Richard Suchenwirth 2002-05-05 - Reading chapter 1.3 of SICP, a highly educational introduction to programming based on the LISP dialect Scheme, I felt challenged to try to reproduce the Scheme examples for summation of a series in Tcl, for example:
b --- \ / f(n) = f(a) + ... + f(b) --- n=a
Here's what I have so far (both f and next are "functional objects", which in Tcl just means "strings that happen to be the name of a proc":
proc sum {f a next b} {
expr {$a>$b? 0 : [$f $a] + [sum $f [$next $a] $next $b]}
}
# --------------------------- small building blocks:
proc cube x {expr {$x*$x*$x}}
proc inc x {incr x} ;# argument x is value, instead of name
proc lambda {argl body} {
set name [info level 0]
proc $name $argl $body
set name
}if 0 {This handful of code allows us to reproduce the Scheme results from SICP - for more info, see there:
sum cube 1 inc 10 ==> 3025, or:
sum cube 1 [lambda x {incr x}] 10 proc identity x {set x}
sum identity 1 inc 10 ==> 55; or:
sum [lambda x {set x}] 1 [lambda x {incr x}] 10 proc pi-term x {expr {1.0 / ($x * ($x+2))}}
proc pi-next x {expr {$x + 4}}
expr {[sum pi-term 1 pi-next 1000]*8} ==> 3.1395926555897828whose run limit could be increased from 1000 until 2756 before raising the "too many nested calls..." error ;-( and still gave a less precise approximation than the good old atan(1)*4...
proc integral0 {f a b dx} {
set ::globaldx $dx
expr {[sum $f [expr {$a+$::globaldx/2}] add-dx $b] * $dx}
}
proc add-dx x {expr {$x+$::globaldx}}
% integral0 cube 0 1 .0016 ==> 0.2499996800000055Here however I had to start to compromise: instead of Scheme's lexical scoping, where dx is visible everywhere inside integral's body, including the add-dx function, I had to elevate dx to global status, which is ugly; and Tcl's recursion depth limit caught me before I could try SICP's dx value of 0.001 - the result is still close (but no cigar) to the correct result of 0.25. Oh wait, at least in this case we can emulate lexical scoping more closely, by embodying $dx into a "live proc body" of add-dx:
proc integral {f a b dx} {
proc add-dx x "expr {\$x+$dx}"
expr {[sum $f [expr {$a+$dx/2}] add-dx $b] * $dx}
}
% integral cube 0 1 .00146 ==> 0.25009974849771255A cleaner way to implement this "closure" would be an added default argument, like they do in Python - the body can remain braced, but the argument list of add-dx now "inherits" from outside:
proc integral {f a b dx} {
proc add-dx "x {dx $dx}" {expr {$x+$dx}}
expr {[sum $f [expr {$a+$dx/2}] add-dx $b] * $dx}
}Slightly off-topic, but as all building blocks are there, here's a stint on the derivative of a function. However, we've reached a certain bound of convolution and ugliness here, escaping and bracing like below (it produces the approximately correct result, but still...):
proc deriv g {
lambda x "expr {(\[{$g} \[expr {\$x+$::dx}]]-\[{$g} \$x])/$::dx}"
}
% set dx 0.00001 ;# well, in SICP they have it global too...
% [deriv [lambda x {expr $x*$x*$x}]] 5 ==> 75.0001499966Anyway, I'm again surprised how many steps towards functional programming are possible with Tcl .. and more to come.