TIP 284: New 'invoke' and 'namespace invoke' Commands

Login
Bounty program for improvements to Tcl and certain Tcl packages.
Author:         Miguel Sofer <msofer@users.sourceforge.net>
State:          Draft
Type:           Project
Vote:           Pending
Created:        01-Oct-2006
Post-History:   
Tcl-Version:    8.7

Abstract

This TIP exposes a Tcl script-level interface to the direct command invokation engine already present in the Tcl library for years.

Proposed New Commands

This TIP proposes a new subcommand to namespace, invoke, with the following syntax:

namespace invoke namespace cmd ?arg ...?

This invokes the command called cmd in the caller's scope, as resolved from namespace namespace, with the supplied arguments. If namespace does not exist, the command returns an error.

This TIP also proposes a new command:

invoke level cmd ?arg ...?

This invokes the command cmd in level level with the supplied arguments. The syntax of level is the same as that required by uplevel.

Rationale

There is currently no script-level equivalent to Tcl_EvalObjv(), though the functionality is provided by one of:

   eval [list cmd ?arg ...?]
   {*}[list cmd ?arg ...?]

Note that the core tries to optimise the first case, but has to be careful to only avoid reparsing when it is guaranteed safe to do so. The notation is rather clumsy too.

The proposed new commands try to improve this situation, with the added functionality of determining the namespace in which the command name is to be resolved (functionality which was very difficult to use previously using the script-level API). In this manner it is possible for the invocation to make good use of namespace path and namespace unknown features.

The new command invoke could be implemented as:

   proc invoke {level args} {
       if {[llength $args] == 0} {
           return -code error SomeMessage
       }
       if {[string is integer $level] && ($level >= 0)} {
           incr level
       }
       uplevel $level $args
   }]

Reference Implementation and Documentation

[RFE 1577324] (which depends on [Patch 1577278]) provides an implementation of namespace invoke. https://sourceforge.net/support/tracker.php?aid=1577324 https://sourceforge.net/support/tracker.php?aid=1577278

Differences to Other Commands

  1. Both these commands perform command invocation, as opposed to the script evaluation done by eval, uplevel, namespace eval and namespace inscope

  2. namespace inscope does a magic expansion of the first argument, namespace invoke takes the first argument as a command name. In other words, namespace inscope can be used with a command prefix. Feedback on the semantics suggest that this is a worthy feature, very useful for packing up command prefixes. This tip may yet be revised or withdrawn to take that into consideration.

  3. Both namespace eval and namespace inscope add a call frame,namespace invoke does not - it invokes in the caller's frame.

Sample Usage

In tcllib's ::math::calculus::romberg we see (edited for brevity):

    # Replace f with a context-independent version

    set fqname [uplevel 1 [list namespace which [lindex $f 0]]]
    set f [lreplace $f 0 0 $fqname]
    ...
    set cmd $f
    lappend cmd [expr {0.5 * ($a + $b)}]
    set v [eval $cmd]

where the command name in the prefix f is replaced with its fully qualified name. A further variable is lappended, and the result is sent to eval.

With namespace invoke and invoke this would be coded as:

    set ns [invoke 1 namespace current]
    set f [list namespace invoke $ns {*}$f]
    ...
    set v [{*}$f [expr {0.5 * ($a + $b)}]]

If both new commands took the cmd argument as a prefix to be expanded (as suggested by some early commenters), it would be even nicer for this usage:

    set ns [invoke 1 [list namespace current]]
    set f [list namespace invoke $ns $f]
    ...
    set v [invoke 0 $f [expr {0.5 * ($a + $b)}]]

The same thing (with slightly different semantics and a performance cost due to script evaluation instead of command invocation) could be obtained today by doing:

    set f [uplevel 1 [list namespace code [list uplevel 1 $f]]]
    ...
    set v [eval [lappend f [expr {0.5 * ($a + $b)}]]]

In order to fully reproduce the semantics and performance it would have to be:

    set f [uplevel 1 [list namespace code [list uplevel 1 $f]]]
    ...
    lset f end end [linsert [lindex $f end end] end [expr {0.5 * ($a + $b)}]]  
    set v [eval $f]

Copyright

This document has been placed in the public domain.

History