Optimization issue?

On Thu, Oct 24, 2019 at 10:54 AM Johannes Echterhoff <echterhoff at interactive-instruments.de> wrote: > > Hello, > > I’m experiencing some strange behavior when executing the following query in the BaseX 9.2.4 GUI (nevermind the awkward query structure – that is just to be able to inspect the results of the individual evaluation steps): > > > > let $t := <x>200.0000</x> > > let $num := number($t) > > let $gt5000 := $num gt 5000 > > let $ltMinus5000 := $num lt -5000 > > return $gt5000 or $ltMinus5000 > > > > The result is true, although it should be false. > > > > The compilation and optimization logs for this query in the GUI tell us the following: > > ---------- > > Compiling: > > - rewrite > comparison to range comparison: ($num_1 > 5000) -> ($num_1 >= 5000.000000000001) > > - rewrite 'gt' comparison to range comparison: ($num_1 gt 5000) -> ($num_1 >= 5000.000000000001) > > - pre-evaluate unary expression to xs:integer item: -5000 > > - rewrite < comparison to range comparison: ($num_1 < -5000) -> ($num_1 <= -5000.000000000001) > > - rewrite 'lt' comparison to range comparison: ($num_1 lt -5000) -> ($num_1 <= -5000.000000000001) > > - inline $gt5000_2 > > - inline $ltMinus5000_3 > > - simplify or: (true()) > > - pre-evaluate or to xs:boolean item: (true()) -> true() > > - remove unused variable $num_1 > > - remove unused variable $t_0 > > - simplify FLWOR expression: true() > > Optimized Query: > > true() > > ---------- > > > > When the return statement is changed to “return ($num,$gt5000,$ltMinus5000,$gt5000 or $ltMinus5000)” then the result of the boolean operation (last element of the result sequence) is false, as expected. > > The GUI log then is: > > ---------- > > Compiling: > > - rewrite > comparison to range comparison: ($num_1 > 5000) -> ($num_1 >= 5000.000000000001) > > - rewrite 'gt' comparison to range comparison: ($num_1 gt 5000) -> ($num_1 >= 5000.000000000001) > > - pre-evaluate unary expression to xs:integer item: -5000 > > - rewrite < comparison to range comparison: ($num_1 < -5000) -> ($num_1 <= -5000.000000000001) > > - rewrite 'lt' comparison to range comparison: ($num_1 lt -5000) -> ($num_1 <= -5000.000000000001) > > Optimized Query: > > let $t_0 := element x { ("200.0000") } let $num_1 := number($t_0) let $gt5000_2 := ($num_1 >= 5000.000000000001) let $ltMinus5000_3 := ($num_1 <= -5000.000000000001) return ($num_1, $gt5000_2, $ltMinus5000_3, ($gt5000_2 or $ltMinus5000_3)) > > ---------- > > > > Am I missing something, or could this be a bug in BaseX 9.2.4? With BaseX 9.0.2, both queries (with the different return statements) returned the expected value false for the boolean operation. > > > > Best regards, > > Johannes > > > >

-----Ursprüngliche Nachricht----- Von: Christian Grün [mailto:christian.gruen at gmail.com] Gesendet: Donnerstag, 24. Oktober 2019 12:41 An: Johannes Echterhoff <echterhoff at interactive-instruments.de> Cc: BaseX <basex-talk at mailman.uni-konstanz.de> Betreff: Re: [basex-talk] Optimization issue? Dear Johannes, Thanks for the reproducible test case. The bug (a too eager merge of combined range expressions) was fixed and a new snapshot is available [1,2]. Cheers Christian [1] https://github.com/BaseXdb/basex/issues/1744 [2] http://files.basex.org/releases/latest/ On Thu, Oct 24, 2019 at 10:54 AM Johannes Echterhoff <echterhoff at interactive-instruments.de> wrote: > > Hello, > > I’m experiencing some strange behavior when executing the following query in the BaseX 9.2.4 GUI (nevermind the awkward query structure – that is just to be able to inspect the results of the individual evaluation steps): > > > > let $t := <x>200.0000</x> > > let $num := number($t) > > let $gt5000 := $num gt 5000 > > let $ltMinus5000 := $num lt -5000 > > return $gt5000 or $ltMinus5000 > > > > The result is true, although it should be false. > > > > The compilation and optimization logs for this query in the GUI tell us the following: > > ---------- > > Compiling: > > - rewrite > comparison to range comparison: ($num_1 > 5000) -> ($num_1 > >= 5000.000000000001) > > - rewrite 'gt' comparison to range comparison: ($num_1 gt 5000) -> > ($num_1 >= 5000.000000000001) > > - pre-evaluate unary expression to xs:integer item: -5000 > > - rewrite < comparison to range comparison: ($num_1 < -5000) -> > ($num_1 <= -5000.000000000001) > > - rewrite 'lt' comparison to range comparison: ($num_1 lt -5000) -> > ($num_1 <= -5000.000000000001) > > - inline $gt5000_2 > > - inline $ltMinus5000_3 > > - simplify or: (true()) > > - pre-evaluate or to xs:boolean item: (true()) -> true() > > - remove unused variable $num_1 > > - remove unused variable $t_0 > > - simplify FLWOR expression: true() > > Optimized Query: > > true() > > ---------- > > > > When the return statement is changed to “return ($num,$gt5000,$ltMinus5000,$gt5000 or $ltMinus5000)” then the result of the boolean operation (last element of the result sequence) is false, as expected. > > The GUI log then is: > > ---------- > > Compiling: > > - rewrite > comparison to range comparison: ($num_1 > 5000) -> ($num_1 > >= 5000.000000000001) > > - rewrite 'gt' comparison to range comparison: ($num_1 gt 5000) -> > ($num_1 >= 5000.000000000001) > > - pre-evaluate unary expression to xs:integer item: -5000 > > - rewrite < comparison to range comparison: ($num_1 < -5000) -> > ($num_1 <= -5000.000000000001) > > - rewrite 'lt' comparison to range comparison: ($num_1 lt -5000) -> > ($num_1 <= -5000.000000000001) > > Optimized Query: > > let $t_0 := element x { ("200.0000") } let $num_1 := number($t_0) let > $gt5000_2 := ($num_1 >= 5000.000000000001) let $ltMinus5000_3 := > ($num_1 <= -5000.000000000001) return ($num_1, $gt5000_2, > $ltMinus5000_3, ($gt5000_2 or $ltMinus5000_3)) > > ---------- > > > > Am I missing something, or could this be a bug in BaseX 9.2.4? With BaseX 9.0.2, both queries (with the different return statements) returned the expected value false for the boolean operation. > > > > Best regards, > > Johannes > > > >

On Fri, Oct 25, 2019 at 11:23 AM Johannes Echterhoff <echterhoff at interactive-instruments.de> wrote: > > Dear Christian, > Tested the snapshot yesterday, and now the expression works as expected. Thank you. > Do you have a target date for the release of BaseX 9.3? > Best regards, > Johannes > > > -----Ursprüngliche Nachricht----- > Von: Christian Grün [mailto:christian.gruen at gmail.com] > Gesendet: Donnerstag, 24. Oktober 2019 12:41 > An: Johannes Echterhoff <echterhoff at interactive-instruments.de> > Cc: BaseX <basex-talk at mailman.uni-konstanz.de> > Betreff: Re: [basex-talk] Optimization issue? > > Dear Johannes, > > Thanks for the reproducible test case. The bug (a too eager merge of combined range expressions) was fixed and a new snapshot is available [1,2]. > > Cheers > Christian > > [1] https://github.com/BaseXdb/basex/issues/1744 > [2] http://files.basex.org/releases/latest/ > > > > > On Thu, Oct 24, 2019 at 10:54 AM Johannes Echterhoff <echterhoff at interactive-instruments.de> wrote: > > > > Hello, > > > > I’m experiencing some strange behavior when executing the following query in the BaseX 9.2.4 GUI (nevermind the awkward query structure – that is just to be able to inspect the results of the individual evaluation steps): > > > > > > > > let $t := <x>200.0000</x> > > > > let $num := number($t) > > > > let $gt5000 := $num gt 5000 > > > > let $ltMinus5000 := $num lt -5000 > > > > return $gt5000 or $ltMinus5000 > > > > > > > > The result is true, although it should be false. > > > > > > > > The compilation and optimization logs for this query in the GUI tell us the following: > > > > ---------- > > > > Compiling: > > > > - rewrite > comparison to range comparison: ($num_1 > 5000) -> ($num_1 > > >= 5000.000000000001) > > > > - rewrite 'gt' comparison to range comparison: ($num_1 gt 5000) -> > > ($num_1 >= 5000.000000000001) > > > > - pre-evaluate unary expression to xs:integer item: -5000 > > > > - rewrite < comparison to range comparison: ($num_1 < -5000) -> > > ($num_1 <= -5000.000000000001) > > > > - rewrite 'lt' comparison to range comparison: ($num_1 lt -5000) -> > > ($num_1 <= -5000.000000000001) > > > > - inline $gt5000_2 > > > > - inline $ltMinus5000_3 > > > > - simplify or: (true()) > > > > - pre-evaluate or to xs:boolean item: (true()) -> true() > > > > - remove unused variable $num_1 > > > > - remove unused variable $t_0 > > > > - simplify FLWOR expression: true() > > > > Optimized Query: > > > > true() > > > > ---------- > > > > > > > > When the return statement is changed to “return ($num,$gt5000,$ltMinus5000,$gt5000 or $ltMinus5000)” then the result of the boolean operation (last element of the result sequence) is false, as expected. > > > > The GUI log then is: > > > > ---------- > > > > Compiling: > > > > - rewrite > comparison to range comparison: ($num_1 > 5000) -> ($num_1 > > >= 5000.000000000001) > > > > - rewrite 'gt' comparison to range comparison: ($num_1 gt 5000) -> > > ($num_1 >= 5000.000000000001) > > > > - pre-evaluate unary expression to xs:integer item: -5000 > > > > - rewrite < comparison to range comparison: ($num_1 < -5000) -> > > ($num_1 <= -5000.000000000001) > > > > - rewrite 'lt' comparison to range comparison: ($num_1 lt -5000) -> > > ($num_1 <= -5000.000000000001) > > > > Optimized Query: > > > > let $t_0 := element x { ("200.0000") } let $num_1 := number($t_0) let > > $gt5000_2 := ($num_1 >= 5000.000000000001) let $ltMinus5000_3 := > > ($num_1 <= -5000.000000000001) return ($num_1, $gt5000_2, > > $ltMinus5000_3, ($gt5000_2 or $ltMinus5000_3)) > > > > ---------- > > > > > > > > Am I missing something, or could this be a bug in BaseX 9.2.4? With BaseX 9.0.2, both queries (with the different return statements) returned the expected value false for the boolean operation. > > > > > > > > Best regards, > > > > Johannes > > > > > > > >