Should you see the error message

attempt to provide package TclOO 0.1.1 failed: package TclOO 0.1 provided
instead

Duoas If I should see that message, then what?

DKF: Then stop using that old version! I really don't recommend using less than 1.0 nowadays…

TclOO::ext

I collected some TclOO extension in a package here: https://github.com/arthurschreiber/tcloo-ext

Currently included is a Reference Counting system inspired by ObjectiveC/Cocoa to make memory management easier and a class method inheritance through MetaClasses (not using mixins).

Discussion

APN I don't quite follow the discussion below about automatically bringing variables into scope. Isn't that what the variable inside a class definition does?

class create C { variable a b }

would have variables a and b automatically brought into scope in all methods. Perhaps the discussion originated before variable was implemented. If someone can confirm this, please delete this and the discussion below because it is misleading.

DKF: Indeed, that's what the variable declaration does. Now. It didn't when the discussion was held. It also doesn't bring in all variables because that turns out to be more problematic when mixed with inheritance.

jcw 2007-06-04:

Is there some nice idiom for the following?

my variable {*}[info vars [namespace current]::*]

IOW, set up methods to automatically have access to all object state as plain variables (as in Itcl). The above line would be added in front of all method definitions, using a small wrapper (am generating methods dynamically anyway).

This assumes that all variables are defined in the constructor.

escargo Does this use of variable contrast confusingly with the Tcl variable where something like that would assign the names of half the variables as the values of the other half of the variables?

jcw - Well, I'm just trying it out, so don't shoot me ;) - although so far I tend prefer this over the "variable" cmd.

escargo - I was just asking a question. It's just that having my variable work not like variable might be considered confusing.

jcw - Here's an example with some output which may help those who don't have TclOO yet:

oo::class create dog {
    method a {} {
        puts a1-[namespace current]
        puts a2-[namespace path]
        foreach x [namespace path] {
          puts a3-$x-[info commands ${x}::*]
        }
        puts a4-[info vars [namespace current]::*]
        my variable e
        set e f
        puts a5-[info vars [namespace current]::*]
    }
    self method b {} {
        puts b1-[namespace current]
        puts b2-[namespace path]
        foreach x [namespace path] {
            puts b3-$x-[info commands ${x}::*]
        }
        puts b4-[info vars [namespace current]::*]
        my variable e
        set e f
        puts b5-[info vars [namespace current]::*]
    }
}

dog create fifi
fifi a
dog b

# Output:
#   a1-::oo::Obj4
#   a2-::oo::Helpers
#   a3-::oo::Helpers-::oo::Helpers::self ::oo::Helpers::next
#   a4-
#   a5-::oo::Obj4::e
#   b1-::oo::Obj3
#   b2-::oo::Helpers ::oo
#   b3-::oo::Helpers-::oo::Helpers::self ::oo::Helpers::next
#   b3-::oo-::oo::InfoClass ::oo::class ::oo::InfoObject ::oo::object \
#                                                     ::oo::copy ::oo::define
#   b4-
#   b5-::oo::Obj3::e

MHo 2015-02-24 This doesn't work for me. There's no command 'dog b'. Should it be 'self method b…' instead of 'self.method b…' ? Would be fine if there where a 'TclOO for OO-Dummies' somewhere....

DKF: Yes, it should. The self.blah syntax was an interim thing that got removed once I had the more general self syntax.

As you can see, each class and each instance gets its own namespace. Note also that class Dog ended up as namespace Obj3, even though it was constructed with a specific name (which is not necessarily globally unique, though).

DKF: I've been experimenting for a while now with OO systems, and a my variable that brings every variable in the object into scope is probably not what you want as it causes great problems in subclasses (i.e. clashes between subclasses' variables and local variables). Instead, I think that being able to pick up all the variables defined within a particular class would be far more useful. OTOH, it's not at all trivial to implement (I think I expose enough API that it can be done through the use of class metadata, but I'm not certain) so I'm not in a hurry to change things.

It was a deliberate decision to make my variable different from variable. The latter has syntax that is almost never what anyone wants, and so is a priori suboptimal. Hence I went for something that was more likely to be what you need. :-)

escargo: Well, I wondered if you did it on purpose, and you did. And I see it was for what you think is a good reason.

Also, as noted, the internal namespaces have names that are not necessarily those of the created object, and their names are deliberately not documented (i.e. they could change between point releases with no warning). Use the introspection facilities to go from one to the other.

APW 2007-06-05:

@ jcw: I am working on an extended version of dkf's code suitable for Itcl, which would have solutions for some of the problems you have mentioned above. For example I am using apply and namespace upvar and namespace unknown to generate Itcl methods, which have access to all the class variables in the class hierarchy (if they are not private) etc. Using only the name of a variable as in Itcl is possible, same for call of Itcl methods without the "my" in front. If you are interested please contact me directly, the implementation is not yet ready for real testing.

@dkf: If I know what I really need additionally for Itcl I will contact you and we can try to perhaps get a merged version. At the moment all my modifications are additional functionality so that the original functionality should work as before with also running all your tests without problems.

Aspects with TclOO

DKF: The following example is ripped from the documentation of next:

oo::class create cache {
    filter Memoize
    method Memoize args {
        # Do not filter the core method implementations
        if {[lindex [self target] 0] eq "::oo::object"} {
            return [next {*}$args]
        }

        # Check if the value is already in the cache
        my variable ValueCache
        set key [self target],$args
        if {[info exist ValueCache($key)]} {
            return $ValueCache($key)
        }

        # Compute value, insert into cache, and return it
        return [set ValueCache($key) [next {*}$args]]
    }
    method flushCache {} {
        my variable ValueCache
        unset ValueCache
        # Skip the cacheing
        return -level 2 ""
    }
}

oo::object create demo
oo::objdefine demo {
    mixin cache
    method compute {a b c} {
        after 3000 ;# Simulate deep thought
        return [expr {$a + $b * $c}]
    }
    method compute2 {a b c} {
        after 3000 ;# Simulate deep thought
        return [expr {$a * $b + $c}]
    }
}

puts [demo compute  1 2 3]      ? prints "7" after delay
puts [demo compute2 4 5 6]      ? prints "26" after delay
puts [demo compute  1 2 3]      ? prints "7" instantly
puts [demo compute2 4 5 6]      ? prints "26" instantly
puts [demo compute  4 5 6]      ? prints "34" after delay
puts [demo compute  4 5 6]      ? prints "34" instantly
puts [demo compute  1 2 3]      ? prints "7" instantly
demo flushCache
puts [demo compute  1 2 3]      ? prints "7" after delay

jcw: ''Way cool, great separation of functionality. Next step: a Metakit backed cache? ;)

DKF: I've thought more about this, and have written an Aspect Support Class for TclOO.

APW 2007-06-06: For discussion of Itcl related topics for a new implementation based on this TIP see tclOO missing features for Itcl

Performance Measurement

DKF 2007-09-20: As seen in my Tck2k7 paper, TclOO is really quite fast. Here's those performance figures (correct with 8.5b1 on my ancient laptop).

Notes: It turns out that stooop non-virtual methods are much faster, but they're not comparable with the other methods as they do not allow overriding of the method by subclasses, so I used the (comparable) virtual methods instead, which are much slower.

DKF: Reanalysed with modern hardware and new versions of XOTcl, TclOO and snit (~3 sig. figures) - this machine doesn't seem to have stooop:

All are packages built by ActiveState, running on Tcl 8.5.2. (I see differences in my own tests, but can't tell what their origin is.)

Analyzing those figures, we see (from the itcl baseline) that the machine is about 4.25 times faster (average). We therefore can see that TclOO is getting a (little) better at method dispatch though it is slower at object construction. We also see that XOTcl has improved a lot on the construction front, though it is still not very fast at method dispatch. Looking at snit, we see that it is now very fast at method dispatch (nearly as fast as TclOO) but is still very slow to create objects.

Sarnold: Indeed snit is very slow at object instantiation, but building a snit-like system on top of XOtcl, namely xoins, give it far better creation times at the cost that method dispatch is slower. I do not know if William Duquette is still working on a TclOO snit implementation - I hope it will succeed, because snit is really good for Megawidgets and has a nice syntax that fits well upon Tcl's one.

DKF: I'm working on two sets of (independent) improvements to TclOO that should make doing snit on top of it easier. Firstly, I'm doing something to make variables easier to use, so you can declare them once per class instead of once per method. Secondly, I'm trying to make submethods work (one of the neatest features of Snit). Apart from these, I suspect that the current method forwarding implementation is over-simplistic in its model; need someone to point me at a use-case case or two in order to figure out how to improve on that.

The script used to create the performance data is below. (Well, I actually did it interactively over a few runs, so I don't warrant that this script will really work...)

Performance Analysis Script

package require Tcl 8.5b1

# Compute Calls Per Second of a script
# Note that this script is self-tuning; it prefers to execute a script for around a second

proc cps {script} {
    # Eat the script compilation costs
    uplevel 1 [list time $script]
    # Have a guess at how many iterations to run for around a second
    set s [uplevel 1 [list time $script 5]]
    set iters [expr {round(1/([lindex $s 0]/1e6))}]
    if {$iters < 50} {
        puts "WARNING: number of iterations low"
    }
    # The main timing run
    set s [uplevel 1 [list time $script $iters]]
    set cps [expr {round(1/([lindex $s 0]/1e6))}]
    puts "$cps calls per second of: $script"
}

#--------------------------------------------------------------------------
puts "using Itcl..."
package require Itcl 3.4
namespace path itcl

class foo {
    variable x
    constructor {} {
        set x 1
    }
    method bar {} {
        set x [expr {!$x}]
    }
}

foo f
cps {f bar}
delete object f

cps {delete object [foo f]}

delete class foo

#--------------------------------------------------------------------------
puts "using XOTcl..."
package require XOTcl 1.5.5
namespace path xotcl

Class create Foo
Foo parameter x
Foo instproc init {} {
    my set x 1
}
Foo instproc bar {} {
    my instvar x
    set x [expr {!$x}]
}

Foo create f
cps {f bar}
f destroy

cps {[Foo create f] destroy}

Foo destroy

#--------------------------------------------------------------------------
puts "using TclOO..."
package require TclOO 0.1
namespace path oo

class create foo {
    constructor {} {
        variable x 1
    }
    method bar {} {
        variable x
        set x [expr {!$x}]
    }
}

foo create f
cps {f bar}
f destroy

cps {[foo create f] destroy}

foo destroy

#--------------------------------------------------------------------------
puts "using snit..."
package require snit 2.1

snit::type foo {
    variable x
    constructor {} {
        set x 1
    }
    method foo {} {
        set x [expr {!$x}]
    }
}

foo f
cps {f bar}
f destroy

cps {[foo f] destroy}

foo destroy

#--------------------------------------------------------------------------
puts "using stooop..."
# Must go last because it plays games with proc which might disturb the
# performance of other OO systems. Note that stooop has both virtual and
# non-virtual methods, with very different performance profiles. The virtual
# ones are much more comparable in capability to other OO systems...
package require stooop 4.4.1
namespace path stooop

class foo {
    proc foo {this} {
        set ($this,x) 1
    }
    proc ~foo {this} {}
    virtual proc bar {this} {
        set ($this,x) [expr {!$($this,x)}]
    }
    proc bar-nv {this} {
        set ($this,x) [expr {!$($this,x)}]
    }
}

set f [new foo]
cps {$f bar}
cps {$f bar-nv}
delete $f

cps {delete [new foo]}

Simple examples of design patterns using TclOO