RE: draft-reddy-dasl-protocol-00.txt

In this e-mail, I'd like to comment on just one part of
the spec. I'll make other comments in other e-mails.
I'm new to the notation and terminology, so
please bear with me for a while until I get up to speed.
I want to talk about the "where" component of a query.
It seems to me that it could be a bit simpler and become
more general in the process.

First, if I understand what you've done, you've chosen 
prefix notation. I think that's great.
There are only three approaches:
prefix, infix, and postfix. I've put some thought into it,
and I have convinced myself that prefix notation is the
correct choice for DASL.

Prefix notation allows operators that are associative
and commutative (e.g., AND, addition, multiplication, etc.)
to become n-ary operators, instead of being stuck as
binary operators. And, you've taken advantage
of this in the spec.

But, more importantly, prefix notation enables the
scope of an operator and its operands in the parse tree
to be processed programmatically without knowing anything
about the operator. (You only have to know about XML's
recursive begin and end bracketing scheme.)
This will be critical for applying three valued
elimination when querying across
multiple servers with different capabilities.
You can "prune" a subtree of the "parse tree",
because you know the exact extent of the parse
tree that's affected by an operator.

Since you've chosen prefix notation, we can forget about
operator precedence (which is a great benefit).
We can just define well known operators indefinitely.

Here is the "where" part of the example in section 5.3:

<d:where>
    <d:expr>
        <d:term>
            <d:gt> 
                <d:value>
                    <d:prop><d:getcontentlength/></d:prop>
                </d:value>
                <d:value>
                    <d:literal>10000</d:literal>
                </d:value>
            </d:gt>
        </d:term>
    </d:expr>
</d:where>

This is what I propose we simplify it to:

<d:where>
    <d:gt_integer> 
        <d:prop_integer>
            <d:getcontentlength/>
        </d:prop_integer>
        <d:literal_integer>10000</d:literal_integer>
    </d:gt_integer>
</d:where>

All I've done is to eliminate the syntactic metavariables
that we don't need, and introduced a bit stronger typing
to simplify life for the programmer who has to write the
program that digests this stuff. 

For example, I propose we would have gt_integer,
gt_float, and gt_string relational operators
that have integer, floating point, and string operands,
respectively, instead of a single polymorphic
gt operator that works with all types, and, worse,
with mixtures of types (e.g., gt(7, "cow") ). That way, you
don't have to define the promotions have to do,
e.g., promoting integers to floating point,
converting strings to integers, etc.
The promotion stuff complicates life for the programmer
writing the XML parsing stuff, and there is no
corresponding benefit.)

Each well known operator must be defined syntatically
and semantically in the spec. Syntatically, we must specify
the return value type. (First, we must specify the
set of base types, of course. I'm talking about
strings, integers, floating point, Booleans, and
maybe datetimes, etc.) Then, we must specify the
number of operands, and the type of each operand.
We must also specify if the operator can safely be
stntatically replaced by the "undefined" integer
or the "undefined" string, or the UNKNOWN Boolean
value using the rules of three valued elimination without
knowing anything about what the operator is
supposed to do.

If the thing is not an operator, then, it is a literal
or a property. I propose there be a pair of beginning and 
ending brackets for literals and properties 
of each base type, so you can tell from the syntax
that whether the types are correct.

Then, with my proposal, you can parse the "where"
expression for correct syntax, check that the types
of all operands are correct, and check that each operator
has the appropriate number of operands, all without
knowing anything about what the operator does.
You just need the operator descriptions with the
information I called out above. You can save the
operator descriptions in an array of C struct's
compiled into your program from just the information
in the spec.

There is no need for separators between the operands,
and you have not proposed any, which I think is good.
Because XML has begin and end brackets for everything,
you know where each operand starts and ends, and
when you have seen the last operator. In general,
you have to do this recursively, but that's not
a problem.

Then, we add a sentence to the spec. that says that
the first and only operator in the "where" brackets must
return a Boolean result, i.e., it must be a Boolean
operator (AND, OR, or NOT), or a relational operator
(gt_integer, ge_integer, eq_integer, ne_integer, ls_integer, le_integer,
gt_float, ge_float, eq_float, ne_float, ls_float, le_float,
gt_string, ge_string, eq_string, ne_string, ls_string, le_string, etc.).

I'm unfamiliar with the syntatic notation, so
I'll take a stab at it in standard BNF
notation. For section 5.6.2, we would have:

Name:  where
Namespace: DAV:
Purpose: Defines the search critera as a Boolean expression.

Unfortunately, BNF uses angle brackets to enclose syntatic
variables. So, in order to avoid confusion with XML's angle
brackets, I'll use curly braces instead. The other syntatically
special characters are "::=" ("is defined as") and "|" (separates
alternatives. English explanations are given within double quoted
strings.

{where} ::= {expr}
{expr} ::= {operator} {operand_list} {/operator}
{operator} ::= {Boolean_operator} | {relational_operator}
{Boolean_operator} ::= {AND} | {OR} | {NOT}
{AND} ::= <DAV: AND>
{OR} ::= <DAV: OR>
{NOT} ::=<DAV: NOT>
{relational_operator} ::= {integer_relational} | {float_relational} |
                          {string_relational} | {content_relational}
{integer_relational}::= 
               <DAV: lt_integer> | <DAV: le_integer> | 
               <DAV: eq_integer> | <DAV: ne_integer> |
               <DAV: ls_integer> | <DAV: le_integer
{float_relational} ::=
               <DAV: gt_float> | <DAV: ge_float> |
               <DAV: eq_float> | <DAV: ne_float> |
               <DAV: ls_float> | <DAV: le_float>
{string_relational}::=
               <DAV: gt_string> | <DAV: ge_string> |
               <DAV: eq_string> | <DAV: ne_string> |
               <DAV: ls_string> | <DAV: le_string>
{content_relational} ::= <DAV: contains>
{/operator} ::= "the same as {operator} but with a '/'
               after the '<' so that it is the XML ending bracket"
{operand_list} ::= {empty} | {operand} | {operand_list} {operand}
{empty} ::= 
{operand} ::= {expr} | {property} | {literal}
{property} ::= {property_type} {property_value} {/property_type}
{property_type} ::= <DAV:prop_integer> | <DAV:prop_float> |
                    <DAV:prop_string>
{/property_type} ::= "the same as {property_type} but with a '/'
                      after '<' so that it is an ending XML bracket"
{property_value} ::= {resource_property_name} | {intrinsic_name}
{resource_property_name ::= {identifier}
                            "this is a property of the resource,
                             e.g., Author, Title, etc."
{intrinsic_name} ::= <DAV: {identifier} />
                            "the spec. uses the example
                             getcontentlength as an intrinsic"
{literal} ::= {literal_type} {literal_value} {/literal_type}
{/literal_type} ::= "the same as {literal_type}, but with a '/'
                    after '<' so that it is an ending XML bracket"
{literal_type} ::= <DAV: literal_integer> | <DAV: literal_float> |
                   <DAV: literal_string> | <DAV: literal_Boolean>
{literal_value} ::= {integer_literal} | {float_literal} | 
                    {string_literal} | {Boolean_literal}
{integer_literal} ::= "signed or unsigned decimal integer"
{float_literal} ::= "signed or unsigned floating pointer number.
                    Contains decimal point or 'E' for exponent, or
                    both."
{string_literal} ::= "string enclosed in double quotes. A double
                     quote within the string is represented
                     as two adjacent quotes"
{Boolean_literal} ::= TRUE | FALSE | UNKNOWN

NOTE: Replacing a Boolean subtree by the Boolean literal UNKNOWN
is done as necessary when performing three valued elimination.

Additional restrictions not expressed in the above BNF:

* The operator immediately after <DAV:where> must return
a Boolean result. A resource is selected for inclusion
in the result set if and only if the 
Boolean result for that particular resource is TRUE (i.e., the
resource is not included if the result is FALSE or UNKNOWN).

* The {AND} and {OR} operators take two or more operands.
All the operands must be of type Boolean. The rules for
{AND} are applied in the following order: 
(1) If any operand is FALSE, the result is FALSE.
(2) If all operands are TRUE, the result is TRUE.
(3) Otherwise, at least one operand has the value UNKNOWN,
no operands are FALSE, and the value is UNKNOWN.
The rules for {OR} are applied in the following order:
(1) If any operand is TRUE, the result is TRUE.
(2) If all operands are FALSE, the result is FALSE.
(3) Otherwise, at least one operand has the value UNKNOWN,
no operands are TRUE, and the value is UNKNOWN.

* The {NOT} takes a single operand of type Boolean.
The rules for {NOT} are NOT TRUE is FALSE, NOT FALSE is TRUE,
and NOT UNKNOWN is UNKNOWN.

* The {string_relational}, {integer_relational}, and
{float_relational} relational operators take exactly two operands.
The type of the two operands must match the type required
by the relational operator. The relational operators
return a Boolean result. If either or both operands of
a relational operator are the undefined, then
the result of the relational operator is the Boolean
value UNKNOWN. Otherwise, the result is TRUE or FALSE
depending upon the result of the comparison.

* The {contains_relational} operator takes one or more
string valued expressions as its operands.

DISCUSSION:

The operators above are just the minimal subset of the
well known operators that must be implemented by all
DASL implementations. These are already in the spec.
I expect that there will be
many more well known operators defined later. Note that
this does not change the general scheme at all.
Any operator can be accommodated now or later.

If an implementation of an old version of the spec.
encounters an operator from a newer version of the spec.
that it doesn't have implemented, it can get the description
of this operator at run time, and can then deal with it
intelligently: (1) It can check that the operator has
the proper number and type of operands, and that 
what takes it as an argument expects the type that
the operator returns. (2) It can also find out if it can
used three valued elimination to get rid of the
operator. If the operator returns a string, a float, or
an integer, you go up recursively to the first relational
operator that is its parent, and replace the entire
section of the where expression for its parent relational 
operator by the Boolean literal UNKNOWN. If the operator
returns a Boolean, then if it is marked as "safe to
eliminate", you replace its entire subsection of the where
expression by UNKNOWN. Otherwise, it is marked as
not safe to eliminate, and you fail the query.

Alan Babich
> -----Original Message-----
> From:	Saveen Reddy (Exchange) [SMTP:saveenr@Exchange.Microsoft.com]
> Sent:	March 11, 1998 1:22 PM
> To:	'www-webdav-dasl@w3.org'
> Subject:	FW: draft-reddy-dasl-protocol-00.txt
> 
> 
> I just submitted the "strawman" DASL protocol sketch as an ID. I've
> included
> it below for you now, and I really look forward to your comments. 
> 
> Thanks,
> Saveen
> 
> 
> >  <<draft-reddy-dasl-protocol-00.txt>> 
> > 
> >  <<draft-reddy-dasl-protocol-00.doc>>  << File:
> draft-reddy-dasl-protocol-00.txt >>  << File:
> draft-reddy-dasl-protocol-00.doc >> 

Received on Friday, 13 March 1998 21:29:40 UTC