[basex-talk] Predicate referring to following-sibling::node()[1]/self::text() incorrectly optimized

Hi Christian,

I wrote a query for a customer who wants to analyze their legacy ISO 12083 math formulas, in this case for detecting multiple subsequent <roman> elements of length >= 3 with only whitespace in between.

This is a synthetic test document:

<doc> <p> <formula><roman>tan</roman> <roman>tan</roman></formula> </p> <p> <formula><roman>sin</roman> ² <roman>sin</roman></formula> </p> <p> <formula><roman>cos</roman>³ <roman>cos</roman></formula> </p> </doc>

And this is the query I wrote:

let $rms := //(formula | dformula)//roman[string-length() gt 2] [

following-sibling::node()[1]/self::text()[not(normalize-space())] ] [

following-sibling::*[1]/self::roman[string-length() gt 2] ]/.., $docs := for $rm-context in $rms let $path := db:path($rm-context) group by $path return <doc path="{$path}">{ $rm-context }</doc> return <result count="{count($rms)}" docs="{count($docs)}">{ $docs }</result>

BaseX (up to version 9.6.3) erroneously reports all three <formula> elements as a result, while only the first should be reported.

This can be remedied by using parentheses, as in (following-sibling::node())[1]/self::text() and (following-sibling::*)[1]/self::roman. But this is inefficient, and the original query should just work™.

In the optimized original query there is following-sibling::text()[fn:position() = 1] and following-sibling::roman[fn:position() = 1]. These are incorrect optimizations of following-sibling::node()[1]/self::text() and following-sibling::*[1]/self::roman.

Gerrit