World: r3wp

i got that bounce email after my last post today too.

maybe someone subscribed to the list is bouncing? although the error 
message refers to a mail loop.

select reduce [1 none] 1 is the same problem as pick reduce [none] 
1 and so on... the problem with switch can be considered a bug, however 
switch expects a block after the value so this is debatable.

i think, that since we have other ways to check for existence (find 
for select, or length? for pick), the behavior is still acceptable 
because useful in many cases.

of course, Ladislav's /default approach would probably be better.

(but can never solve the switch problem completely, because any value 
used as default could appear in the block  - you have to use find)

right, Gabriele, but SWITCH can be implemented using FIND, which 
would solve this issue

switch expects a block after the value so this is debatable.
 - how about switch/default 1 [1 #[none]] [2] then?

sorry, forget about it, I just want to say, that SWITCH doesn't check 
the value is followed by a block

yes, i agree it's a problem, it's just not a problem in practice 
:)

i'd actually implement switch using parse. i find it very useful 
to be able to specify multiple values for the same block.

i.e. switch val [2 4 6 ['even] 1 3 5 ['odd]]

that is what Cyphre wants too

a possible implementation:

switch1: func [
    "Selects a choice and evaluates what follows it."
    [throw]
    value "Value to search for."
    cases [block!] "Block of cases to search."
    /default case [block!] "Default case if no others are found."
    /local blk
][
	value: find cases value
	if value [value: find next value block!]
    either value [do first value] [if default [do case]]
]

it is slower than SWITCH due to "double search", though

I like that behaviour. I planned to write something similar for a 
long time... :)

Mmm.. parsing numbers...

How much effort is a switch in c?

switch2: func [

    "Selects a choice and evaluates the first block that follows it."
    [throw]
    value "Value to search for."
    cases [block!] "Block of cases to search."
    /default case [block!] "Default case if no others are found."
    /local rule
][
	rule: [
		1 1 () ; <-- value
		to block! set value block! (return do value) 
		to end
		| skip to () ; <-- type? value
	]
	rule/3: value
	change back tail rule type? value
	any [
		parse cases [some rule]
		do case
	]
]

;test
repeat n 10 [
	print [
		n

  switch2/default n [2 4 6 ['even] 1 3 5 ['odd]] [mold "--default--"]
	]
]
switch2 1 []
switch2/default 1 [] ["--default--"]

Actually, don't need to wrap the final two lines in the ANY block.

>> switch 1 [1 [2]]
== 2
>> switch1 1 [1 [2]]
== 2
>> switch2 1 [1 [2]]
** Throw Error: Return or exit not in function
** Where: switch2
** Near: return do value

fixing it right now...

I think it's fixed, but I'm also working on a version which evaluates 
the action block for every value and following block found.

That could be useful, allowing intersections of behaviours.

(and should not interfere with the usual usage we are used to.)

How about this:

switch2: func [

    "Selects a choice and evaluates the first block that follows it."
    [throw]
    value "Value to search for."
    cases [block!] "Block of cases to search."
    /default case [block!] "Default case if no others are found."
    /local rule
][
	rule: [
		1 1 () ; <-- value
		to block! set case block! ; <- re-use the 'case variable
		to end
		| skip to () ; <-- type? value
	]
	rule/3: value
	change back tail rule type? value
	parse cases [some rule]
	do case
]

{switch2: func [

    "Selects a choice and evaluates the first block that follows it. 
    This occurs for every matching value and following block found."
    [throw]
    value "Value to search for."
    cases [block!] "Block of cases to search."
    /default case [block!] "Default case if no others are found."
    /local rule
][
	rule: [
		1 1 () ; <-- value

  to block! set case block! (case: do case) ; <- re-use the 'case variable, 
  twice...
		| [skip to ()] ; <-- type? value
		| skip
	]
	rule/3: value
	rule/11/3: type? value
	any [
		all [
			parse cases [some rule]
			case
		]
		do case
	]
]}

;test
repeat n 10 [
	print [
		n

  switch2/default n [2 4 6 ['even] 1 3 5 ['odd]] [mold "--default--"]
	]
]
switch2 1 []
switch2/default 1 [] [probe "--default, ok--"]
switch2 1 [1 [probe "ok"]]
switch2 2 [1 [probe "bad"]]
switch2 1 [1 2 [probe "ok"]]
switch2 2 [1 2 [probe "ok"]]
switch2 3 [1 2 [probe "bad"]]

; multiple action blocks
switch2 1 [1 2 [probe "ok"] 1 3 4 [probe "ok#2"]] ; <-- 
switch2 2 [1 2 [probe "ok"] 1 3 4 [probe "bad"]]
switch2 3 [1 2 [probe "bad"] 1 3 4 [probe "ok"]]
switch2 4 [1 2 [probe "bad"] 1 3 4 [probe "ok"]]
switch2 5 [1 2 [probe "bad"] 1 3 4 [probe "bad"]]


switch2/default 5 [1 2 [probe "bad"] 1 3 4 [probe "bad"]] [probe 
"--default, ok--"]

looks slightly slower than the FIND-based version according to my 
measurements

parse block [to value to block! ...etc...]

if TO doesn't behave (i remember the "feature" with integers...) 
then maybe:

parse find block value [to block! ..etc..]

Ladislav. Hmm.. that's a bit disappointing, although understandable. 
It's a complex parse rule with some time spent in setup.

I'm enamoured with the multiple action block version, so I'll see 
if I can make a nice FIND-based version of that.

I may not understand, doesn't the SWITCH1 above do what you want?

Uncomment my second switch2 and have a look at the first test under 
"multiple action blocks".

For that test, both "ok" and "ok#2" should be printed.

aha, this is what you are after

it would be interesting to know how many users would want this feature 
- I am afraid it can help to hide a programming error

That's true. I will definitely be keeping it in my library though.

Can you remind us of the effect of [throw] on return etc.. ?

I got an error "return or exit not in function" when I had [throw] 
in the function header.

[throw]: "Normal" REBOL functions are expected to return a value 
when "encountering" RETURN. This behaviour differs from REBOL natives, 
though. See e.g. loop 1 [return 1], which does not mean the programmer 
wants the LOOP function to return one. The intended meaning is that 
the RETURN is meant for some other function, not for LOOP. To immitate 
this behaviour (necessary for control functions like SWITCH), Carl 
invented the [throw] attribute, which is "telling", that the function 
should "throw" RETURN to some other function instead of "using it" 
for itself.

For REBOL3 I proposed a change compatible with the above mentioned 
explanation allowing to discern return types.

Ah.. thankyou.

This means if I want to allow the user to use 'return in my switch2 
to jump out of her enclosing function, then I must add [throw] to 
the function header.

yes

unfortunately (in R2) this means, that you are unable to use RETURN 
for the SWITCH implementation purposes

Yes, I see now why you did not use it.

(unless you use "black magic", like my TFUNC instead of FUNC)