Updated 2018-09-28 13:24:33 by PeterLewerin

set, a built-in command, reads and writes variables.

Synopsis  edit

set varName ?value?

Documentation  edit

official reference

See Also  edit

Can be used as a variant of set that returns an error if the variable doesn't already exist.
often used to set multiple variables in one stroke

Description  edit

set returns the value of the variable named varName, or if value is given, stores that value to the named variable, first creating the variable if it doesn't already exist. When creating a variable, set resolves the name relative to the current namespace.

If varName is not fully qualified, set searches first for a variable having that name in the current namespace, and then in the global namespace. This may lead to inadvertent modification of a variable in the global namespace. To ensure that this doesn't happen, use variable to first declare a variable in a namespace before using set to modify that variable.

If varName contains an open parenthesis and ends with a close parenthesis, then it refers to a variable in an array: The characters the first open parenthesis are the name of the array, and the characters between the first parenthesis and the parenthesis at the end of the word are the name of the variable in the array.

If no procedure is active, then varName refers to a namespace variable, which may be in the global namespace. If a procedure is active, then varName refers to a parameter or local variable of the procedure. Commands such as global, variable, or upvar, can be used to link other variables into the local scope of the procedure.

set can entirely replace variable substitution.

Basic Examples  edit

set greeting hello
set greeting ;# ->hello
set person(name) bob
set person(name) ;#-> bob
set (name) bob ;# the is an array variable, where the array name is the empty string
set (name) ;#-> bob
set {} hello 
set {} ;#->hello

Gotcha: Variable Resolution  edit

See Also, Dangers of creative writing

ulis 2003-11-16:

Try this:
set ::version 1.0
namespace eval ns {
    set version 0.9
puts $::version
catch {puts $ns::version} msg
puts $msg

can't read "ns::version": no such variable


As stated in the Tcl manual: if the name does not start with a :: (i.e., is relative), Tcl follows a fixed rule for looking it up: Command and variable names are always resolved by looking first in the current namespace, and then in the global namespace.

In the above script the variable version wasn't defined inside the namespace so Tcl used the existing global variable.

To avoid that, always declare namespace variables with the variable command:
set ::version 1.0
namespace eval ns {
    variable version 0.9
puts $::version
catch { puts $ns::version } msg
puts $msg

New result:

I (ulis) think that it would be better if the search in the global space was used when refering a variable and avoided when setting a variable.

Setting Multiple Variables at Once  edit

MSW: For those who dislike doing multiple assignments at once with foreach in the style
foreach {a b c} {1 2 3} {}

and who don't want to use lassign, here is a multiple argument set (with help from RS):
if {[info procs tcl::set]=={}} then {rename set tcl::set}
proc set {args} {
    switch [llength $args] {
        0 {return -code error {wrong # args: should be set varname ?newvalue? ?varname ?newvalue?? ...}}
        1 {return [uplevel [list tcl::set [lindex $args 0]]]}
        2 {return [uplevel [list tcl::set [lindex $args 0] [lindex $args 1]]]}
        default {
            uplevel [list tcl::set [lindex $args 0] [lindex $args 1]]
            return [uplevel [list set [lrange $args 2 end]]]

Use like this
% set a 1 b 2 c 3
=> 3
% set d 15 e [expr int(100*rand())] c
=> 3
% list $a $b $c $d
=> 1 2 3 15

Duoas: The foreach..break idiom is so prevalent in Tcl, and so common, that experienced Tcler's automatically recognize it as a set replacement idiom:
set ls [list 1 2 3]
foreach {var1 var2 ...} $ls break

However, something about it has always bothered me: I just dislike programming to the side-effects. I've submitted TIP #58 to extend set such that it can assign to multiple variables, but not as above, where the values to assign are interleaved with the variable names. Usually the values come from a list and the above implementation would require zipping variable names and values together before use, then evaling or expanding. It doesn't obviate the need to use that silly foreach..break idiom.

Littered throughout my own code is the use of this simple little routine:
proc sets args {
    set names [lrange $args 0 end-1]
    set values [lindex $args end]
    uplevel 1 [list foreach $names $values break]
    return [lrange $values [llength $names] end]

And an example of use:
sets x0 y0 x1 y1 [.canvas coords my-rectangle-tag]

This is much more Tclish and intuitive. Note also that you can get what is not used for later use (foreach requires you use it now or not at all):
set ls [sets a b $ls]
# do something with $a and $b, and maybe sometime later with the rest of $ls

A more concrete example:
% set ls [sets a b {1 2 3 4 5}]
3 4 5
% puts $ls
3 4 5
% puts $b

As per my TIP submission, set is easily extended to have such functionality without slowing it down when used as per the current specification (well, except one or two processor instructions when errors occur). When used in the extended form it is faster than using foreach, which has a lot of extra stuff to handle multiple, concurrent lists.

MJ: in 8.5 we have lassign, which is set with the arguments reversed. The example above then translates to:
% set ls [lassign {1 2 3 4 5} a b]
3 4 5
% puts $ls
3 4 5
% puts $b

Duoas: Me feels stupid for having missed that... I learned Tcl moving into 8.0 and I'm still a little behind in a lot of 8.5 improvements.

MJ: No need to feel stupid, Tcl 8.5 has a lot of new goodies, see Changes in Tcl/Tk 8.5.

Double Indirection  edit

See also: An Essay on Tcl Dereferencing

In some languages, notably PHP, an additional dollar sign can be added to a variable to achieve double-indirection. e.g. $$var. Tcl doesn't support such syntax, but set can be used to the same effect:
% puts [set $var] ;# This works safely

  proc $ {var {# 1}} {for {set i 1} {$i<=${#}} {upvar $var v;set var $v};uplevel return $v}   ;# multi indirection, which tcl doesn't support
% puts [$ var 2] ;# This works safely

The following example, which uses set instead of $, is equivalent:
% puts [set [set var]] ;# as does this

Similarly, to print the values of var1, var2, and var3:
set var1 3.14159
set var2 hello
set var3 13
foreach num {1 2 3} {
    puts "var$num = [set var$num]"

var1 = 3.14159
var2 = hello
var3 = 13

upvar can also provides access to to other variables, even when they are in the same scope:
set var1 hello
upvar 0 var1 var2

eval could also be used to achieve double indirection (but there are major caveats):
% set a 5
% set var a
% puts $$var              ;# This doesn't work
% eval puts $$var         ;# This does  - but it's dangerous

One caveat is that if $var has a value containing any special characters (e.g. whitespace, semicolon), they'll get interpreted, and where this is inadvertent, could result in an error or an exploit.

An Alternative to return  edit

proc returns its last evaluated result, so it's a common idiom to use set instead of return as last command.
set res
return $res

Some Tcl style guides recommend using the explicit return alternative. The rationale is is that using an explicit return guards against inadvertantly adding addtional code after the set command.

Before return got byte-compiled (i.e., before Tcl 8.4), the set idiom was faster than the return idiom for returning a variable value. This is no-longer true.

A Verbose set  edit


As Tcl has no reserved words, you can even write your own set command (make sure its effects are like the original, or most Tcl code might break). For instance, this version reports its actions on stdout:
rename set _set
proc set {var args} {
    puts [list set $var $args]
    uplevel 1 _set $var $args

This might help in finding what was going on before a crash. When sourced in a wish app, shows what's up on the Tcl side of Tk (as long as you can find the program's stdout).

Swap the Contents of Two Variables  edit

Thanks to copy-on-write and the special-case of appending an empty string, this is efficient:
set a one
set b two
set a $b[set b $a; lindex {}]

In the previous example, lindex is used as the identity function.

AMG: Can also take advantage of the fact that [list] with zero arguments returns empty string:
set a $b[set b $a; list]

Bug: array and $  edit

As of Tcl 8.6.3, set myarray($) one results in an array variable named the empty string because set interprets $, even at the end of index, as variable substitution, and removes it. Meanwhile, variable substitution handles the same syntax just fine:
% set a($) one
% puts $a($)
can't read "a($)": no such element in array
% set {a($)} two
% puts $a($)

AMG: This is extremely confusing to me. [set] has no need to interpret $, ever. If a $ survives the Tcl interpreter and manages to be part of [set]'s first or second argument, it should be treated literally. The above set {a($)} two command demonstrates this fact.

Let me attempt to reproduce using Tcl 8.6.1::
% set a($) one; array get a
{} one
% set {b($)} one; array get b
{$} one
% set c(\$) one; array get c
{$} one

So yeah, the bug is there. But what's going on? Let's ask [list] what the arguments are:
% list set a($) one
set {a($)} one
% list set {b($)} one
set {b($)} one
% list set c(\$) one
set {c($)} one

Uh oh. If [list] gives the same results regardless of quoting, it means the different quoting methods (or lack thereof) are equivalent, so [set] should not be able to discern.

Okay, another test:
% proc myset {args} {tailcall set {*}$args}
% myset d($) one; array get d
{$} one

Wow. This seems to indicate the bug is in the bytecode compiler for [set].

Yet another test, to bypass any attempts at optimization:
% interp hide {} set
% interp invokehidden {} set e($) one; array get e
{$} one

Pretty much confirms it.

PYK 2015-03-26: That was a nice writeup. This bug is fixed in 8.6.4, but I'd like to see this analysis preserved on some page about investigating the behaviour of the interpreter, maybe with some additional notes about the role of tailcall.

Why C Programmers Hate Tcl Variable Scope Resolution!  edit

[JoGusto]: Yes, that is meant to be provocative. But, it is a serious problem for software developers who have spent years, or decades, programming in languages such as C, where a global variable is just that: GLOBAL!

It is absolutely NOT intuitive to the average C programmer that all global variables disappear from scope (YES! Ridiculous!) when you enter a proc... in C, a global remains in scope, with no special syntax required, unless that global name is hidden because of a more enclosed declaration inside a block. Even then, the global is accessible (in C++ at least) via the same notation Tcl uses: the double colon.

This scope resolution problem is just really abhorrent to those of us who "Think in C" and try valiantly to script in Tcl, only to repeatedly bump up against that old bugaboo: "I forgot the double colon! My global variable is "undefined"" It is ridiculous to have to repeatedly declare variables that are global within the local scope just to access them.

Yea, I know it's way too late to change, but its a common complaint I get from many people about Tcl, and Python, that "they got the whole global variable thing WRONG..." and it's definitely something that trips up a lot of people who are not used to it.

dbohdan 2015-02-16: OTOH, having to explicitly declare your globals prevents accidental mutation of global state. This is important because in Tcl unlike in C variables are generally set without prior declaration and the compiler won't catch a redeclared variable for you.

[name redacted]: there are always differences going from programming language A to programming language B. The correct way to deal with them is to adapt, not to demand that B be reworked to more closely resemble A. It's quite too late to change the exclusive-scope model that Tcl uses, but more importantly it would be a bad idea since it works well. And if you're going to put C forward as an example of handling variables in a better way, I'll just whisper "extern..." ;)

AMG: "Global" means a variable can be reached from any stack level. That is not an abuse of the term. The difference between Tcl and many other languages is whether or not the global frame is automatically searched if the variable can't be found locally.

PYK 2015-03-11: "This thing in language X is abhorrent to those of us who think in language Y!".

Tcl design decisions often come down on the side of allowing programmers to have it their way, You want to access global variables from inside a function? No problem. You want zero pollution from any other scope? Also no problem. Or maybe you specifically want to access a certain variable two levels up the call stack. Go right ahead. If you want C-style variable scoping, write a command that gives you that, and use that command to read or write variables. Variable scoping rules are notoriously hard to get right, but fortunately, Tcl did (except for the creative writing issue).

[Jnk] - 2016-12-02 23:01:07

How can we use set like read command in bash? Eg. In BASH shell using the read command as
read -e -p "Enter your name:" -i "Bob" name

AMG: Use [gets].
puts -nonewline "Enter your name:"
flush stdout
set name [gets stdin]

There's also tclreadline which may be of service for advanced editing, but I've never tried it.