namespace

namespace is a built-in ensemble of commands for working with namespaces.

Synopsis

namespace children ?namespace? ?pattern?
namespace code script
namespace current
namespace delete ?namespace namespace ...?
namespace ensemble option ?arg ...?
namespace eval namespace arg ?arg ...?
namespace exists namespace
namespace export ?-clear? ?pattern pattern ...?
namespace forget ?pattern pattern ...?
namespace import ?-force? ?pattern pattern ...?
namespace inscope namespace script ?arg ...?
namespace origin command
namespace parent ?namespace?
namespace path ?namespaceList?
namespace qualifiers string
namespace tail string
namespace upvar namespace otherVar myVar ?otherVar myVar ...?
namespace unknown ?script?
namespace which ?-command? ?-variable? name

See Also

::
The namespace delimiter.
ns, by stu
An additional set of commands for working with namespaces.
namespace split
Split a name into its namespace components.
namespace size
Thoughts on Namespaces and OO
Namespace variables 2
Namespace resolution of Variables & Procedures
data is code
_namespace import_ and _rename_
Illustrates the flexibility of namespaces.
closures
Used in other languages to do what is done in Tcl using namespaces, apply, and coroutines.

Documentation

official reference documentation
Namespaces and Packages , by William Duquette
XML Namespaces FAQ
Information about the general concept of namespaces.
A Gentle Introduction to Namespaces , by Sam Ruby 2002
A description of namespace in XML.

Proposals

TIP 283, Modify Ensemble Command Resolution Behaviour

Description

Each namespace is actually three separate namespaces: One for commands, one for variables, and one for nested namespaces. Configurable aspects of a namspace include routine resolution, handling of routine lookup failures, namespace ensemble. :: is used delimit the namespace name and the name of an object in that namespace For example:

namespace eval ::one::two::three {
    set four {} 
    proc four {} {}
    namespace eval four {}
}

namespace which -variable ::one::two::three::four ;# -> ::one::two::three::four
namespace which -command ::one::two::three::four ;# -> ::one::two::three::four
namespace children ::one::two::three ;# -> ::one::two::three::four

Namespaces make it possible to encapsulate data and functionality into logical units. Changing the content of a namespace you don't control can have unanticipated consequences. Read the documentation for that namespace to understand how to interact with it.

Commands that create namespaces include namespace eval, interp alias, oo::class, oo::object, and oo::copy.

Introspection

Can someone discuss how a Tcl script can determine:

  • what namespaces are available at any particular point in time
  • what procs and variables available in any one of these namespaces

A simple proc to list all child namespaces, recursively:

proc listns {{parentns ::}} {
    foreach ns [namespace children $parentns] {
        eval lappend result [listns $ns]
        lappend result $ns
    }
    return $result
}

output:

% namespace eval ::test {set foo 1}
1

% namespace eval ::test2 {set foo 1}
1

% namespace eval ::test::test3 {set foo 1}
1

% listns
::test::test3 ::test ::test2 ::tcl

From 8.5 onwards, use this instead:

proc listns {{parentns ::}} {
    set result [list]
    foreach ns [namespace children $parentns] {
        lappend result {*}[listns $ns] $ns
    }
    return $result
}

list the variables in a namespace:

proc listnsvars {{ns ::}} {
    return [info vars ${ns}::*]
}

list the procs in a namespace:

proc listnsprocs {{ns ::}} {
    return [info procs ${ns}::*]
}

list the commands in a namespace:

proc listnscommands {{ns ::}} {
    return [info commands ${ns}::*]
}

The previous examples only work well with fully qualified namespace names.

Beware that in some apps these may take a long time to complete (for example, if there are hundreds of thousands of commands in ::).

Avoiding variable declarations in procedures

JCW wrote in comp.lang.tcl: For variables, I've found a child namespace to be convenient. Instead of:

namespace eval TO {
    variable foo
    proc proc1 {} {
        variable foo
        ... use $foo ...
    }
}

Consider using:

namespace eval TO {
    namespace eval v {
        variable foo
    }
    proc proc1 {} {
        ... use $v::foo ...
    }
}

No more lists of variable calls inside each proc. The other way is to put all your variables inside a single array, but my impression is that the v::* approach is slightly faster (variable seems to take a bit of time, don't know why).

Command Name Resolution

To resolve a command name, Tcl first looks in the current namespace, then in the namespaces in the path of the current namespace, and then in the global namespace. It does not look in any intermediate containing namespaces, unless they happen to be in the path of the current namespace:

proc ::p {} {
    puts {I'm ::p}
}

namespace eval ns1 {
    proc p {} {
        puts {I'm ns::p}
    }
    namespace eval child {
        p
    }
}

Result:

I'm ::p

Tcl resolved the name in the global namespace and didn't search in the containing ns1 namespace.

This behaviour can be modified via namespace path.


LVwikignoming 2010-07-01 13:03:52 PYK 2020-01-24:

A recent email on the TCLCORE mailing list discussed a problem. The op was attempting to add a new routine to array, but the results were unexpected. The solution, as provided by dkf, was to use ::set instead of just plain set in the body of the new routine, because ::tcl::array::set overrides ::set in that namespace.

So, this note is not specific to set, but applies in general. If you are seeing behavior that reports a problem that is unexpected, examine the code closely to see if you need to add namespacing to the commands in use.

Make an Alias for a Command

kruzalex 2013-04-04:

Currently, the typical idiom for importing commands as functions into tcl::mathfunc is to use interp alias, which has the nasty effect of incurring some performance degradation on each invocation of the alias.

If the source command has been exported from its own namespace, then it can be imported into another, which makes execution of the imported command considerably faster, and makes the new command robust against later renaming of the origin command.

Sometimes one might wish for more than what namespace import currently provides:

  1. one might want to access commands from a namespace where they are not exported from. e.g.: ::tcl::unsupported doesn't export anything, so disassemble cannot just be imported.
  2. one might want to import commands to a different name, e.g., importing tcl::string::length to a local strlen
  3. a special one: if one tries to import only the multiplication operator from ::tcl::mathop::* then the asterisk is interpreted as a pattern, and everything is going to get imported.

I've written a procedure that looks a bit complicated/awkward, but addresses all these shortcomings. Here it is:

proc alias {src tgt} {
    set nsrc [namespace qualifiers $src]
    set tsrc [namespace tail $src]
    set oe [uplevel 1 [list namespace eval $nsrc {namespace export}]]
    uplevel 1 [list namespace eval $nsrc {namespace export *}]
    set n 0
    while {[namespace exists tmp-[incr n]]} {}
    set fqsrc [uplevel 1 [list namespace which $src]]
    namespace eval tmp-$n [list namespace import $fqsrc]
    uplevel 1 [list rename [namespace which -command "tmp-${n}::$tsrc"] $tgt]
    uplevel 1 [list namespace eval $nsrc [list namespace export -clear {*}$oe]]
    namespace delete tmp-$n
}

PYK 2014-06-23: See also proc alias for another implementation of the same.

Since Tcl8.6, you could create this procedure named as ::tcl::namespace::alias instead, and add it to the ensemble with three more awkward lines:

set nsmap [namespace ensemble configure namespace -map]
::dict set nsmap alias ::tcl::namespace::alias
namespace ensemble configure namespace -map $nsmap

In Tcl8.5 namespace isn't a namespace ensemble yet, but the global proc alias works just fine with Tcl8.5.

Examples:

# import "string length" as "strlen" to be used within expr
alias ::tcl::string::length ::tcl::mathfunc::strlen

# save some typing in an interactive shell :)
alias ::tcl::unsupported::disassemble disasm

Note: unlike interp alias, you can not append any extra arguments with this "alias". However, if the intended argument is really just the subcommand of an ensemble (as is string length), then you can simply alias the subcommand directly :-)

Known bugs: specifying bad arguments may leave some empty namespace and export-patterns behind. That could be fixed with try or catch, but I thought it was already complicated enough as it is without fancy error handling. Also, it would be nice to be able to specify more src/tgt pairs on one invocation...

kwilsoft: The copied command is still the imported command from the origin namespace. Hence, when removing command from origin namespace, the alias command will also cease to work. Better to copy the args and body of the original proc to the new namespace proc.

PYK 2020-01-24: It's a question of knowing what you want to do. Sometimes, namespace import semantics are exactly what you want. interp alias is an alternative that can dispatch back to another routine, whatever it happens to be at the moment. Copying the args and the body of the original procedure is a distant third option. If you're doing it, your design is probably suboptimal.

Duplicate a Namespace

There is no built-in namespace dup command, but such an action can be useful when a namespace is part of the implementation of an object system. Tcl's copy-on-write semantics make this less costly than it might at first seem. Duplicating a namespace involves duplicating some or all of the following:

variables
Distinguish variables declared by upvar or namespace upvar. Also distinguish arrays. Duplicate variables that are declared but not defined.
variable traces
TO DO
commands
make sure to distinguish between native and imported commands. Since it is impossible to import by exact name a command whose name contains a glob pattern character such as * or ?, extra care must be taken to ensure that the proper commands were imported. Since commands that are not procedures can't be copied, make an alias to them. Also make sure to respect import aliases that have been renamed.
command traces
TO DO
exports
simple copy
path
simple copy
nested namespaces
recursively duplicate, but make sure not to descend into the target if a namespace is duplicated to some descendant.
namespace unknown
simple copy

ycl::ns::duplicate implements namespace duplication as described above, except for duplicating traces.

Misc

JeremyM 2007-09-19: I'm struggling with the decision as to whether to use a namespace or an object to hold code and variables that are a singleton, i.e. if I use an object, I will always instantiate exactly one instance of the object. Because of this, a namespace seems like a more appropriate container, but I'm getting the feeling that namespaces are somewhat obsolete. Plus I'm not crazy about mixing the two syntaxes in one set of scripts. So the question is, what is the best way to hold a block of code and variables that are a singleton?

Bryan Oakley 2007-09-19: There is no definitive "best way". I use namespaces all the time. I even use them for "objects" that aren't singletons. I've done it so many times I can (almost, but I keep trying!) do it with my eyes closed. Namespaces are much more portable than objects since, as of this writing, they are part of the core and OO is not. If portability is not a concern and you already use an object system, it makes sense to continue to use an object system.

PYK 2014-06-23: namespace isn't obsolete, just lower-level.

Private namespaces

Anonymous: There should be a facility in Tcl to define a private namespace. And, in a completely different direction, Darren New points out that tcllib would be a good home for a namespace serializer, that is, a proc which writes out a namespace's content in a form that can be read back in to recreate the namespace exactly.

Stu 2007-09-30: Creating a randomly named namespace with global vars cleanup. Left-to-right evaluation and the fact that for and unset return empty strings are what makes it possible.

# Random namespace (10 chars, A-Z) with vars cleanup
namespace eval [
    for {set ns {};set i 0} {$i < 10} {incr i} {
        append ns [format %c [expr {65 + int(rand() * 26)}]]
    }
][unset i][set ns][unset ns] {
        puts [namespace current]

        # Other stuff: variables, procs, etc.
}

Lars H 2007-10-02: Edited the above to make it more readable. Technically it's sufficient that unset returns an empty string, as the first three command substitutions in the namespace argument of namespace eval could be combined into one. Realistically, the for loop should be in a procedure instead to avoid using global variables at all...

Stu 2008-11-16: Lars, could you please date your comments? The one you added below in the middle of the paragraph arrived a year after the original discussion, when apply did not even exist. So yes, now with Tcl 8.5, it can be done with apply ... I guess ... you didn't provide an example.

Stu 2007-10-02: But where would you put the command? The idea here is to pollute the global namespace as little as possible and to create a unique namespace so that we're not "stepping on anyone's toes." (Lars H: Well, use apply then.) If one writes namespace-agnostic code then one should be able to source their code into unknown environments without problems. Even with a proc, we still pollute the global namespace - one proc versus two variables seems like a step in the right direction.

This is a bit better (I like it in one line but this is easier to read/understand):

# Random namespace (10 chars, A-Z) with cleanup, using [proc] with empty proc name.
namespace eval [
    proc {} {} {
        set ns {}
        for {set i 0} {$i < 10} {incr i} {
            append ns [format %c [expr {65 + int(rand() * 26)}]]
        }
        return $ns
    }
][{}][rename {} {}] {
        puts [namespace current]
        # Other stuff: variables, procs, etc.
}

Using eval, I can eliminate all procs and vars and it's much smaller more readable:

# Random namespace (10 chars, A-Z) with cleanup.
namespace eval [eval return [string repeat {[format %c [expr {65 + int(rand() * 26)}]]} 10]] {     
    puts [namespace current]
    # Other stuff: variables, procs, etc.
}

This gives an interesting result which I don't understand just yet. Put the above (with eval) code in a file: namespacetest.tcl

$ tclsh namespacetest.tcl                                                                        
::HTUZNJTKOE

$ tclsh
% source namespacetest.tcl
NIDZGBNLJP

Where is the :: ?

DKF: Seems to me like the return might be confusing things, causing the evaluation to stop before the namespace eval's body fires. In 8.5 use return -level 0 instead.

Stu 2007-10-02 Ok then, (and even better) without eval:

namespace eval [subst [string repeat {[format %c [expr {65 + int(rand() * 26)}]]} 10]] {
    puts [namespace current]

    # Other stuff: variables, procs, etc.
}

Stu 2008-10-31 Ultimate random namespace, like ::TkTwig!(20081031023934)1500<omzpnyxzif>

#! /bin/sh
# \
exec tclsh "$0" ${1+"$@"}

package require Tcl; namespace eval \
MyProg!([clock format [clock seconds] -format %Y%m%d%H%M%S])[pid]<[subst \
[string repeat {[format %c [expr {97 + int(rand() * 26)}]]} 10]]> {


# program goes here


}; #End of namespace
 
# EOF

PYK 2014-06-23: I usually just use info cmdcount

namespace eval [info cmdcount] {
    ...
}

And for good measure, here's an example using apply:

namespace eval [apply [list {} {
    while {[namespace exists [set ns [info cmdcount]]]} {}
    return $ns
} [namespace current]]] {
    #script goes here
}

An Anonymous Critique

This section was transplanted here from Tcl Warts.

anonymous 2006-03-10: The concept is good, but the execution is flawed, especially if you're doing anything object-oriented. Which is kind of ironic, given that (AFAIK) namespace were created to support OO, specifically incr Tcl. The implementation has the fingerprints of C++ all over the syntax. Tcl is many things, but C++ is not one of them. The two don't fit well together.

A problem that I keep running into is that I don't typically use the global scope operator :: for objects at global scope, so there's usually a string inequality between what I'm typing and the value of $self, $this, self, whatever the OO system is using. And if you try to compose object names by combining object names, you get spurious ::'s lying around in the middle of names, which makes Tcl very unhappy.

DKF 2015-06-11: Namespaces are not objects. They form the foundation for object systems, but are not objects (because you need commands as well, and in fact quite a lot of other stuff too). What's more, the names of objects generally should always be fully-qualified command names; anything else is usually just an abbreviated form. (The exception are widget names, which are unqualified but always start with a . character.)

PYK 2020-01-24: The flavor of object orientation offered by object systems is different than that offered by C++. The C++ version is wrong for Tcl because Tcl is a more dynamic language, where simple namespace like those implemented in Tcl are a better fit. The most common way to pass a handle to an object to a routine is to pass the name of the routine representing the command as its first argument. This name should be fully-qualified. the "unknown" mechanismes in both namespace and namespace ensemble can help with this, as can [namespace which and namespace origin.


anonymous 2017-01-18: You may want to look at this [L1 ] implementation which was done in Tcl 8.0 syntax back in 2001 for an example of a pure script based OOP implementation which deals with namespace separators in order to allow proper shadowing of members in child classes. It hasn't been actively developed since 2001 but is public domain if you find it or studying it useful.


PYK 2017-01-22: C++ is one the first things the phrase, "flawed concept" evokes in my mind. In contrast, Tcl namespaces are minimal and effective. Namespace separators don't appear in my scripts very much at all anymore, having been supplanted somewhere along the way by things like namespace which, namespace eval, and namespace ensembles. Perhaps some of these things didn't exist when the critique was written.

History

TCLCORE Re: incr Tcl in the core? , tcl-core mailing list, 2000-11-09
KBK recounts the panel discussion at the Third Annual Tcl/Tk Workshop that led to the inclusion of namespace in the core