return , a built-in Tcl command, returns a value, a code and other options from a particular level.


return ?result
return ?-code code? ?result
return ?option value? ?result


-errorcode list
-errorinfo info
-errorstack list
New in Tcl8.6.
-level level
-options options


official reference
TIP 90 , Enable return -code in Control Structure Procs


Delivers a return code for some level to the interpreter, which then takes some action based on the received code. By far the most common use case is to return a value from the current procedure:

proc knightswhosay {} {
    return Ni!

-level is a positive integer indicating the number of levels up from the current level. Level 0 is the evaluation of return itself, level 1 is the caller of return, and so on.

-code is a positive integer or the symbolic value as specified later on this page. By default, the value of -code is 1 (ok) and the value of -level is 1. The following three commands all effectively instruct the caller to return normally from the current procedure (level 1):

return -level 0 -code return
return -level 1 -code ok

The return value is either the final argument or the empty string. catch and try can be used to intercept a return.

Any additional option/value pairs are added to the options dictionary for the level.

[TODO: Document Tcl 8.5's extended handling]

Return Codes

-code may be any of the following values:

Success. return -level 1 -code ok is the same as return. This is the default. (TCL_OK)
An error. return -level 0 -code error is the same as error. (TCL_ERROR)
Cause the level up to return. return -level 0 -code return has the same effect as return, as well as return -level 1 -code ok. (TCL_RETURN).
Instruct the next level up break out of the innermost nested loop in the current script. return -level 0 -code break is the same as break. (TCL_BREAK)
Instruct the next level up to terminate the current iteration of the innermost nested loop in the code that invoked the current procedure and start the next iteration. (TCL_CONTINUE)
Any integer. User-defined behaviour. Rarely used with values outside the range 0..4.

-code is rarely used, as commands such as error, break and continue handle the common cases. However, a routine that implements a new control structure that acts like break, continue, or error might use return -level 1 .... Another use case for -code is to have the caller, rather than the current procedure, report that the wrong number of arguments were provided, since this makes the error message more clear. In this case, use return -level 1 -code error.

-errorinfo specifies an initial stack trace for $errorInfo. If it is not specified, the stack trace left in $errorInfo includes the call to the current routine and higher levels on the stack but does not include any information about the context of the error within the procedure. Typically the info is derived from the value left in $errorInfo when catch traps an error within the procedure.

If -errorcode is specified, list provides a value for $errorCode. Otherwise, $errorCode defaults default to NONE. (from: Tcl Help)

Hack: Arbitrary Data after return

Text occurring in a script after return is never evaluated, and thus doesn't even have to be valid Tcl:

proc foo {} {
    puts Foo
    This is not Tcl - code after the return is never evaluated so 
    may be  used for commenting...
} ;# RS

DGP: In Tcl 7 and in recent enough Tcl 8.5 that is correct. In the releases in between, due to some limitations in the bytecode compiler/execution machinery it could not be just any text:

  • braces still needed to be balanced
  • Commands like set get byte-compiled early, so a syntax error is found if a line in that post-return comment starts with set and has more than two other words.

Joe English also responds that not just any arbitrary text can appear after return. proc interprets its third argument as a script. It's therefore unwise, and one could argue even illegal, to pass in something that's not at least syntactically valid as a script, even if you know that parts of it will never be executed. By way of analogy: lindex interprets its first argument as a list, so you'd better only pass it valid lists. In Tcl 7.6 and earlier you could actually get away with things like

lindex "a b c {bad{list" 1

as long as the examined part of the list was syntactically valid. However, this was more of an accidental artifact of implementation details than anything guaranteed by the language, and in fact this raises an error in more recent Tcl versions. Similarly, if a command expects a script, you'd better pass it a script.

PYK 2013-12-10: However, if lindex is missing its second value, the first value can still be any value, even one that isn't a valid list.

AMG: There's a reason for this. When [lindex] is given only one argument, it interprets that as an instruction to not perform any list indexing. No list indexing means no requirement to be a list. It's that simple.

jenglish's statement is correct, though it's more philosophical than practical. If [return] is redefined, code following the [return] could certainly come into play. Or perhaps the code isn't being executed but rather is being analyzed by Nagelfar, which will surely take issue with the invalidity of the code. One real possibility which would upset Tcl in any event is if the text following [return] contains mismatched braces.

And last, I really ought to mention that the first time Tcl runs the proc, it tries to bytecode the whole thing, which means wasting time analyzing garbage. If there's any doubt about the finality of the [return] (e.g. a conditional is in play), the compiled proc will contain code to return any errors found in parsing.

proc moo x {
    if {$x} {return 5}
% tcl::unsupported::disassemble proc moo
ByteCode 0x0x913eb0, refCt 1, epoch 15, interp 0x0x8c6ac0 (epoch 15)
  Source "\nif {$x} {return 5}\n\"invalid\"a"...
  Cmds 2, src 34, inst 26, litObjs 4, aux 0, stkDepth 2, code/src 0.00
  Proc 0x0x95b5e0, refCt 1, args 1, compiled locals 1
      slot 0, scalar, arg, "x"
  Commands 2:
      1: pc 0-11, src 1-18        2: pc 4-6, src 10-17
  Command 1: "if {$x} {return 5}"...
    (0) loadScalar1 %v0         # var "x"
    (2) jumpFalse1 +7   # pc 9
  Command 2: "return 5"...
    (4) push1 0         # "5"
    (6) done
    (7) nop
    (8) nop
    (9) push1 1         # ""
    (11) pop
    (12) push1 2        # "extra characters after close-quote"
    (14) push1 3        # "-code 1 -level 0 -errorcode NONE -errori"...
    (16) syntax +1 0
    (25) done

return from source

The fact that return also terminates source can be used for loading array contents without specifying an array name. Let the file t.tcl contain:

return {
    one 1
    two 2
    three 3

Then you can write it like this:

array set myArrayName [source t.tcl] ;# RS

wdb: This works, but being a purist, I prefer this text in the file to source:

list one 1 two 2 three 3

RS 2006-06-23: sure. Just if you have hundreds and thousands of array elements, with list you'd have to backslash-escape the newlines, while with bracing they need not.

DKF: I'm of the kind of purist which doesn't insist on using list to indicate listishness; that's an illusion and the list-compiler knows it.

Use in pkgIndex.tcl

See package index script interface guidelines for another use of return in sourced scripts: The main use for return outside procedures is in pkgIndex.tcl:

if {![package vsatisfies [package provide Tcl] 8.4]} return

which avoids presenting the package to interps that cannot use it.

return -code error

RS 2005-08-08: Using return -code error in place of plain error, you get a leaner error traceback which is possibly better to read:

% proc 1 x {if {$x<=0} {error              "too small"}}

% proc 2 x {if {$x<=0} {return -code error "too small"}}

% 1 0
too small
% set errorInfo
too small
   while executing
"error "too small""
   (procedure "1" line 1)
   invoked from within
"1 0"

% 2 0
too small
% set errorInfo
too small
   while executing
"2 0"

DKF: I find this is useful when distinguishing between problems with the values passed in by the caller (such as if the code really wants a string of length 37, has documented this, and yet the user has given a string of length 28), and problems internal to the code that it isn't reasonable for the caller to try to get right.

return to the Current Level

AMG PYK: return -level 0 $x simply sets the interpreter result to $x. This is the canonical identity function; see the linked page for more information.

Cancel an Error

HaO: When a return code should be forwarded to the caller, one could remove the level 0 to not directly trigger an eventual exception here:

if {[catch {script goes here} err options]} {
    dict unset options -level
    return -options $options

This was useful to me in the context of Tk bind scripts, which return break if no further bind scripts should process.

PYK 2019-08-08: What the example above is doing is for practical purposes a no-op. To cancel an error, just catch it and be done with it. In order to not hide real errors, catch only the break condition:

try {
    some script
} on break {bres bopts} {
    do break struff

The return equivalent is:

catch {
    some script
} cres copts
dict with copts {}
if {${-code} == 3} {
    do break stuff
} else {
    return -options $opts $res

return vs Falling off the End

PYK 2016-09-15: Historically, return at the end of a procedure was slightly more performant than a final command that didn't explicitly return. This isn't the case in modern versions of Tcl, where procedures are byte-compiled.

Related Commands

includes a TclChat discussion on the future of return

tailcall is also related to return in that it terminates execution of the current proc. However, unlike return, tailcall's continuation is not the caller. yieldTo is also related to tailcall in that it has a custom continuation.

See Also

Errors management
namespace eval
Funky Tcl extensibility
tricks to play with return -code return and error on return -code error
try ... finally ...
KBK 2001-01-02: how to use return -code to implement a new control structure. Lars H: Other pages which do that kind of thing are breakeval (using -code 10) and returneval (using -code -1).