag is present, or no `tt((j))' flag is present but
the string is to be split as given by rules tt(8.) or tt(9.), and joining
did not take place at step tt(4.), any words in the value are joined
together using the given string or the first character of tt($IFS) if none.
Note that the `tt((F))' flag implicitly supplies a string for joining in this
manner.
)
item(tt(9.) em(Forced Splitting))(
If one of the `tt((s))', `tt((f))' or `tt((z))' flags are present, or the `tt(=)'
specifier was present (e.g. tt(${=)var(var)tt(})), the word is split on
occurrences of the specified string, or (for tt(=) with neither of the two
flags present) any of the characters in tt($IFS).
)
item(tt(10.) em(Shell Word Splitting))(
If no `tt((s))', `tt((f))' or `tt(=)' was given, but the word is not
quoted and the option tt(SH_WORD_SPLIT) is set, the word is split on
occurrences of any of the characters in tt($IFS).  Note this step, too,
takes place at all levels of a nested substitution.
)
item(tt(11.) em(Uniqueness))(
If the result is an array and the `tt((u))' flag was present, duplicate
elements are removed from the array.
)
item(tt(12.) em(Ordering))(
If the result is still an array and one of the `tt((o))' or `tt((O))' flags
was present, the array is reordered.
)
item(tt(13.) em(Re-Evaluation))(
Any `tt((e))' flag is applied to the value, forcing it to be re-examined
for new parameter substitutions, but also for command and arithmetic
substitutions.
)
item(tt(14.) em(Padding))(
Any padding of the value by the `tt(LPAR()l.)var(fill)tt(.RPAR())' or
`tt(LPAR()r.)var(fill)tt(.RPAR())' flags is applied.
)
item(tt(15.) em(Semantic Joining))(
In contexts where expansion semantics requires a single word to
result, all words are rejoined with the first character of tt(IFS)
between.  So in `tt(${LPAR()P)tt(RPAR()${LPAR()f)tt(RPAR()lines}})'
the value of tt(${lines}) is split at newlines, but then must be
joined again before the tt(P) flag can be applied.

If a single word is not required, this rule is skipped.
)
enditem()

subsect(Examples)
The flag tt(f) is useful to split a double-quoted substitution line by
line.  For example, tt(${(f)"$LPAR()<)var(file)tt(RPAR()"})
substitutes the contents of var(file) divided so that each line is
an element of the resulting array.  Compare this with the effect of
tt($)tt(LPAR()<)var(file)tt(RPAR()) alone, which divides the file
up by words, or the same inside double quotes, which makes the entire
content of the file a single string.

The following illustrates the rules for nested parameter expansions.
Suppose that tt($foo) contains the array tt(LPAR()bar baz)tt(RPAR()):

startitem()
item(tt("${(@)${foo}[1]}"))(
This produces the result tt(b).  First, the inner substitution
tt("${foo}"), which has no array (tt(@)) flag, produces a single word
result tt("bar baz").  The outer substitution tt("${(@)...[1]}") detects
that this is a scalar, so that (despite the `tt((@))' flag) the subscript
picks the first character. 
)
item(tt("${${(@)foo}[1]}"))(
This produces the result `tt(bar)'.  In this case, the inner substitution
tt("${(@)foo}") produces the array `tt(LPAR()bar baz)tt(RPAR())'.  The outer
substitution tt("${...[1]}") detects that this is an array and picks the
first word.  This is similar to the simple case tt("${foo[1]}").
)
enditem()

As an example of the rules for word splitting and joining, suppose tt($foo)
contains the array `tt(LPAR()ax1 bx1)tt(RPAR())'.  Then

startitem()
item(tt(${(s/x/)foo}))(
produces the words `tt(a)', `tt(1 b)' and `tt(1)'.
)
item(tt(${(j/x/s/x/)foo}))(
produces `tt(a)', `tt(1)', `tt(b)' and `tt(1)'.
)
item(tt(${(s/x/)foo%%1*}))(
produces `tt(a)' and `tt( b)' (note the extra space).  As substitution
occurs before either joining or splitting, the operation  first generates
the modified array tt(LPAR()ax bx)tt(RPAR()), which is joined to give
tt("ax bx"), and then split to give `tt(a)', `tt( b)' and `'.  The final
empty string will then be elided, as it is not in double quotes.
)
enditem()

texinode(Command Substitution)(Arithmetic Expansion)(Parameter Expansion)(Expansion)
sect(Command Substitution)
cindex(command substitution)
cindex(substitution, command)
A command enclosed in parentheses preceded by a dollar sign, like
`tt($LPAR())...tt(RPAR())', or quoted with grave 
accents, like `tt(`)...tt(`)', is replaced with its standard output, with
any trailing newlines deleted.
If the substitution is not enclosed in double quotes, the
output is broken into words using the tt(IFS) parameter.
vindex(IFS, use of)
The substitution `tt($LPAR()cat) var(foo)tt(RPAR())' may be replaced
by the equivalent but faster `tt($LPAR()<)var(foo)tt(RPAR())'.
In either case, if the option tt(GLOB_SUBST) is set,
the output is eligible for filename generation.
texinode(Arithmetic Expansion)(Brace Expansion)(Command Substitution)(Expansion)
sect(Arithmetic Expansion)
cindex(arithmetic expansion)
cindex(expansion, arithmetic)
A string of the form `tt($[)var(exp)tt(])' or
`tt($LPAR()LPAR())var(exp)tt(RPAR()RPAR())' is substituted
with the value of the arithmetic expression var(exp).  var(exp) is
subjected to em(parameter expansion), em(command substitution)
and em(arithmetic expansion) before it is evaluated.
See noderef(Arithmetic Evaluation).
texinode(Brace Expansion)(Filename Expansion)(Arithmetic Expansion)(Expansion)
sect(Brace Expansion)
cindex(brace expansion)
cindex(expansion, brace)
A string of the form
`var(foo)tt({)var(xx)tt(,)var(yy)tt(,)var(zz)tt(})var(bar)'
is expanded to the individual words
`var(fooxxbar)', `var(fooyybar)' and `var(foozzbar)'.
Left-to-right order is preserved.  This construct
may be nested.  Commas may be quoted in order to
include them literally in a word.

An expression of the form `tt({)var(n1)tt(..)var(n2)tt(})',
where var(n1) and var(n2) are integers,
is expanded to every number between
var(n1) and var(n2) inclusive.  If either number begins with a
zero, all the resulting numbers will be padded with leading zeroes to
that minimum width.  If the numbers are in decreasing order the
resulting sequence will also be in decreasing order.

If a brace expression matches none of the above forms, it is left
unchanged, unless the option tt(BRACE_CCL) (an abbreviation for `brace
character class') is set.
pindex(BRACE_CCL, use of)
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
tt([)...tt(]) expression in filename generation:
`tt(-)' is treated specially to denote a range of characters, but `tt(^)' or
`tt(!)' as the first character is treated normally.  For example,
`tt({abcdef0-9})' expands to 16 words tt(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 tt(*/{foo,bar}) is split into two separate words
tt(*/foo) and tt(*/bar) before filename generation takes place.  In
particular, note that this is liable to produce a `no match' error if
em(either) of the two expressions does not match; this is to be contrasted
with tt(*/(foo|bar)), which is treated as a single pattern but otherwise
has similar effects.

To combine brace expansion with array expansion, see the
tt(${^)var(spec)tt(}) form described
ifzman(in the section Parameter Expansion)\
ifnzman(in noderef(Parameter Expansion))
above.

texinode(Filename Expansion)(Filename Generation)(Brace Expansion)(Expansion)
sect(Filename Expansion)
cindex(filename expansion)
cindex(expansion, filename)
Each word is checked to see if it begins with an unquoted `tt(~)'.
If it does, then the word up to a `tt(/)',
or the end of the word if there is no `tt(/)',
is checked to see if it can be substituted in one of the ways
described here.  If so, then the `tt(~)' and the checked portion are
replaced with the appropriate substitute value.

A `tt(~)' by itself is replaced by the value of tt($HOME).
A `tt(~)' followed by a `tt(PLUS())' or a `tt(-)' is replaced by the value of
tt($PWD) or tt($OLDPWD), respectively.

A `tt(~)' followed by a number is replaced by the directory at that
position in the directory stack.
`tt(~0)' is equivalent to `tt(~PLUS())',
and `tt(~1)' is the top of the stack.
`tt(~PLUS())' followed by a number is replaced by the directory at that
position in the directory stack.
`tt(~PLUS()0)' is equivalent to `tt(~PLUS())',
and `tt(~PLUS()1)' is the top of the stack.
`tt(~-)' followed by a number is replaced by the directory that
many positions from the bottom of the stack.
`tt(~-0)' is the bottom of the stack.
pindex(PUSHD_MINUS, use of)
The tt(PUSHD_MINUS)
option exchanges the effects of `tt(~PLUS())' and `tt(~-)' where they are
followed by a number.

subsect(Dynamic named directories)
cindex(directories, named, dynamic)
cindex(named directories, dynamic)
cindex(dynamic named directories)

The feature described here is only available if the shell function
tt(zsh_directory_name) exists.

A `tt(~)' followed by a string var(namstr) in unquoted square brackets is
treated specially as a dynamic directory name.  Note that the first
unquoted closing square bracket always terminates var(namstr).  The shell
function is passed two arguments: the string tt(n) (for name) and
var(namstr).  It should either set the array tt(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 and tt(NOMATCH) handling is applied if the option is set.

The function tt(zsh_directory_name) is also used to see if a directory can
be turned into a name, for example when printing the directory stack or
when expanding tt(%~) in prompts.  In this case the function is passed two
arguments: the string tt(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 `tt(~[)var(...)tt(])'), and the second is the
prefix length of the directory to be replaced.  For example, if the trial
directory is tt(/home/myname/src/zsh) and the dynamic name for
tt(/home/myname/src) (which has 16 characters) is tt(s), then the function
sets

example(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.

As a working example, here is a function that expands any dynamic names
beginning with the string tt(p:) to directories below
tt(/home/pws/perforce).  In this simple case a static name for the
directory would be just as effective.

example(zsh_directory_name+LPAR()RPAR() {
  emulate -L zsh
  setopt extendedglob
  local -a match mbegin mend
  if [[ $1 = d ]]; then
    if [[ $2 = (#b)(/home/pws/perforce/)([^/]##)* ]]; then
      typeset -ga reply
      reply=(p:$match[2] $(( ${#match[1]} + ${#match[2]} )) )
    else
      return 1
    fi
  else
    [[ $2 != (#b)p:(?*) ]] && return 1
    typeset -ga reply
    reply=(/home/pws/perforce/$match[1])
  fi
  return 0
})

subsect(Static named directories)
cindex(directories, named, static)
cindex(named directories, static)
cindex(static named directories)
A `tt(~)' followed by anything not already covered consisting
of any number of alphanumeric characters or underscore (`tt(_)'),
hyphen (`tt(-)'), or dot (`tt(.)') is looked up as a
named directory, and replaced by the value of that named directory if found.
Named directories are typically home directories for users on the system.
They may also be defined if the text after the `tt(~)' is the name
of a string shell parameter whose value begins with a `tt(/)'.
Note that trailing slashes will be removed from the path to the directory
(though the original parameter is not modified).

It is also possible to define directory names using the tt(-d) option to the
tt(hash) builtin.

In certain circumstances (in prompts, for instance), when the shell
prints a path, 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 `tt(~)' followed by the name of the directory.
The shortest way of referring to the directory is used,
with ties broken in favour of using a named directory,
except when the directory is tt(/) itself.  The parameters tt($PWD) and
tt($OLDPWD) are never abbreviated in this fashion.

subsect(`=' expansion)

If a word begins with an unquoted `tt(=)'
and the tt(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.

subsect(Notes)

Filename expansion is performed on the right hand side of a parameter
assignment, including those appearing after commands of the
tt(typeset) family.  In this case, the right hand side will be treated
as a colon-separated list in the manner of the tt(PATH) parameter,
so that a `tt(~)' or an `tt(=)' following a `tt(:)' is eligible for expansion.
All such behaviour can be
disabled by quoting the `tt(~)', the `tt(=)', or the whole expression (but not
simply the colon); the tt(EQUALS) option is also respected.

If the option tt(MAGIC_EQUAL_SUBST) is set, any unquoted shell
argument in the form `var(identifier)tt(=)var(expression)' becomes eligible
for file expansion as described in the previous paragraph.  Quoting the
first `tt(=)' also inhibits this.

texinode(Filename Generation)()(Filename Expansion)(Expansion)
sect(Filename Generation)
cindex(filename generation)
If a word contains an unquoted instance of one of the characters
`tt(*)', `tt(LPAR())', `tt(|)', `tt(<)', `tt([)', or `tt(?)', it is regarded
as a pattern for filename generation, unless the tt(GLOB) option is unset.
pindex(GLOB, use of)
If the tt(EXTENDED_GLOB) option is set,
pindex(EXTENDED_GLOB, use of)
the `tt(^)' and `tt(#)' characters also denote a pattern; otherwise
they are not treated specially by the shell.

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 tt(NULL_GLOB) option is set,
pindex(NULL_GLOB, use of)
in which case the word is deleted; or unless the tt(NOMATCH)
option is unset, in which case the word is left unchanged.
pindex(NOMATCH, use of)

In filename generation,
the character `tt(/)' must be matched explicitly;
also, a `tt(.)' must be matched
explicitly at the beginning of a pattern or after a `tt(/)', unless the
tt(GLOB_DOTS) option is set.
pindex(GLOB_DOTS, use of)
No filename generation pattern
matches the files `tt(.)' or `tt(..)'.  In other instances of pattern
matching, the `tt(/)' and `tt(.)' are not treated specially.
subsect(Glob Operators)
startitem()
item(tt(*))(
Matches any string, including the null string.
)
item(tt(?))(
Matches any character.
)
item(tt([)...tt(]))(
Matches any of the enclosed characters.  Ranges of characters
can be specified by separating two characters by a `tt(-)'.
A `tt(-)' or `tt(])' may be matched by including it as the
first character in the list.
cindex(character classes)
There are also several named classes of characters, in the form
`tt([:)var(name)tt(:])' with the following meanings.
The first set use the macros provided by
the operating system to test for the given character combinations,
including any modifications due to local language settings, see
manref(ctype)(3):

startitem()
item(tt([:alnum:]))(
The character is alphanumeric
)
item(tt([:alpha:]))
(
The character is alphabetic
)
item(tt([:ascii:]))(
The character is 7-bit, i.e. is a single-byte character without
the top bit set.
)
item(tt([:blank:]))(
The character is either space or tab
)
item(tt([:cntrl:]))(
The character is a control character
)
item(tt([:digit:]))(
The character is a decimal digit
)
item(tt([:graph:]))(
The character is a printable character other than whitespace
)
item(tt([:lower:]))(
The character is a lowercase letter
)
item(tt([:print:]))(
The character is printable
)
item(tt([:punct:]))(
The character is printable but neither alphanumeric nor whitespace
)
item(tt([:space:]))(
The character is whitespace
)
item(tt([:upper:]))(
The character is an uppercase letter
)
item(tt([:xdigit:]))(
The character is a hexadecimal digit
)
enditem()

Another set of named classes is handled internally by the shell and
is not sensitive to the locale:

startitem()
item(tt([:IDENT:]))(
The character is allowed to form part of a shell identifier, such
as a parameter name
)
item(tt([:IFS:]))(
The character is used as an input field separator, i.e. is contained in the
tt(IFS) parameter
)
item(tt([:IFSSPACE:]))(
The character is an IFS white space character; see the documentation
for tt(IFS) in
ifzman(the zmanref(zshparam) manual page)\
ifnzman(noderef(Parameters Used By The Shell))\
.
)
item(tt([:WORD:]))(
The character is treated as part of a word; this test is sensitive
to the value of the tt(WORDCHARS) parameter
)
enditem()

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 `tt([[:alnum:]])'.  Named
character sets can be used alongside other types,
e.g. `tt([[:alpha:]0-9])'.
)
xitem(tt([^)...tt(]))
item(tt([!)...tt(]))(
Like tt([)...tt(]), except that it matches any character which is
not in the given set.
)
item(tt(<)[var(x)]tt(-)[var(y)]tt(>))(
Matches any number in the range var(x) to var(y), inclusive.
Either of the numbers may be omitted to make the range open-ended;
hence `tt(<->)' matches any number.  To match individual digits, the
tt([)...tt(]) form is more efficient.

Be careful when using other wildcards adjacent to patterns of this form;
for example, tt(<0-9>*) will actually match any number whatsoever at the
start of the string, since the `tt(<0-9>)' will match the first digit, and
the `tt(*)' 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 `tt(<0-9>[^[:digit:]]*)' can be
used instead.
)
item(tt(LPAR())...tt(RPAR()))(
Matches the enclosed pattern.  This is used for grouping.
If the tt(KSH_GLOB) option is set, then a
`tt(@)', `tt(*)', `tt(+)', `tt(?)' or `tt(!)' immediately preceding
the `tt(LPAR())' is treated specially, as detailed below. The option
tt(SH_GLOB) prevents bare parentheses from being used in this way, though
the tt(KSH_GLOB) option is still available.

Note that grouping cannot extend over multiple directories: it is an error
to have a `tt(/)' within a group (this only applies for patterns used in
filename generation).  There is one exception:  a group of the form
tt(LPAR())var(pat)tt(/RPAR()#) appearing as a complete path segment can
match a sequence of directories.  For example, tt(foo/(a*/)#bar) matches
tt(foo/bar), tt(foo/any/bar), tt(foo/any/anyother/bar), and so on.
)
item(var(x)tt(|)var(y))(
Matches either var(x) or var(y).
This operator has lower precedence than any other.
The `tt(|)' character
must be within parentheses, to avoid interpretation as a pipeline.
)
item(tt(^)var(x))(
(Requires tt(EXTENDED_GLOB) to be set.)
Matches anything except the pattern var(x).
This has a higher precedence than `tt(/)', so `tt(^foo/bar)'
will search directories in `tt(.)' except `tt(./foo)'
for a file named `tt(bar)'.
)
item(var(x)tt(~)var(y))(
(Requires tt(EXTENDED_GLOB) to be set.)
Match anything that matches the pattern var(x) but does not match var(y).
This has lower precedence than any operator except `tt(|)', so
`tt(*/*~foo/bar)' will search for all files in all directories in `tt(.)'
and then exclude `tt(foo/bar)' if there was such a match.
Multiple patterns can be excluded by
`var(foo)tt(~)var(bar)tt(~)var(baz)'.
In the exclusion pattern (var(y)), `tt(/)' and `tt(.)' are not treated
specially the way they usually are in globbing.
)
item(var(x)tt(#))(
(Requires tt(EXTENDED_GLOB) to be set.)
Matches zero or more occurrences of the pattern var(x).
This operator has high precedence; `tt(12#)' is equivalent to `tt(1(2#))',
rather than `tt((12)#)'.  It is an error for an unquoted `tt(#)' to follow
something which cannot be repeated; this includes an empty string, a
pattern already followed by `tt(##)', or parentheses when part of a
tt(KSH_GLOB) pattern (for example, `tt(!LPAR())var(foo)tt(RPAR()#)' is
invalid and must be replaced by
`tt(*LPAR()!LPAR())var(foo)tt(RPAR()RPAR())').
)
item(var(x)tt(##))(
(Requires tt(EXTENDED_GLOB) to be set.)
Matches one or more occurrences of the pattern var(x).
This operator has high precedence; `tt(12##)' is equivalent to `tt(1(2##))',
rather than `tt((12)##)'.  No more than two active `tt(#)' characters may
appear together.  (Note the potential clash with glob qualifiers in the
form `tt(1(2##))' which should therefore be avoided.)
)
enditem()
subsect(ksh-like Glob Operators)
pindex(KSH_GLOB, use of)
If the tt(KSH_GLOB) option is set, the effects of parentheses can be
modified by a preceding `tt(@)', `tt(*)', `tt(+)', `tt(?)' or `tt(!)'.
This character need not be unquoted to have special effects,
but the `tt(LPAR())' must be.

startitem()
item(tt(@LPAR())...tt(RPAR()))(
Match the pattern in the parentheses.  (Like `tt(LPAR())...tt(RPAR())'.)
)
item(tt(*LPAR())...tt(RPAR()))(
Match any number of occurrences.  (Like `tt(LPAR())...tt(RPAR()#)'.)
)
item(tt(PLUS()LPAR())...tt(RPAR()))(
Match at least one occurrence.  (Like `tt(LPAR())...tt(RPAR()##)'.)
)
item(tt(?LPAR())...tt(RPAR()))(
Match zero or one occurrence.  (Like `tt(LPAR()|)...tt(RPAR())'.)
)
item(tt(!LPAR())...tt(RPAR()))(
Match anything but the expression in parentheses.
(Like `tt(LPAR()^LPAR())...tt(RPAR()RPAR())'.)
)
enditem()
subsect(Precedence)
cindex(precedence of glob operators)
The precedence of the operators given above is (highest) `tt(^)', `tt(/)',
`tt(~)', `tt(|)' (lowest); the
remaining operators are simply treated from left to right as part of a
string, with `tt(#)' and `tt(##)' applying to the shortest possible
preceding unit (i.e. a character, `tt(?)', `tt([)...tt(])',
`tt(<)...tt(>)', or a parenthesised expression).  As mentioned
above, a `tt(/)' used as a directory separator may not appear inside
parentheses, while a `tt(|)' must do so; in patterns used in other contexts
than filename generation (for example, in tt(case) statements and tests
within `tt([[)...tt(]])'), a `tt(/)' is not special; and `tt(/)' is also
not special after a `tt(~)' appearing outside parentheses in a filename
pattern.
subsect(Globbing Flags)
There are various flags which affect any text to their right up to the
end of the enclosing group or to the end of the pattern; they require
the tt(EXTENDED_GLOB) option. All take the form
tt(LPAR()#)var(X)tt(RPAR()) where var(X) may have one of the following
forms:

startitem()
item(tt(i))(
Case insensitive:  upper or lower case characters in the pattern match
upper or lower case characters.
)
item(tt(l))(
Lower case characters in the pattern match upper or lower case
characters; upper case characters in the pattern still only match
upper case characters.
)
item(tt(I))(
Case sensitive:  locally negates the effect of tt(i) or tt(l) from
that point on.
)
item(tt(b))(
Activate backreferences for parenthesised groups in the pattern;
this does not work in filename generation.  When a pattern with a set of
active parentheses is matched, the strings matched by the groups are
stored in the array tt($match), the indices of the beginning of the matched
parentheses in the array tt($mbegin), and the indices of the end in the array
tt($mend), with the first element of each array corresponding to the first
parenthesised group, and so on.  These arrays are not otherwise special to
the shell.  The indices use the same convention as does parameter
substitution, so that elements of tt($mend) and tt($mbegin) may be used in
subscripts; the tt(KSH_ARRAYS) option is respected.  Sets of globbing flags
are not considered parenthesised groups; only the first nine active
parentheses can be referenced.

For example,

example(foo="a string with a message"
if [[ $foo = (a|an)' '(#b)(*)' '* ]]; then
  print ${foo[$mbegin[1],$mend[1]]}
fi)

prints `tt(string with a)'.  Note that the first parenthesis is before the
tt((#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 tt(${)var(array)tt(#)var(pattern)tt(}), or a global substitution, such
as tt(${)var(param)tt(//)var(pat)tt(/)var(repl)tt(}), 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 tt(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 `tt(#)' or `tt(##)'.  Only the last match of the parenthesis is
remembered: for example, in `tt([[ abab = (#b)([ab])# ]])', only the final
`tt(b)' is stored in tt(match[1]).  Thus extra parentheses may be necessary
to match the complete segment: for example, use
`tt(X+LPAR()(ab|cd)#+RPAR()Y)' to match
a whole string of either `tt(ab)' or `tt(cd)' between `tt(X)' and `tt(Y)',
using the value of tt($match[1]) rather than tt($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 DASH()- which happens if they are in an alternate
branch which fails to match, or if they are followed by tt(#) and matched
zero times DASH()- 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.
)
item(tt(B))(
Deactivate backreferences, negating the effect of the tt(b) flag from that
point on.
)
item(tt(c)var(N)tt(,)var(M))(
The flag tt(LPAR()#c)var(N)tt(,)var(M)tt(RPAR()) can be used anywhere
that the tt(#) or tt(##) operators can be used; it cannot be combined
with other globbing flags and a bad pattern error occurs if it is
misplaced.  It is equivalent to the form tt({)var(N)tt(,)var(M)tt(}) in
regular expressions.  The previous character or group is required to
match between var(N) and var(M) times, inclusive.  The form
tt(LPAR()#c)var(N)tt(RPAR()) requires exactly tt(N) matches;
tt(LPAR()#c,)var(M)tt(RPAR()) is equivalent to specifying var(N) as 0;
tt(LPAR()#c)var(N)tt(,RPAR()) specifies that there is no maximum limit
on the number of matches.
)
item(tt(m))(
Set references to the match data for the entire string matched; this is
similar to backreferencing and does not work in filename generation.  The
flag must be in effect at the end of the pattern, i.e. not local to a
group. The parameters tt($MATCH),  tt($MBEGIN) and tt($MEND) will be set to
the string matched and to the indices of the beginning and end of the
string, respectively.  This is most useful in parameter substitutions, as
otherwise the string matched is obvious.

For example,

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
`tt(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.
)
item(tt(M))(
Deactivate the tt(m) flag, hence no references to match data will be
created.
)
item(tt(a)var(num))(
Approximate matching: var(num) errors are allowed in the string matched by
the pattern.  The rules for this are described in the next subsection.
)
item(tt(s), tt(e))(
Unlike the other flags, these have only a local effect, and each must
appear on its own:  `tt((#s))' and `tt((#e))' are the only valid forms.
The `tt((#s))' flag succeeds only at the start of the test string, and the
`tt((#e))' flag succeeds only at the end of the test string; they
correspond to `tt(^)' and `tt($)' in standard regular expressions.  They
are useful for matching path segments in patterns other than those in
filename generation (where path segments are in any case treated
separately).  For example, `tt(*((#s)|/)test+LPAR()(#e)|/+RPAR()*)' matches
a path segment `tt(test)' in any of the following strings: tt(test),
tt(test/at/start), tt(at/end/test), tt(in/test/middle).

Another use is in parameter substitution; for example
`tt(${array/(#s)A*Z+LPAR()#e+RPAR()})' will remove only elements of an
array which
match the complete pattern `tt(A*Z)'.  There are other ways of performing
many operations of this type, however the combination of the substitution
operations `tt(/)' and `tt(//)' with the `tt((#s))' and `tt((#e))' flags
provides a single simple and memorable method.

Note that assertions of the form `tt((^(#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 `tt((""~(#s)))' to match a zero-length portion of the string
not at the start.
)
item(tt(q))(
A `tt(q)' and everything up to the closing parenthesis of the globbing
flags are ignored by the pattern matching code.  This is intended to
support the use of glob qualifiers, see below.  The result is that
the pattern `tt((#b)(*).c+LPAR()#q.+RPAR())' can be used both for globbing
and for
matching against a string.  In the former case, the `tt((#q.))' will be
treated as a glob qualifier and the `tt((#b))' will not be useful, while in
the latter case the `tt((#b))' is useful for backreferences and the
`tt((#q.))' will be ignored.  Note that colon modifiers in the glob
qualifiers are also not applied in ordinary pattern matching.
)
item(tt(u))(
Respect the current locale in determining the presence of multibyte
characters in a pattern, provided the shell was compiled with 
tt(MULTIBYTE_SUPPORT).  This overrides the tt(MULTIBYTE)
option; the default behaviour is taken from the option.  Compare tt(U).
(Mnemonic: typically multibyte characters are from Unicode in the UTF-8
encoding, although any extension of ASCII supported by the system
library may be used.)
)
item(tt(U))(
All characters are considered to be a single byte long.  The opposite
of tt(u).  This overrides the tt(MULTIBYTE) option.
)
enditem()

For example, the test string tt(fooxx) can be matched by the pattern
tt(LPAR()#i)tt(RPAR()FOOXX), but not by tt(LPAR()#l)tt(RPAR()FOOXX),
tt(LPAR()#i)tt(RPAR()FOO)tt(LPAR()#I)tt(RPAR()XX) or
tt(LPAR()LPAR()#i)tt(RPAR()FOOX)tt(RPAR()X).  The string
tt(LPAR()#ia2)tt(RPAR()readme) specifies case-insensitive matching of
tt(readme) with up to two errors.

When using the ksh syntax for grouping both tt(KSH_GLOB) and
tt(EXTENDED_GLOB) must be set and the left parenthesis should be
preceded by tt(@).  Note also that the flags do not affect letters
inside tt([...]) groups, in other words tt(LPAR()#i)tt(RPAR()[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
tt(LPAR()#i)tt(RPAR()/foo/bar/...) is potentially slow.

subsect(Approximate Matching)
When matching approximately, the shell keeps a count of the errors found,
which cannot exceed the number specified in the
tt(LPAR()#a)var(num)tt(RPAR()) flags.  Four types of error are recognised:

startitem()
item(1.)(
Different characters, as in tt(fooxbar) and tt(fooybar).
)
item(2.)(
Transposition of characters, as in tt(banana) and tt(abnana).
)
item(3.)(
A character missing in the target string, as with the pattern tt(road) and
target string tt(rod).
)
item(4.)(
An extra character appearing in the target string, as with tt(stove)
and tt(strove).
)
enditem()

Thus, the pattern tt(LPAR()#a3)tt(RPAR()abcd) matches tt(dcba), with the
errors occurring by using the first rule twice and the second once,
grouping the string as tt([d][cb][a]) and tt([a][bc][d]).

Non-literal parts of the pattern must match exactly, including characters
in character ranges: hence tt(LPAR()#a1)tt(RPAR()???)  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 tt(?) must match.  Other characters
which must match exactly are initial dots in filenames (unless the
tt(GLOB_DOTS) option is set), and all slashes in filenames, so that
tt(a/bc) is two errors from tt(ab/c) (the slash cannot be transposed with
another character).  Similarly, errors are counted separately for
non-contiguous strings in the pattern, so that tt(LPAR()ab|cd)tt(RPAR()ef)
is two errors from tt(aebf).

When using exclusion via the tt(~) operator, approximate matching is
treated entirely separately for the excluded part and must be activated
separately.  Thus, tt(LPAR()#a1)tt(RPAR()README~READ_ME) matches
tt(READ.ME) but not tt(READ_ME), as the trailing tt(READ_ME) is matched
without approximation.  However,
tt(LPAR()#a1)tt(RPAR()README~LPAR()#a1)tt(RPAR()READ_ME)
does not match any pattern of the form tt(READ)var(?)tt(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,
tt(LPAR()#a1)tt(RPAR()cat)tt(LPAR()LPAR()#a0)tt(RPAR()dog)tt(RPAR()fox)
allows one error in total, which may not occur in the tt(dog) section, and
the pattern
tt(LPAR()#a1)tt(RPAR()cat)tt(LPAR()#a0)tt(RPAR()dog)tt(LPAR()#a1)tt(RPAR()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,
tt((#a1)abc+LPAR()#a0+RPAR()xyz) will not match tt(abcdxyz), because the
error occurs at the `tt(x)', where approximation is turned off.

Entire path segments may be matched approximately, so that
`tt((#a1)/foo/d/is/available/at/the/bar)' allows one error in any path
segment.  This is much less efficient than without the tt((#a1)), however,
since every directory in the path must be scanned for a possible
approximate match.  It is best to place the tt((#a1)) after any path
segments which are known to be correct.

subsect(Recursive Globbing)
A pathname component of the form `tt(LPAR())var(foo)tt(/RPAR()#)'
matches a path consisting of zero or more directories
matching the pattern var(foo).

As a shorthand, `tt(**/)' is equivalent to `tt((*/)#)'; note that this
therefore matches files in the current directory as well as
subdirectories.
Thus:

example(ls (*/)#bar)

or

example(ls **/bar)

does a recursive directory search for files named `tt(bar)' (potentially
including the file `tt(bar)' in the current directory).  This form does not
follow symbolic links; the alternative form `tt(***/)' 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 `tt(*)'
operators revert to their usual effect.
subsect(Glob Qualifiers)
cindex(globbing, qualifiers)
cindex(qualifiers, globbing)
Patterns used for filename generation may end in a
list of qualifiers enclosed in parentheses.
The qualifiers specify which filenames that otherwise match the given pattern
will be inserted in the argument list.

pindex(BARE_GLOB_QUAL, use of)
If the option tt(BARE_GLOB_QUAL) is set, then a trailing set of parentheses
containing no `tt(|)' or `tt(LPAR())' characters (or `tt(~)' 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 `tt((^x))', can be forced to be treated as part of
the glob pattern by doubling the parentheses, in this case producing
`tt(((^x)))'.

If the option tt(EXTENDED_GLOB) is set, a different syntax for glob
qualifiers is available, namely `tt((#qx))' where tt(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 `tt(|)', `tt(LPAR())' or
`tt(~)' 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 `tt(*(#q*)(.))' will recognise executable regular
files if both options are set; however, mixed syntax should probably be
avoided for the sake of clarity.

A qualifier may be any one of the following:

startitem()
item(tt(/))(
directories
)
item(tt(F))(
`full' (i.e. non-empty) directories.  Note that the
opposite sense tt(LPAR()^F)tt(RPAR()) expands to empty directories
and all non-directories.  Use tt(LPAR()/^F)tt(RPAR()) for
empty directories
)
item(tt(.))(
plain files
)
item(tt(@))(
symbolic links
)
item(tt(=))(
sockets
)
item(tt(p))(
named pipes (FIFOs)
)
item(tt(*))(
executable plain files (0100)
)
item(tt(%))(
device files (character or block special)
)
item(tt(%b))(
block special files
)
item(tt(%c))(
character special files
)
item(tt(r))(
owner-readable files (0400)
)
item(tt(w))(
owner-writable files (0200)
)
item(tt(x))(
owner-executable files (0100)
)
item(tt(A))(
group-readable files (0040)
)
item(tt(I))(
group-writable files (0020)
)
item(tt(E))(
group-executable files (0010)
)
item(tt(R))(
world-readable files (0004)
)
item(tt(W))(
world-writable files (0002)
)
item(tt(X))(
world-executable files (0001)
)
item(tt(s))(
setuid files (04000)
)
item(tt(S))(
setgid files (02000)
)
item(tt(t))(
files with the sticky bit (01000)
)
item(tt(f)var(spec))(
files with access rights matching var(spec). This var(spec) may be a
octal number optionally preceded by a `tt(=)', a `tt(PLUS())', or a
`tt(-)'. If none of these characters is given, the behavior is the
same as for `tt(=)'. The octal number describes the mode bits to be
expected, if combined with a `tt(=)', the value given must match the
file-modes exactly, with a `tt(PLUS())', at least the bits in the
given number must be set in the file-modes, and with a `tt(-)', the
bits in the number must not be set. Giving a `tt(?)' instead of a
octal digit anywhere in the number ensures that the corresponding bits 
in the file-modes are not checked, this is only useful in combination
with `tt(=)'.

If the qualifier `tt(f)' is followed by any other character anything
up to the next matching character (`tt([)', `tt({)', and `tt(<)' match 
`tt(])', `tt(})', and `tt(>)' respectively, any other character
matches itself) is taken as a list of comma-separated
var(sub-spec)s. Each var(sub-spec) may be either an octal number as
described above or a list of any of the characters `tt(u)', `tt(g)',
`tt(o)', and `tt(a)', followed by a `tt(=)', a `tt(PLUS())', or a
`tt(-)', followed by a list of any of the characters `tt(r)', `tt(w)', 
`tt(x)', `tt(s)', and `tt(t)', or an octal digit. The first list of
characters specify which access rights are to be checked. If a `tt(u)'
is given, those for the owner of the file are used, if a `tt(g)' is
given, those of the group are checked, a `tt(o)' means to test those
of other users, and the `tt(a)' says to test all three groups. The
`tt(=)', `tt(PLUS())', and `tt(-)' 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: `tt(r)' for read access, `tt(w)' for write access, 
`tt(x)' for the right to execute the file (or to search a directory),
`tt(s)' for the setuid and setgid bits, and `tt(t)' for the sticky
bit.

Thus, `tt(*(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
`tt(*(f-100))' gives all files for which the owner does not have
execute permission, and `tt(*(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.
)
xitem(tt(e)var(string))
item(tt(PLUS())var(cmd))(
The var(string) will be executed as shell code.  The filename will be
included in the list if and only if the code returns a zero status (usually
the status of the last command).  The first character after the `tt(e)'
will be used as a separator and anything up to the next matching separator
will be taken  as the var(string); `tt([)', `tt({)', and `tt(<)' match
`tt(])', `tt(})', and `tt(>)', respectively, while any other character
matches itself. Note that expansions must be quoted in the var(string)
to prevent them from being expanded before globbing is done.

vindex(REPLY, use of)
vindex(reply, use of)
During the execution of var(string) the filename currently being tested is
available in the parameter tt(REPLY); the parameter may be altered to
a string to be inserted into the list instead of the original
filename.  In addition, the parameter tt(reply) may be set to an array or a
string, which overrides the value of tt(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 `tt(lonely)'.  Then
the expression `tt(*(e:'reply=(${REPLY}{1,2})':))' will cause the words
`tt(lonely1 lonely2)' to be inserted into the command line.  Note the
quotation marks.

The form tt(PLUS())var(cmd) has the same effect, but no delimiters appear
around var(cmd).  Instead, var(cmd) is taken as the longest sequence of
characters following the tt(PLUS()) that are alphanumeric or underscore.
Typically var(cmd) will be the name of a shell function that contains the
appropriate test.  For example,

example(nt+LPAR()RPAR() { [[ $REPLY -nt $NTREF ]] }
NTREF=reffile
ls -l *(+nt))

lists all files in the directory that have been modified more recently than
tt(reffile).
)
item(tt(d)var(dev))(
files on the device var(dev)
)
item(tt(l)[tt(-)|tt(PLUS())]var(ct))(
files having a link count less than var(ct) (tt(-)), greater than
var(ct) (tt(PLUS())), or equal to var(ct)
)
item(tt(U))(
files owned by the effective user ID
)
item(tt(G))(
files owned by the effective group ID
)
item(tt(u)var(id))(
files owned by user ID var(id) if that is a number.  Otherwise,
var(id) specifies a user name: the
character after the `tt(u)' will be taken as a separator and the string
between it and the next matching separator will be taken as a user name.
The starting separators `tt([)', `tt({)', and `tt(<)'
match the final separators `tt(])', `tt(})', and `tt(>)', respectively;
any other character matches itself.  The selected files are those
owned by this user.  For example, `tt(u:foo:)' or `tt(u[foo])' selects
files owned by user `tt(foo)'.
)
item(tt(g)var(id))(
like tt(u)var(id) but with group IDs or names
)
item(tt(a)[tt(Mwhms)][tt(-)|tt(PLUS())]var(n))(
files accessed exactly var(n) days ago.  Files accessed within the last
var(n) days are selected using a negative value for var(n) (tt(-)var(n)).
Files accessed more than var(n) days ago are selected by a positive var(n)
value (tt(PLUS())var(n)).  Optional unit specifiers `tt(M)', `tt(w)',
`tt(h)', `tt(m)' or `tt(s)' (e.g. `tt(ah5)') cause the check to be
performed with months (of 30 days), weeks, hours, minutes or seconds
instead of days, respectively.

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, `tt(echo *(ah-5))' would echo files accessed within the last
five hours, while `tt(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.
)
item(tt(m)[tt(Mwhms)][tt(-)|tt(PLUS())]var(n))(
like the file access qualifier, except that it uses the file modification
time.
)
item(tt(c)[tt(Mwhms)][tt(-)|tt(PLUS())]var(n))(
like the file access qualifier, except that it uses the file inode change
time.
)
item(tt(L)[tt(PLUS())|tt(-)]var(n))(
files less than var(n) bytes (tt(-)), more than var(n) bytes (tt(PLUS())), or
exactly var(n) bytes in length. If this flag is directly followed by a `tt(k)'
(`tt(K)'), `tt(m)' (`tt(M)'), or `tt(p)' (`tt(P)') (e.g. `tt(Lk-50)')
the check is performed with kilobytes, megabytes, or blocks (of 512 bytes)
instead.
)
item(tt(^))(
negates all qualifiers following it
)
item(tt(-))(
toggles between making the qualifiers work on symbolic links (the
default) and the files they point to
)
item(tt(M))(
sets the tt(MARK_DIRS) option for the current pattern
pindex(MARK_DIRS, setting in pattern)
)
item(tt(T))(
appends a trailing qualifier mark to the filenames, analogous to the
tt(LIST_TYPES) option, for the current pattern (overrides tt(M))
)
item(tt(N))(
sets the tt(NULL_GLOB) option for the current pattern
pindex(NULL_GLOB, setting in pattern)
)
item(tt(D))(
sets the tt(GLOB_DOTS) option for the current pattern
pindex(GLOB_DOTS, setting in pattern)
)
item(tt(n))(
sets the tt(NUMERIC_GLOB_SORT) option for the current pattern
pindex(NUMERIC_GLOB_SORT, setting in pattern)
)
item(tt(o)var(c))(
specifies how the names of the files should be sorted. If var(c) is
tt(n) they are sorted by name (the default); if it is tt(L) they
are sorted depending on the size (length) of the files; if tt(l)
they are sorted by the number of links; if tt(a), tt(m), or tt(c)
they are sorted by the time of the last access, modification, or
inode change respectively; if tt(d), files in subdirectories appear before
those in the current directory at each level of the search DASH()- this is best
combined with other criteria, for example `tt(odon)' to sort on names for
files within the same directory; if tt(N), no sorting is performed.
Note that tt(a), tt(m), and tt(c) compare
the age against the current time, hence the first name in the list is the
youngest file. Also note that the modifiers tt(^) and tt(-) are used,
so `tt(*(^-oL))' gives a list of all files sorted by file size in descending
order, following any symbolic links.  Unless tt(oN) is used, multiple order
specifiers may occur to resolve ties.

tt(oe) and tt(o+) are special cases; they are each followed by shell code,
delimited as for the tt(e) glob qualifier and the tt(+) glob qualifier
respectively (see above).  The code is executed for each matched file with
the parameter tt(REPLY) set to the name of the file on entry.  The code
should modify the parameter tt(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, tt(oe) and tt(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.
)
item(tt(O)var(c))(
like `tt(o)', but sorts in descending order; i.e. `tt(*(^oc))' is the
same as `tt(*(Oc))' and `tt(*(^Oc))' is the same as `tt(*(oc))'; `tt(Od)'
puts files in the current directory before those in subdirectories at each
level of the search.
)
item(tt([)var(beg)[tt(,)var(end)]tt(]))(
specifies which of the matched filenames should be included in the
returned list. The syntax is the same as for array
subscripts. var(beg) and the optional var(end) may be mathematical
expressions. As in parameter subscripting they may be negative to make 
them count from the last match backward. E.g.: `tt(*(-OL[1,3]))'
gives a list of the names of the three largest files.
)
enditem()

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 `tt(M)', `tt(T)',
`tt(N)', `tt(D)', `tt(n)', `tt(o)', `tt(O)' and the subscripts given
in brackets (`tt([...])').

If a `tt(:)' appears in a qualifier list, the remainder of the expression in
parenthesis is interpreted as a modifier (see noderef(Modifiers)
in noderef(History Expansion)).  Each modifier must be introduced by a
separate `tt(:)'.  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 `tt((:..))' 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 tt(NULL_GLOB). Thus:

example(ls *(-/))

lists all directories and symbolic links that point to directories,
and

example(ls *(%W))

lists all world-writable device files in the current directory, and

example(ls *(W,X))

lists all files in the current directory that are
world-writable or world-executable, and

example(echo /tmp/foo*(u0^@:t))

outputs the basename of all root-owned files beginning with the string
`tt(foo)' in tt(/tmp), ignoring symlinks, and

example(ls *.*~(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 tt(GLOB_DOTS) is explicitly
switched off) except for tt(lex.c), tt(lex.h), tt(parse.c) and tt(parse.h).

example(print b*.pro+LPAR()#q:s/pro/shmo/+RPAR()(#q.:s/builtin/shmiltin/))

demonstrates how colon modifiers and other qualifiers may be chained
together.  The ordinary qualifier `tt(.)' is applied first, then the colon
modifiers in order from left to right.  So if tt(EXTENDED_GLOB) is set and
the base pattern matches the regular file tt(builtin.pro), the shell will
print `tt(shmiltin.shmo)'.