Observe these lines of BPF filters in tcpdump/libpcap syntax:
1: not host x or host y
2: not (host x or host y)
3: not (host x or y)
4: not host x or y
5: (not host x) or host y
6: (not host x) or y
It is my opinion that host z matches all of the above (with the exception of 6 because that one has an invalid syntax).
My problem is with line 4. The tcpdump program sees that as equivalent to 5, but I think that is not intuitive and therefore not correct.
Line 5 is unambiguous, as is line 3. However, line 4 can mean both things, depending how you look at things. I am of the opinion that because you cannot see y separately from the "host" keyword, it is wrong to treat line 4 like line 5.
What is the parse logic behind this? Who can explain why 1 == 4 == 5 and why 2 != 4 and 3 != 4 ?
"I think that is not intuitive and therefore not correct."
Perhaps. But often intuition is in the eye of the beholder, and a precise specification is always more useful than "the parser does the intuitive thing". (Unless you like Perl, I suppose. But then you need the correct intuitions.)
Having said that, I can't find a precise specification of the pcap grammar, but man pcap-filter does explain how expressions are disambiguated in combinations of primitives with boolean operators:
Negation has highest precedence. Alternation and concatenation have equal precedence and associate left to right.
Many primitives consist of a keyword followed by an identifier, but the keyword may be omitted:
If an identifier is given without a keyword, the most recent keyword is assumed.
That has no impact on grouping. The omitted keyword is inserted without changing the parse. The example makes that clear:
For example,
not host vs and ace
is short for
not host vs and host ace
which should not be confused with
not ( host vs or ace )
What the description doesn't really make clear is the reason that your example 6 is a syntax error, which is that the parse is performed recursively inside parentheses, and consequently keywords inside the parenthetic expression do not change "the most recent keyword".
Related
I am writing a basic parser for a Scheme interpreter and here are the definitions I have set up to define the various type of tokens:
# 1. Parens
Type:
PAREN
Subtype:
LEFT_PAREN
Value:
'('
# 2. Operators (<=, =, +, ...)
Type:
OPERATOR
Subtype:
EQUALS
Value:
'='
Arity:
2
# 3. Types (2.5, "Hello", #f, etc.)
Type:
DATA
Subtype:
NUMBER
Value:
2.4
# 4. Procedures, builtins, and such
Type:
KEYWORD
Subtype:
BUILTIN
Value:
"set"
Arity:
2
PROCEDURE:
... // probably need a new class for this
Does the above seem like it's a good starting place? Are there some obvious things I'm missing here, or does this give me a "good-enough" foundation?
Your approach makes distinctions which really don't exist in the syntax of the language, and also makes decisions far too early. For example consider this program:
(let ((x 1))
(with-assignment-notes
(set! x 2)
(set! x 3)
x))
When I run this:
> (let ((x 1))
(with-assignment-notes
(set! x 2)
(set! x 3)
x))
setting x to 2
setting x to 3
3
In order for this to work with-assignment-notes has to somehow redefine what (set! ...) means in its body. Here's a hacky and probably incorrect (Racket) implementation of that:
(define-syntax with-assignment-notes
(syntax-rules (set!)
[(_ form ...)
(let-syntax ([rewrite/maybe
(syntax-rules (set!)
[(_ (set! var val))
(let ([r val])
(printf "setting ~A to ~A~%" 'var r)
(set! var r))]
[(_ thing)
thing])])
(rewrite/maybe form) ...)]))
So the critical features of any parser for a Lisp-family language are:
it should not make any decision about the semantics of the language that it can avoid making;
the structure it constructs must be available to the language itself as first-class objects;
(and optionally) the parser should be modifiable from the language itself.
As examples:
it is probably inevitable that the parser needs to make decisions about what is and is not a number and what sort of number it is;
it would be nice if it had default handling for strings, but this should ideally be controllable by the user;
it should make no decision at all about what, say (< x y) means but rather should return a structure representing it for interpretation by the language.
The reason for the last, optional, requirement is that Lisp-family languages are used by people who are interested in using them for implementing languages. Allowing the reader to be altered from within the language makes that hugely easier, since you don't have to start from scratch each time you want to make a language which is a bit like the one you started with but not completely.
Parsing Lisp
The usual approach to parsing Lisp-family languages is to have machinery which will turn a sequence of characters into a sequence of s-expressions consisting of objects which are defined by the language itself, notably symbols and conses (but also numbers, strings &c). Once you have this structure you then walk over it to interpret it as a program: either evaluating it on the fly or compiling it. Critically, you can also write programs which manipulate this structure itself: macros.
In 'traditional' Lisps such as CL this process is explicit: there is a 'reader' which turns a sequence of characters into a sequence of s-expressions, and macros explicitly manipulate the list structure of these s-expressions, after which the evaluator/compiler processes them. So in a traditional Lisp (< x y) would be parsed as (a cons of a symbol < and (a cons of a symbol x and (a cons of a symbol y and the empty list object)), or (< . (x . (y . ()))), and this structure gets handed to the macro expander and hence to the evaluator or compiler.
In Scheme it is a little more subtle: macros are specified (portably, anyway) in terms of rules which turn a bit of syntax into another bit of syntax, and it's not (I think) explicit whether such objects are made of conses & symbols or not. But the structure which is available to syntax rules needs to be as rich as something made of conses and symbols, because syntax rules get to poke around inside it. If you want to write something like the following macro:
(define-syntax with-silly-escape
(syntax-rules ()
[(_ (escape) form ...)
(call/cc (λ (c)
(define (escape) (c 'escaped))
form ...))]
[(_ (escape val ...) form ...)
(call/cc (λ (c)
(define (escape) (c val ...))
form ...))]))
then you need to be able to look into the structure of what came from the reader, and that structure needs to be as rich as something made of lists and conses.
A toy reader: reeder
Reeder is a little Lisp reader written in Common Lisp that I wrote a little while ago for reasons I forget (but perhaps to help me learn CL-PPCRE, which it uses). It is emphatically a toy, but it is also small enough and simple enough to understand: certainly it is much smaller and simpler than the standard CL reader, and it demonstrates one approach to solving this problem. It is driven by a table known as a reedtable which defines how parsing proceeds.
So, for instance:
> (with-input-from-string (in "(defun foo (x) x)")
(reed :from in))
(defun foo (x) x)
Reeding
To read (reed) something using a reedtable:
look for the next interesting character, which is the next character not defined as whitespace in the table (reedtables have a configurable list of whitespace characters);
if that character is defined as a macro character in the table, call its function to read something;
otherwise call the table's token reader to read and interpret a token.
Reeding tokens
The token reader lives in the reedtable and is responsible for accumulating and interpreting a token:
it accumulates a token in ways known to itself (but the default one does this by just trundling along the string handling single (\) and multiple (|) escapes defined in the reedtable until it gets to something that is whitespace in the table);
at this point it has a string and it asks the reedtable to turn this string into something, which it does by means of token parsers.
There is a small kludge in the second step: as the token reader accumulates a token it keeps track of whether it is 'denatured' which means that there were escaped characters in it. It hands this information to the token parsers, which allows them, for instance, to interpret |1|, which is denatured, differently to 1, which is not.
Token parsers are also defined in the reedtable: there is a define-token-parser form to define them. They have priorities, so that the highest priority one gets to be tried first and they get to say whether they should be tried for denatured tokens. Some token parser should always apply: it's an error if none do.
The default reedtable has token parsers which can parse integers and rational numbers, and a fallback one which parses a symbol. Here is an example of how you would replace this fallback parser so that instead of returning symbols it returns objects called 'cymbals' which might be the representation of symbols in some embedded language:
Firstly we want a copy of the reedtable, and we need to remove the symbol parser from that copy (having previously checked its name using reedtable-token-parser-names).
(defvar *cymbal-reedtable* (copy-reedtable nil))
(remove-token-parser 'symbol *cymbal-reedtable*)
Now here's an implementation of cymbals:
(defvar *namespace* (make-hash-table :test #'equal))
(defstruct cymbal
name)
(defgeneric ensure-cymbal (thing))
(defmethod ensure-cymbal ((thing string))
(or (gethash thing *namespace*)
(setf (gethash thing *namespace*)
(make-cymbal :name thing))))
(defmethod ensure-cymbal ((thing cymbal))
thing)
And finally here is the cymbal token parser:
(define-token-parser (cymbal 0 :denatured t :reedtable *cymbal-reedtable*)
((:sequence
:start-anchor
(:register (:greedy-repetition 0 nil :everything))
:end-anchor)
name)
(ensure-cymbal name))
An example of this. Before modifying the reedtable:
> (with-input-from-string (in "(x y . z)")
(reed :from in :reedtable *cymbal-reedtable*))
(x y . z)
After:
> (with-input-from-string (in "(x y . z)")
(reed :from in :reedtable *cymbal-reedtable*))
(#S(cymbal :name "x") #S(cymbal :name "y") . #S(cymbal :name "z"))
Macro characters
If something isn't the start of a token then it's a macro character. Macro characters have associated functions and these functions get called to read one object, however they choose to do that. The default reedtable has two-and-a-half macro characters:
" reads a string, using the reedtable's single & multiple escape characters;
( reads a list or a cons.
) is defined to raise an exception, as it can only occur if there are unbalanced parens.
The string reader is pretty straightforward (it has a lot in common with the token reader although it's not the same code).
The list/cons reader is mildly fiddly: most of the fiddliness is dealing with consing dots which it does by a slightly disgusting trick: it installs a secret token parser which will parse a consing dot as a special object if a dynamic variable is true, but otherwise will raise an exception. The cons reader then binds this variable appropriately to make sure that consing dots are parsed only where they are allowed. Obviously the list/cons reader invokes the whole reader recursively in many places.
And that's all the macro characters. So, for instance in the default setup, ' would read as a symbol (or a cymbal). But you can just install a macro character:
(defvar *qr-reedtable* (copy-reedtable nil))
(setf (reedtable-macro-character #\' *qr-reedtable*)
(lambda (from quote table)
(declare (ignore quote))
(values `(quote ,(reed :from from :reedtable table))
(inch from nil))))
And now 'x will read as (quote x) in *qr-reedtable*.
Similarly you could add a more compllicated macro character on # to read objects depending on their next character in the way CL does.
An example of the quote reader. Before:
> (with-input-from-string (in "'(x y . z)")
(reed :from in :reedtable *qr-reedtable*))
\'
The object it has returned is a symbol whose name is "'", and it didn't read beyond that of course. After:
> (with-input-from-string (in "'(x y . z)")
(reed :from in :reedtable *qr-reedtable*))
`(x y . z)
Other notes
Everything works one-character-ahead, so all of the various functions get the stream being read, the first character they should be interested in and the reedtable, and return both their value and the next character. This avoids endlessly unreading characters (and probably tells you what grammar class it can handle natively (obviously macro character parsers can do whatever they like so long as things are sane when they return).
It probably doesn't use anything which isn't moderately implementable in non-Lisp languages. Some
Macros will cause pain in the usual way, but the only one is define-token-parser. I think the solution to that is the usual expand-the-macro-by-hand-and-write-that-code, but you could probably help a bit by having an install-or-replace-token-parser function which dealt with the bookkeeping of keeping the list sorted etc.
You'll need a language with dynamic variables to implement something like the cons reeder.
it uses CL-PPCRE's s-expression representation of regexps. I'm sure other languages have something like this (Perl does) because no-one wants to write stringy regexps: they must have died out decades ago.
It's a toy: it may be interesting to read but it's not suitable for any serious use. I found at least one bug while writing this: there will be many more.
I'm having trouble working out how to use any of the functions in the Text.Parsec.Indent module provided by the indents package for Haskell, which is a sort of add-on for Parsec.
What do all these functions do? How are they to be used?
I can understand the brief Haddock description of withBlock, and I've found examples of how to use withBlock, runIndent and the IndentParser type here, here and here. I can also understand the documentation for the four parsers indentBrackets and friends. But many things are still confusing me.
In particular:
What is the difference between withBlock f a p and
do aa <- a
pp <- block p
return f aa pp
Likewise, what's the difference between withBlock' a p and do {a; block p}
In the family of functions indented and friends, what is ‘the level of the reference’? That is, what is ‘the reference’?
Again, with the functions indented and friends, how are they to be used? With the exception of withPos, it looks like they take no arguments and are all of type IParser () (IParser defined like this or this) so I'm guessing that all they can do is to produce an error or not and that they should appear in a do block, but I can't figure out the details.
I did at least find some examples on the usage of withPos in the source code, so I can probably figure that out if I stare at it for long enough.
<+/> comes with the helpful description “<+/> is to indentation sensitive parsers what ap is to monads” which is great if you want to spend several sessions trying to wrap your head around ap and then work out how that's analogous to a parser. The other three combinators are then defined with reference to <+/>, making the whole group unapproachable to a newcomer.
Do I need to use these? Can I just ignore them and use do instead?
The ordinary lexeme combinator and whiteSpace parser from Parsec will happily consume newlines in the middle of a multi-token construct without complaining. But in an indentation-style language, sometimes you want to stop parsing a lexical construct or throw an error if a line is broken and the next line is indented less than it should be. How do I go about doing this in Parsec?
In the language I am trying to parse, ideally the rules for when a lexical structure is allowed to continue on to the next line should depend on what tokens appear at the end of the first line or the beginning of the subsequent line. Is there an easy way to achieve this in Parsec? (If it is difficult then it is not something which I need to concern myself with at this time.)
So, the first hint is to take a look at IndentParser
type IndentParser s u a = ParsecT s u (State SourcePos) a
I.e. it's a ParsecT keeping an extra close watch on SourcePos, an abstract container which can be used to access, among other things, the current column number. So, it's probably storing the current "level of indentation" in SourcePos. That'd be my initial guess as to what "level of reference" means.
In short, indents gives you a new kind of Parsec which is context sensitive—in particular, sensitive to the current indentation. I'll answer your questions out of order.
(2) The "level of reference" is the "belief" referred in the current parser context state of where this indentation level starts. To be more clear, let me give some test cases on (3).
(3) In order to start experimenting with these functions, we'll build a little test runner. It'll run the parser with a string that we give it and then unwrap the inner State part using an initialPos which we get to modify. In code
import Text.Parsec
import Text.Parsec.Pos
import Text.Parsec.Indent
import Control.Monad.State
testParse :: (SourcePos -> SourcePos)
-> IndentParser String () a
-> String -> Either ParseError a
testParse f p src = fst $ flip runState (f $ initialPos "") $ runParserT p () "" src
(Note that this is almost runIndent, except I gave a backdoor to modify the initialPos.)
Now we can take a look at indented. By examining the source, I can tell it does two things. First, it'll fail if the current SourcePos column number is less-than-or-equal-to the "level of reference" stored in the SourcePos stored in the State. Second, it somewhat mysteriously updates the State SourcePos's line counter (not column counter) to be current.
Only the first behavior is important, to my understanding. We can see the difference here.
>>> testParse id indented ""
Left (line 1, column 1): not indented
>>> testParse id (spaces >> indented) " "
Right ()
>>> testParse id (many (char 'x') >> indented) "xxxx"
Right ()
So, in order to have indented succeed, we need to have consumed enough whitespace (or anything else!) to push our column position out past the "reference" column position. Otherwise, it'll fail saying "not indented". Similar behavior exists for the next three functions: same fails unless the current position and reference position are on the same line, sameOrIndented fails if the current column is strictly less than the reference column, unless they are on the same line, and checkIndent fails unless the current and reference columns match.
withPos is slightly different. It's not just a IndentParser, it's an IndentParser-combinator—it transforms the input IndentParser into one that thinks the "reference column" (the SourcePos in the State) is exactly where it was when we called withPos.
This gives us another hint, btw. It lets us know we have the power to change the reference column.
(1) So now let's take a look at how block and withBlock work using our new, lower level reference column operators. withBlock is implemented in terms of block, so we'll start with block.
-- simplified from the actual source
block p = withPos $ many1 (checkIndent >> p)
So, block resets the "reference column" to be whatever the current column is and then consumes at least 1 parses from p so long as each one is indented identically as this newly set "reference column". Now we can take a look at withBlock
withBlock f a p = withPos $ do
r1 <- a
r2 <- option [] (indented >> block p)
return (f r1 r2)
So, it resets the "reference column" to the current column, parses a single a parse, tries to parse an indented block of ps, then combines the results using f. Your implementation is almost correct, except that you need to use withPos to choose the correct "reference column".
Then, once you have withBlock, withBlock' = withBlock (\_ bs -> bs).
(5) So, indented and friends are exactly the tools to doing this: they'll cause a parse to immediately fail if it's indented incorrectly with respect to the "reference position" chosen by withPos.
(4) Yes, don't worry about these guys until you learn how to use Applicative style parsing in base Parsec. It's often a much cleaner, faster, simpler way of specifying parses. Sometimes they're even more powerful, but if you understand Monads then they're almost always completely equivalent.
(6) And this is the crux. The tools mentioned so far can only do indentation failure if you can describe your intended indentation using withPos. Quickly, I don't think it's possible to specify withPos based on the success or failure of other parses... so you'll have to go another level deeper. Fortunately, the mechanism that makes IndentParsers work is obvious—it's just an inner State monad containing SourcePos. You can use lift :: MonadTrans t => m a -> t m a to manipulate this inner state and set the "reference column" however you like.
Cheers!
So I've been trying to write a calculator with Scala's parser, and it's been fun, except that I found that operator associativity is backwards, and that when I try to make my grammar left-recursive, even though it's completely unambiguous, I get a stack overflow.
To clarify, if I have a rule like:
def subtract: Parser[Int] = num ~ "-" ~ add { x => x._1._1 - x._2 }
then evaluating 7 - 4 - 3 comes out to be 6 instead of 0.
The way I have actually implemented this is that I am composing a binary tree where operators are non-leaf nodes, and leaf nodes are numbers. The way I evaluate the tree is left child (operator) right child. When constructing the tree for 7 - 4 - 5, what I would like for it to look like is:
-
- 5
7 4 NULL NULL
where - is the root, its children are - and 5, and the second -'s children are 7 and 4.
However, the only tree I can construct easily is
-
7 -
NULL NULL 4 5
which is different, and not what I want.
Basically, the easy parenthesization is 7 - (4 - 5) whereas I want (7 - 4) - 5.
How can I hack this? I feel like I should be able to write a calculator with correct operator precedence regardless. Should I tokenize everything first and then reverse my tokens? Is it ok for me to just flip my tree by taking all left children of right children and making them the right child of the right child's parent and making the parent the left child of the ex-right child? It seems good at a first approximation, but I haven't really thought about it too deeply. I feel like there must just be some case that I'm missing.
My impression is that I can only make an LL parser with the scala parsers. If you know another way, please tell me!
Scala's standard implementation of parser combinators (the Parsers trait) do not support left-recursive grammars. You can, however, use PackratParsers if you need left recursion. That said, if your grammar is a simple arithmetic expression parser, you most definitely do not need left recursion.
Edit
There are ways to use right recursion and still keep left associativity, and if you are keen on that, just look up arithmetic expressions and recursive descent parsers. And, of course, as, I said, you can use PackratParsers, which allow left recursion.
But the easiest way to handle associativity without using PackratParsers is to avoid using recursion. Just use one of the repetition operators to get a List, and then foldLeft or foldRight as required. Simple example:
trait Tree
case class Node(op: String, left: Tree, right: Tree) extends Tree
case class Leaf(value: Int) extends Tree
import scala.util.parsing.combinator.RegexParsers
object P extends RegexParsers {
def expr = term ~ (("+" | "-") ~ term).* ^^ mkTree
def term = "\\d+".r ^^ (_.toInt)
def mkTree(input: Int ~ List[String ~ Int]): Tree = input match {
case first ~ rest => ((Leaf(first): Tree) /: rest)(combine)
}
def combine(acc: Tree, next: String ~ Int) = next match {
case op ~ y => Node(op, acc, Leaf(y))
}
}
You can find other, more complete, examples on the scala-dist repository.
I'm interpreting your question as follows:
If you write rules like def expression = number ~ "-" ~ expression and then evalute on each node of the syntax tree, then you find that in 3 - 5 - 4, the 5 - 4 is computed first, giving 1 as a result, and then 3 - 1 is computed giving 2 as a result.
On the other hand, if you write rules like def expression = expression ~ "-" ~ number, the rules are left-recursive and overflow the stack.
There are three solutions to this problem:
Post-process the abstract syntax tree to convert it from a right-associative tree to a left-associative tree. If you're using actions on the grammar rules to do the computation immediately, this won't work for you.
Define the rule as def expression = repsep(number, "-") and then when evaluating the computation, loop over the parsed numbers (which will appear in a flat list) in whichever direction provides you the associativity you need. You can't use this if more than one kind of operator will appear, since the operator will be thrown away.
Define the rule as def expression = number ~ ( "-" ~ number) *. You'll have an initial number, plus a set of operator-number pairs in a flat list, to process in any direction you want (though left-to-right is probably easier here).
Use PackratParsers as Daniel Sobral suggested. This is probably your best and simplest choice.
I've started looking into REBOL, just for fun, and as a fan of programming languages, I really like seeing new ideas and even just alternative syntaxes. REBOL is definitely full of these. One thing I noticed is the use of '/' as the path operator which can be used similarly to the '.' operator in most object-oriented programming languages. I have not programmed in REBOL extensively, just looked at some examples and read some documentation, but it isn't clear to me why there's no ambiguity with the '/' operator.
x: 4
y: 2
result: x/y
In my example, this should be division, but it seems like it could just as easily be the path operator if x were an object or function refinement. How does REBOL handle the ambiguity? Is it just a matter of an overloaded operator and the type system so it doesn't know until runtime? Or is it something I'm missing in the grammar and there really is a difference?
UPDATE Found a good piece of example code:
sp: to-integer (100 * 2 * length? buf) / d/3 / 1024 / 1024
It appears that arithmetic division requires whitespace, while the path operator requires no whitespace. Is that it?
This question deserves an answer from the syntactic point of view. In Rebol, there is no "path operator", in fact. The x/y is a syntactic element called path. As opposed to that the standalone / (delimited by spaces) is not a path, it is a word (which is usually interpreted as the division operator). In Rebol you can examine syntactic elements like this:
length? code: [x/y x / y] ; == 4
type? first code ; == path!
type? second code
, etc.
The code guide says:
White-space is used in general for delimiting (for separating symbols).
This is especially important because words may contain characters such as + and -.
http://www.rebol.com/r3/docs/guide/code-syntax.html
One acquired skill of being a REBOler is to get the hang of inserting whitespace in expressions where other languages usually do not require it :)
Spaces are generally needed in Rebol, but there are exceptions here and there for "special" characters, such as those delimiting series. For instance:
[a b c] is the same as [ a b c ]
(a b c) is the same as ( a b c )
[a b c]def is the same as [a b c] def
Some fairly powerful tools for doing introspection of syntactic elements are type?, quote, and probe. The quote operator prevents the interpreter from giving behavior to things. So if you tried something like:
>> data: [x [y 10]]
>> type? data/x/y
>> probe data/x/y
The "live" nature of the code would dig through the path and give you an integer! of value 10. But if you use quote:
>> data: [x [y 10]]
>> type? quote data/x/y
>> probe quote data/x/y
Then you wind up with a path! whose value is simply data/x/y, it never gets evaluated.
In the internal representation, a PATH! is quite similar to a BLOCK! or a PAREN!. It just has this special distinctive lexical type, which allows it to be treated differently. Although you've noticed that it can behave like a "dot" by picking members out of an object or series, that is only how it is used by the DO dialect. You could invent your own ideas, let's say you make the "russell" command:
russell [
x: 10
y: 20
z: 30
x/y/z
(
print x
print y
print z
)
]
Imagine that in my fanciful example, this outputs 30, 10, 20...because what the russell function does is evaluate its block in such a way that a path is treated as an instruction to shift values. So x/y/z means x=>y, y=>z, and z=>x. Then any code in parentheses is run in the DO dialect. Assignments are treated normally.
When you want to make up a fun new riff on how to express yourself, Rebol takes care of a lot of the grunt work. So for example the parentheses are guaranteed to have matched up to get a paren!. You don't have to go looking for all that yourself, you just build your dialect up from the building blocks of all those different types...and hook into existing behaviors (such as the DO dialect for basics like math and general computation, and the mind-bending PARSE dialect for some rather amazing pattern matching muscle).
But speaking of "all those different types", there's yet another weirdo situation for slash that can create another type:
>> type? quote /foo
This is called a refinement!, and happens when you start a lexical element with a slash. You'll see it used in the DO dialect to call out optional parameter sets to a function. But once again, it's just another symbolic LEGO in the parts box. You can ascribe meaning to it in your own dialects that is completely different...
While I didn't find any written definitive clarification, I did also find that +,-,* and others are valid characters in a word, so clearly it requires a space.
x*y
Is a valid identifier
x * y
Performs multiplication. It looks like the path operator is just another case of this.
I am experimenting with bit pattern matching in Erlang:
-module(test).
-export([test/2]).
%test(P,<<X:P,0:1>>) ->
% X.
test(P,X) ->
<<Y:P,0:1>> = X,
Y.
when compiling the commented out version of test/2 i get a complaint that "variable 'P' is unbound".
Is there any good reason for not allowing the first version to work the same as the second?
Because in the commented out version P is a length - for it to work Erlang would need to perform a double match - match the value of the 2nd parameter with a pattern which is undetermined...
The question you are asking in a clause pattern match is "is this the clause for me" - you can't 'pop into the clause' and then back out if it isn't...
In the second example X is bound before the match, you are committed to going into the clause and if <<Y:P,0:1>> don't match X, well crash time!
The reason is that arguments to the function are evaluated independent of each other. The correctness of bindings to variables in only checked as a second step.
This means that in your first example P will be unbound when evaluating the second argument, which is against the rules of pattern matching. In contrast, in your second example, P is bound at the time of evaluating the pattern match on the binary.