URL guessing is conceptually wrong

[VladimirAlexiev]

As #4 explains, XSPARQL tries to guess whether a term is a URL or literal. Such guessing is error-prone and completely unnecessary: it should know if something is a URL or not from the script, no guessing is necessary.

Eg consider this script

prefix ex: <http://example.org>
declare option saxon:output "method=text";

let $http := "http://this is a literal"
let $ftp := "ftp://this.is.url.com/"
let $url := "relative-URL"
construct {
  ex:subj ex:prop
    "http://this is a literal",
    $http,
    <{$http}>,
    <ftp://this.is.url.com/>,
    $ftp,
    <{$ftp}>,
    $url,
    <{$url}>.
}

The output is as follows

ex:subj  ex:prop  "http://this is a literal" . # OK
ex:subj  ex:prop  <http://this is a literal> . # NOK
ex:subj  ex:prop  <http://this is a literal> . # OK
ex:subj  ex:prop  <ftp://this.is.url.com/> . # OK
ex:subj  ex:prop  "ftp://this.is.url.com/" . # OK
ex:subj  ex:prop  ftp://this.is.url.com/ . # NOK and a syntax error
ex:subj  ex:prop  "relative-URL" . # OK
ex:subj  ex:prop  <relative-URL> . # OK

You see that the guessing makes two mistakes:

• for $http it wrongly assumes a URL because it starts with http://
• for <{$ftp}> it fails to emit a URL (because the guessing is incomplete, see list of URL schemes is incomplete #4) but fails to emit a literal either (because <...> indicates a URL)

[VladimirAlexiev]

#16 shows a specific example and code location