2001-06-03  rasc:


Hi!

as promised a kind of an EBNF for the new parser.
The next seems to be a good job for "yacc"...

ok here it comes: 


--------------------------------------------------------------------------------------

Parser for SWISH (>2.2):

   swishSearchRequest:=   
              search_expr;


   search_expr :=
              "(" search_expr ")"
              unary_search_expr
              binary_search_exr
              post_search_expr
              word_expr;
      

   binary_search_exr:
              search_expr  "<AND>"    search_expr            // Logical AND
              search_expr  "<OR>"     search_expr            // Logical OR
              search_expr  "<ANDNOT>" search_expr;           // Swish   AND_NOT

   unary_search_expr:
              "<NOT>"    search_expr                         // Logical NOT
              "<PLUS>"   word_expr                           // AltaVista "+"    ???
              "<MINUS>"  word_expr;                          // AltaVista "-"    ???

   post_search_expr:                                         // Postprocess current search result list
              "<NARROW>"  property_expr;                     // select from result list

   property_expr:
              "(" property_expr ")"
              unary_property_expr
              binary_property_expr
              property_word_expr;

   binary_property_expr:
              property_expr "<AND>" property_expr
              property_expr "<OR>"  property_expr;

   unary_property_expr:
              "<NOT>" property_expr;


   property_word_expr:
              propertyname "="  regexpr
              propertyname "="  phrase
              propertyname ">"  phrase
              propertyname ">=" phrase
              propertyname "<"  phrase
              propertyname "<=" phrase


   word_expr:
              "<TRUE>"                                        // Logical True, (e.g for stopwords)
              "<FALSE>"                                       // Logical False  (usage ?)
              metaname "=" phrase                             // ------ also ">" "<" ">=" "<=" ?????
              metaname ">"  phrase
              metaname ">=" phrase
              metaname "<"  phrase
              metaname "<=" phrase
              phrase;

   phrase:
              word
              "\""string"\""
              wild_expr;


   wild_expr:
              standard_wild_expr                             //   word*
              extended_wild_expr                             //   word?and*wild*
              regular_expr;                                  //   (\w[a-z])hello.*

   string:
             printable character and whitespace;

   word:
             printable character without whitespace;

   

Operator Priorities:

     ( )                      
     =
     >=
     <=
     >
     <
    <PLUS>
    <MINUS>
    <NOT>
    <NARROW>
    <AND>
    <ANDNOT>
    <OR>


----

Additive:

-   metaname and propertyname implies a possible datatype conversion
    of word. (should be done with a routine, which may be dummy as
    long we don't have datatypes)

-   swish works only with internal operators  "<NOT>", "<AND>", etc.
    there has to be a mapping between uservisible operators  "and"
    to internal operators "<AND>" prior to any parsing.


 

Possible search request:


    author="Joe" AND (NOT "foo bar") NARROW 
          (swishlastmodified>=2001-05-01 AND swishdocname=".*/swishdir/.*")
  




    
---------------------------------------------------------------------------


Please have a look on it.
I tried to have a clear syntax structure tree.
So it should be possible to have recursive parser routine
calls, which makes the search request processing a little easier.


cu - rainer



---------------------------------------------------------------------------

Comments: