World: r3wp

>>a: 1
>> t loop 1000000 [case [block? a [] integer? a []]] tt
0:00:00.534

>> t loop 1000000 [switch type?/word a  [block! [] integer! []]] 
tt
0:00:00.433

BT: It should at least be a separate function, but I suggest separate 
from the language too.  a) it's too much baggage, and b) the scope 
for recognising string patterns could better be handled in a more 
open way.

I did say that errors should complain loudly, but I didn't mean complain 
to me personally :)

so what did you profile Brian ?

Too much baggage for the pros, yes.  Just right for a Professor of 
Geography or home user, imho.   Don't fear those that don't normally 
program, help them.  No?

Too much baggage for the language.  Perhaps it could be incorporated 
into a beginner's binary?  I understand the need, even for pros too...

I look forward to R3 modules - then we won't have to be as picky 
about what gets included because it won't all get loaded by default.

I still don't see how a lexical foreign! would be that expensive. 
 Am I just not getting it?   Again, you cant   ADD email! url!  but 
they are valid types.  Make foreign! a valid type and then REBOL 
loads anything and series! operations can operate freely.

Adding a lexical foreign! means you add the overhead of checking 
for valid syntax to *all* REBOL code, not just LOAD. It removes your 
ability to trust that what comes out of LOAD is valid. That is a 
huge overhead to all REBOL code, and would be the source of most 
REBOL bugs in the future. It is a horrible thing to do to the community.

Well, what do you mean by valid?  foreign! would be valid, no?

It becomes a new datatype! with the semantics of junk.

No, it wouldn't, it would be meaningless. It would be like #[unset!].

I agree that allowing junk to linger, would mean that all code needs 
to handle junk... which can be sort of awkward

It can't throw an error, or you wouldn't be able to parse it to get 
anything ot of it. If it is another string type you'd have to check 
for it at every evaluation, because you aren't getting a nice error 
to escape out of yor code. It's either useless or a nightmare.

Yes, meaningless; but there would be less syntax errors.  And once 
in a block, cleaning could be done, but wouldn't have to be.  That 
would be up to the domain problem at hand.   I may never be convinced 
that having rules for what can be added would be that much different 
if REBOL were allowed to   make foreign!

it depends how rebol handles junked newline in code

Maxim;  I don't think so, no more than what junk is handled now by 
forcing pros to use string! parsing to get round human input foibles.

You *want* syntax errors. It's the only way to trust your input.

IMHO, commas sh

No *you* want syntax errors.  I want to load %kingjames.txt  and 
call sort.   ;)

...should be hendled like newline in code

This is why I prefer the fallback method, so the errors can be handled 
right away before they corrupt your data.

...appearing only when molded

What corruption?   $10,000,000.00 handled as foreign!  is not that 
much different than  do [abcd] when abcd has no value, no?

Right, and that throws an error.

so do [$10,000,000.00] throws, but   a: [$10,000,000.00] is accepted 
as foreign!

If it throws an error, you can't parse it.

(ok i'm in a big trouble now, my bottle of whisky is empty)

Yeah, but throw on do, not on make.  No?

I have a variety of booze here, and thanks for reminding me Steeve 
:)

Can't you see how cool it would be to let a person   load %textfile.txt 
and do analysis?  With foreign! sitting there ready to trigger if 
evaluated?

Errors throw on eval.

How about letting them do
    load/else %textfile.txt 'string

Absolutely.  And ship that with Core.  I'll stop whining and whinging. 
  ;)

That is what I mean by the fallback approach. We could even provide 
a transform function argument :)

just a question: IIRC, we will be able to build our own rebol distro 
to include our own mezzanines, right Brian ?

Some time ago we discussed this on a blog post.  Carl had some valid 
concerns about "where to restart the lexical scan", but I trust him 
and you and Gabriele to get that right, in an explainable way.   
So yes.

But ... come think.  The fallback now leads to professionals wondering 
if the human data was entered in quotes as string! or a fallback 
coercion....so I'd still go for a psuedostring! type to differentiate. 
 Maybe?

I don't know the name of the option, and I wouldn't be able to start 
implementing it until TRANSCODE gets more work.

Steeve, you will be able to build your own host. The core will still 
be the core. All of the platform-specific code is in the host.

You can also replace mezzanines by importing modules with your own 
functions.

so the debate is over, i will include the mezzanines i want for my 
distros

I can't wait to plug the C version of liquid within R3 and be able 
to browse associations using paths!   :-)

Forget "distros", you will be able to make programs :)

MAxim, don't talk about "liquid" i have not anymore

brian..... do you mean:


deliver: make program! true

?  ;-)

steeve what do you mean?

He's out.

He's out of whiskey.