Expected TCL: upvar vs namespace variable performance

Question

Is there an expected, by spec/implementation, difference between accessing namespace variables vs. upvar. I have to use a call-back function. I cannot just pass an argument.Empirically, upvar wins. But is that expected, in all reasonable case? Thanks.

Answer 1

Yes definitely. fully scoped reference is faster than upvar reference which is faster than variable reference.

To find-out, the command 'time' is your friend:

namespace eval toto {
    proc cb_upvar {varname} {
        upvar $varname var
        incr var
    }

    proc cb_scoped {varname} {
        incr $varname
    }

    proc cb_variable {varname} {
        variable $varname
        incr $varname
    }
}

proc benchmark {cmd} {
    set toto::totovar 1
    time $cmd 100
    puts -nonewline "[lindex $cmd 0] =>\t"
    puts [time $cmd 20000000]
}

puts [info tclversion]
benchmark {toto::cb_scoped ::toto::totovar}
benchmark {toto::cb_variable totovar}
benchmark {toto::cb_upvar totovar}

Output:

toto::cb_scoped =>    0.47478505 microseconds per iteration
toto::cb_variable =>  0.7644891 microseconds per iteration
toto::cb_upvar =>     0.6046395 microseconds per iteration

Rem: the huge number of iterations is required to get consistent result.