texinode(Redirection)(Command Execution)(Shell Grammar)(Top)
chapter(Redirection)
cindex(redirection)
ifzman(\
sect(Redirection)
)\
cindex(file descriptors)
cindex(descriptors, file)
If a command is followed by tt(&)
and job control is not active,
then the default standard input
for the command is the empty file tt(/dev/null).
Otherwise, the environment for the execution of a command contains the
file descriptors of the invoking shell as modified by
input/output specifications.

The following may appear anywhere in a simple command
or may precede or follow a complex command.
Expansion occurs before var(word) or var(digit)
is used except as noted below.
If the result of substitution on var(word)
produces more than one filename,
redirection occurs for each
separate filename in turn.

startitem()
item(tt(<) var(word))(
Open file var(word) for reading as standard input.
It is an error to open a file in this fashion if it does not exist.
)
item(tt(<>) var(word))(
Open file var(word) for reading and writing as standard input.
If the file does not exist then it is created.
)
item(tt(>) var(word))(
Open file var(word) for writing as standard output.
If the file does not exist then it is created.
If the file exists, and the tt(CLOBBER) option is unset,
this causes an error;
otherwise, it is truncated to zero length.
)
xitem(tt(>|) var(word))
item(tt(>!) var(word))(
Same as tt(>), except that the file is truncated to zero length
if it exists, regardless of tt(CLOBBER).
)
item(tt(>>) var(word))(
Open file var(word) for writing in append mode as standard output.
If the file does not exist, and the tt(CLOBBER) and tt(APPEND_CREATE)
options are both unset, this causes an error;
otherwise, the file is created.
)
xitem(tt(>>|) var(word))
item(tt(>>!) var(word))(
Same as tt(>>), except that the file is created if it does not
exist, regardless of tt(CLOBBER) and tt(APPEND_CREATE).
)
item(tt(<<)[tt(-)] var(word))(
The shell input is read up to a line that is the same as
var(word), or to an end-of-file.
No parameter expansion, command substitution or
filename generation is performed on var(word).
The resulting document, called a
em(here-document), becomes the standard input.

If any character of var(word) is quoted with
single or double quotes or a `tt(\)',
no interpretation is placed upon the characters of the document.
Otherwise, parameter and command substitution
occurs, `tt(\)' followed by a newline is removed,
and `tt(\)' must be used to quote the characters
`tt(\)', `tt($)', `tt(`)' and the first character of var(word).

Note that var(word) itself does not undergo shell expansion.  Backquotes
in var(word) do not have their usual effect; instead they behave
similarly to double quotes, except that the backquotes themselves are
passed through unchanged.  (This information is given for completeness
and it is not recommended that backquotes be used.)  Quotes in the form
tt($')var(...)tt(') have their standard effect of expanding backslashed
references to special characters.

If tt(<<-) is used, then all leading
tabs are stripped from var(word) and from the document.
)
item(tt(<<<) var(word))(
Perform shell expansion on var(word) and pass the result
to standard input.  This is known as a em(here-string).
Compare the use of var(word) in here-documents above, where var(word)
does not undergo shell expansion.
)
xitem(tt(<&) var(number))
item(tt(>&) var(number))(
The standard input/output is duplicated from file descriptor
var(number) (see manref(dup2)(2)).
)
xitem(tt(<& -))
item(tt(>& -))(
Close the standard input/output.
)
xitem(tt(<& p))
item(tt(>& p))(
The input/output from/to the coprocess is moved to the standard input/output.
)
xitem(tt(>&) var(word))
item(tt(&>) var(word))(
(Except where `tt(>&) var(word)' matches one of the above syntaxes;
`tt(&>)' can always be used to avoid this ambiguity.)
Redirects both standard output and standard error (file descriptor 2)
in the manner of `tt(>) var(word)'.
Note that this does em(not) have the same effect as `tt(>) var(word) tt(2>&1)'
in the presence of multios (see the section below).
)
xitem(tt(>&|) var(word))
xitem(tt(>&!) var(word))
xitem(tt(&>|) var(word))
item(tt(&>!) var(word))(
Redirects both standard output and standard error (file descriptor 2)
in the manner of `tt(>|) var(word)'.
)
xitem(tt(>>&) var(word))
item(tt(&>>) var(word))(
Redirects both standard output and standard error (file descriptor 2)
in the manner of `tt(>>) var(word)'.
)
xitem(tt(>>&|) var(word))
xitem(tt(>>&!) var(word))
xitem(tt(&>>|) var(word))
item(tt(&>>!) var(word))(
Redirects both standard output and standard error (file descriptor 2)
in the manner of `tt(>>|) var(word)'.
)
enditem()

If one of the above is preceded by a digit, then the file
descriptor referred to is that specified by the digit
instead of the default 0 or 1.
The order in which redirections are specified is significant.
The shell evaluates each redirection in terms of the
(em(file descriptor), em(file))
association at the time of evaluation.
For example:

indent(... tt(1>)var(fname) tt(2>&1))

first associates file descriptor 1 with file var(fname).
It then associates file descriptor 2 with the file associated with file
descriptor 1 (that is, var(fname)).
If the order of redirections were reversed,
file descriptor 2 would be associated
with the terminal (assuming file descriptor 1 had been)
and then file descriptor 1 would be associated with file var(fname).

The `tt(|&)' command separator described in
ifzman(em(Simple Commands & Pipelines) in zmanref(zshmisc))\
ifnzman(noderef(Simple Commands & Pipelines))
is a shorthand for `tt(2>&1 |)'.

The various forms of process substitution, `tt(<LPAR())var(list)tt(RPAR())',
and `tt(=LPAR())var(list)tt(RPAR())' for input and
`tt(>LPAR())var(list)tt(RPAR())' for output, are often used together with
redirection.  For example, if var(word) in an output redirection is of the
form `tt(>LPAR())var(list)tt(RPAR())' then the output is piped to the
command represented by var(list).  See
ifzman(\
em(Process Substitution) in zmanref(zshexpn))\
ifnzman(\
noderef(Process Substitution)).
sect(Opening file descriptors using parameters)
cindex(file descriptors, use with parameters)
cindex(parameters, for using file descriptors)

When the shell is parsing arguments to a command, and the shell option
tt(IGNORE_BRACES) is not set, a different form of redirection is allowed:
instead of a digit before the operator there is a valid shell identifier
enclosed in braces.  The shell will open a new file descriptor that
is guaranteed to be at least 10 and set the parameter named by the
identifier to the file descriptor opened.  No whitespace is allowed
between the closing brace and the redirection character.  For example:

indent(... tt({myfd}>&1))

This opens a new file descriptor that is a duplicate of file descriptor
1 and sets the parameter tt(myfd) to the number of the file descriptor,
which will be at least 10.  The new file descriptor can be written to using
the syntax tt(>&$myfd).  The file descriptor remains open in subshells
and forked external executables.

The syntax tt({)var(varid)tt(}>&-), for example tt({myfd}>&-), may be used
to close a file descriptor opened in this fashion.  Note that the
parameter given by var(varid) must previously be set to a file descriptor
in this case.

It is an error to open or close a file descriptor in this fashion when the
parameter is readonly.  However, it is not an error to read or write a file
descriptor using tt(<&$)var(param) or tt(>&$)var(param) if var(param) is
readonly.

If the option tt(CLOBBER) is unset, it is an error to open a file
descriptor using a parameter that is already set to an open file descriptor
previously allocated by this mechanism.  Unsetting the parameter before
using it for allocating a file descriptor avoids the error.

Note that this mechanism merely allocates or closes a file descriptor; it
does not perform any redirections from or to it.  It is usually convenient
to allocate a file descriptor prior to use as an argument to tt(exec).
The syntax does not in any case work when used around complex commands
such as parenthesised subshells or loops, where the opening brace is
interpreted as part of a command list to be executed in the current shell.

The following shows a typical sequence of allocation, use, and closing of a
file descriptor:

example(integer myfd
exec {myfd}>~/logs/mylogfile.txt
print This is a log message. >&$myfd
exec {myfd}>&-)

Note that the expansion of the variable in the expression tt(>&$myfd)
occurs at the point the redirection is opened.  This is after the expansion
of command arguments and after any redirections to the left on the command
line have been processed.
sect(Multios)
cindex(multios)
pindex(MULTIOS, use of)
If the user tries to open a file descriptor for writing more than once,
the shell opens the file descriptor as a pipe to a process that copies
its input to all the specified outputs, similar to bf(tee),
provided the tt(MULTIOS) option is set, as it is by default.  Thus:

example(date >foo >bar)

writes the date to two files, named `tt(foo)' and `tt(bar)'.
Note that a pipe is an implicit redirection; thus

example(date >foo | cat)

writes the date to the file `tt(foo)', and also pipes it to cat.

Note that the shell opens all the files to be used in the multio process
immediately, not at the point they are about to be written.

Note also that redirections are always expanded in order.  This happens
regardless of the setting of the tt(MULTIOS) option, but with the option
in effect there are additional consequences. For example,
the meaning of the expression tt(>&1) will change after a previous
redirection:

example(date >&1 >output)

In the case above, the tt(>&1) refers to the standard output at the
start of the line; the result is similar to the tt(tee) command.
However, consider:

example(date >output >&1)

As redirections are evaluated in order, when the tt(>&1) is encountered
the standard output is set to the file tt(output) and another copy of
the output is therefore sent to that file.  This is unlikely to be what
is intended.

If the tt(MULTIOS)
option is set, the word after a redirection operator is also subjected
to filename generation (globbing).  Thus

example(: > *)

will truncate all files in the current directory,
assuming there's at least one.  (Without the tt(MULTIOS)
option, it would create an empty file called `tt(*)'.)
Similarly, you can do

example(echo exit 0 >> *.sh)

If the user tries to open a file descriptor for reading more than once,
the shell opens the file descriptor as a pipe to a process that copies
all the specified inputs to its output in the order specified, provided
the tt(MULTIOS) option is set.  It should be noted that each file is
opened immediately, not at the point where it is about to be read:
this behaviour differs from tt(cat), so if strictly standard behaviour
is needed, tt(cat) should be used instead.

Thus

example(sort <foo <fubar)

or even

example(sort <f{oo,ubar})

is equivalent to `tt(cat foo fubar | sort)'.

Expansion of the redirection argument occurs at the point the redirection
is opened, at the point described above for the expansion of the variable
in tt(>&$myfd).

Note that a pipe is an implicit redirection; thus

example(cat bar | sort <foo)

is equivalent to `tt(cat bar foo | sort)' (note the order of the inputs).

If the tt(MULTIOS) option is em(un)set,
each redirection replaces the previous redirection for that file descriptor.
However, all files redirected to are actually opened, so

example(echo Hello > bar > baz)

when tt(MULTIOS) is unset will truncate `tt(bar)', and write `tt(Hello)'
into `tt(baz)'.

There is a problem when an output multio is attached to an external
program.  A simple example shows this:

example(cat file >file1 >file2
cat file1 file2)

Here, it is possible that the second `tt(cat)' will not display the full
contents of tt(file1) and tt(file2) (i.e. the original contents of
tt(file) repeated twice).

The reason for this is that the multios are spawned after the tt(cat)
process is forked from the parent shell, so the parent shell does not
wait for the multios to finish writing data.  This means the command as
shown can exit before tt(file1) and tt(file2) are completely written.
As a workaround, it is possible to run the tt(cat) process as part of a
job in the current shell:

example({ cat file } >file >file2)

Here, the tt({)var(...)tt(}) job will pause to wait for both files to be
written.

sect(Redirections with no command)
When a simple command consists of one or more redirection operators
and zero or more parameter assignments, but no command name, zsh can
behave in several ways.

vindex(NULLCMD, use of)
pindex(CSH_NULLCMD, use of)
If the parameter tt(NULLCMD) is not set or the option tt(CSH_NULLCMD) is
set, an error is caused.  This is the bf(csh) behavior and tt(CSH_NULLCMD)
is set by default when emulating bf(csh).

pindex(SH_NULLCMD, use of)
If the option tt(SH_NULLCMD) is set, the builtin `tt(:)' is inserted as a
command with the given redirections.  This is the default when emulating
bf(sh) or bf(ksh).

vindex(READNULLCMD, use of)
Otherwise, if the parameter tt(NULLCMD) is set, its value will be used as a
command with the given redirections.  If both tt(NULLCMD) and
tt(READNULLCMD) are set, then the value of the latter will be used instead
of that of the former when the redirection is an input.  The default for
tt(NULLCMD) is `tt(cat)' and for tt(READNULLCMD) is `tt(more)'. Thus

example(< file)

shows the contents of tt(file) on standard output, with paging if that is a
terminal.  tt(NULLCMD) and tt(READNULLCMD) may refer to shell functions.