World: r3wp

Petr, I have a copy with some notes here:
http://www.ross-gill.com/techniques/rsp/

I think, that Tim Peters knew what he wrote here: http://mail.python.org/pipermail/python-dev/1999-December/001770.html

Very nice comments. But comparing a parser with a regex is a bit 
unfair ;-)

Yeah, for either :)


Still, Peters seems to think that REBOL parse rule blocks are closures, 
for various reasons that are clear from context.

They aren't, though something like parse closures has been suggested 
during the latest round of enhancement proposals.

They are quite like smalltalk-closures. Without locals, but locals 
are not his point. (he may miss them for recursion).

Hey, locals and arguments (practically the same thing in REBOL) are 
the most important difference between closures and plain blocks. 
The difference is significant but Peters' background with Smalltalk 
made him miss it - Smalltalk "blocks" look like REBOL blocks but 
act like functions.

No, the main point is, easy definitions of code and referencing the 
original context. Rebol-blocks do that.

You can  have closures without any arguments.

The highlights he mentions is: lexically-scoped, code and data,  
freely mix computations in

That scoping is the difference between a closure and doing a "string" 
here.

REBOL blocks don't reference a context, but they may contain words 
that reference a context. Still, this distinction makes no difference 
to the argument that Peters was making - REBOL text processing is 
more powerful than regex and easier to use. It would be easier to 
replicate REBOL-style parsing in Python using closures and generators 
anyway (Peters' real subject), since that is the closest Python gets 
to Icon-style backtracking.

its not important what references the context, but that a variable 
can find one.

result := a > b
    ifTrue:[ 'greater' ]
    ifFalse:[ 'less' ]

There are two closures here. Rebol could do it the same way.

besides, Tim was a REBOL 1.x user

didn't parse come with 2.0?

it did, so it looks, that Tim is still silently watching REBOL

most probably he didn't leave as soon as we think :) i think i remember 
him on the list in 2.0 times too.

seems you are right

Maybe someone find it usefull:

remove-tags: func[html /except allowed-tags /local new x tag name 
tagchars][
	new: make string! length? html
	tagchars: charset [#"a" - #"z" #"A" - #"Z"]
	parse/all html [
		any [
			copy x to {<} copy tag thru {>}  (
				if not none? x [insert tail new x]
				if all [
					except
					parse tag ["<" opt #"/" copy name some tagchars to end]
					find allowed-tags name
				][	insert tail new tag ]
			)
		]
		copy x to end (if not none? x [insert tail new x])
	]
	new
]

Wouldn't it be nice, if the /case refinement for parse also worked 
with words, when parsing blocks?

>> parse [aBc] ['abc]
== true
>> parse/case [aBc] ['abc]
== true

It should work like when parsing strings, I think!

>> parse "aBc" ["abc"]
== true
>> parse/case "aBc" ["abc"]
== false

Why? words are not case sensitive by definition.

I would like the functionality, when parsing things like TeX. There 
the greek letter gamma is called gamma, and the same in capital is 
called Gamma. Now I have to invent the word capgamma or something.

>> parse ["Gamma"] ["gamma"]
== true
>> parse/case ["Gamma"] ["gamma"]
== false

If it were a safe and easy thing to change, I can see some value 
in it as an option but, since words--and REBOL--are case insensitive, 
I'm inclined to live with things as they are, and use string parsing 
if case sensitivity is needed. I think it's Oldes or Rebolek that 
sometimes requests the ability to parse non-loadable strings, using 
percentage values as an example. I think loading percentages would 
be awesome, but then there are other values we might want to load 
as well; where do you draw the line? I'm waiting to see what R3 holds 
with custom datatypes and such.

Yes, it's me who is calling to add posibility to load anything what 
is now throwing invalid datatype error.

And didn't you suggest that values throwing errors could be coerced 
to string! or another type? e.g. add an /any refinement to load, 
and any value in the string that can't be loaded would become a string 
(or maybe you could say you want them to be tags for easy identification).

I'm not sure how custom datatype lexing would work, unless it did 
something similar, calling custom lexers when running up against 
values the standard lexer doesn't understand. I can't remember how 
Gabriele's custom type mezz code works either; need to look at that.

I think, load/next can be used to handle invalid datatypes now:
>> b: {1 2 3 'x' ,}
== "1 2 3 'x' ,"
>> while [v: load/next b not empty? second v][probe v b: v/2]
[1 " 2 3 'x' ,"]
[2 " 3 'x' ,"]
[3 " 'x' ,"]
['x' " ,"]
** Syntax Error: Invalid word -- ,
** Near: (line 1) ,

Just add some hadler to convert the invalid datatype to something 
else what is loadable and then parse as a block

But such a preloader will slow down:(

I would like to know if string based parsing witch would handle all 
curent rebol datatypes can be faster or same fast as block parsing

Gabriele, yes it works with strings. But I have words! Thing is, 
I parse the string input from the user and produce words in an internal 
format. Then I parse those words for the final output, which can 
be different formats. I would expect parse/case to be case-sensitive, 
when parsing words, but parse/case is only for strings, therefore 
my suggestion.

what i'd suggest is - if case is important, don't make them into 
words :)

:D But it makes so much sense to work with words.

sure, but you can only have 8k or them (unless you make sure they 
never end up in system/words), so if you also counted case...

another way to counter the word limit is to use #issue datatype.

And there is some parse example how to deal with recursions while 
parsing strings? If you parse block, it's easy detect, what is string! 
and what is other type, but if you need to parse string, it's not 
so easy to detect for example strings like {some text {other "text"}}

(it should be a question - is there such a example?)

Words should be non-case sensitive, but is it always the case? I've 
found this today accidentaly:

>> a: [small Small]
== [small Small]
>> find/case a to word! "small"
== [small Small]
>> find/case a to word! "Small"
== [Small]

so /case with words works, at least in 'find

if it's working in find, it should be working on parse as well

well... case insensitivity for words is done via automatic aliasing 
of words that differ in case only. (i know this because we found 
a bug related to this :)

so internally, words are case-sensitive?

yes

Here's an idea to toss into the mix:

I am thinking of a new notation for strings using underscore (eg. 
 _"hello"_  ) in a parse block, which allows to specify whether they 
are delimited by whitespace or not. This would allow you to enable/disable 
the necessity for delimiters per-string. eg:

parse input [

 _"house"_   ; a complete word surrounded both sides by whitespace

 _"hous"   ;  this would match "house", "housing", "housed" or even 
 "housopoly" etc.. but left side must be whitespace

 "ad"_ ; this would match "ad", "fad", "glad" and right side must 
 be whitespace
]

But this would need string datatype to change.

On the other hand, I could just set underscore _ to a charset of 
whitespace, then use that with parse/all eg:

	_: charset " ^-^/"

parse/all input [
	[ _ "house" _ ]
]


though that wouldn't be as comfortable. Maybe I can create parse 
rules from a simpler dialect which understands the underscore _.
Just an idea...

Anton, Andrew had defined white space patterns in his patterns.r 
script which seems usable then you can use [ ws* "house" ws*] or 
other combinations as needed without underscore.  Andrew's solution 
for this and a lot of other things have given me some good mileage 
over the past few years.   WS*: [some WS]   and WS?: [any WS].   
 It makes for clean parse scripts clear once you adopt it.

I think either approach above can work well. I like the "look" of 
the underscore, and have done similar things with standard function 
names. For SOME, ANY, and OPT, the tag chars I prefer are +, *, and 
? resepctively; which are EBNF standard.

Oh yes, I've seen Andrew's patterns.r. I was just musing how to make 
it more concise without even using a short word like WS.  Actually 
the use case which sparked this idea was more of a "regex-level" 
pattern matcher, just a simple pattern matcher where the user writes 
the pattern to match filenames and to match strings appearing in 
file contents.

Gregg, + * ? could be a good idea. I'll throw that into my mix-bowl.