LIMIT in subquery ignored

[divral]

Hello,

I think there might be something wrong with the SUM() function in QLever.
I calculated the correct value by hand. The output of the Wikidata Query Service matches it. The one from QLever doesn't.

Correct output: 1.106.581.000.000
WDQS
Output by QLever: 1.223.190.000.000
QLever
QLever with outer query commented out, so values to be summed up are visible

For reference, the query takes the three largest Big Tech companies (according to the most recent number of their employees), then takes the most recent entry for their total revenue and sums it up.

Is there something one can do to get the correct value using QLever?

Best regards
Christina

[joka921]

Hi Christina,
Thanks for your report.
The problem is not the SUM function, but the LIMIT is ignored in the Subquery (The reported sum does not change if you remove the LIMIT.

There is currently nothing that you can change to make this query work correctly, but it should be a rather quick fix for me.

[hannahbast]

The problem is that the LIMIT 3 in the subquery has currently no effect for QLever. If you remove the LIMIT 3, you get the same result in Qlever and in WDQS:

[divral]

Thank you both! Yes, it is apparently also adding the latest total revenue value for Meta Platforms.

[joka921]

I have written a fix that seems to solve the problem. It is not yet merged into QLever, but already runs on the Wikidata instance.
In particular, your query example from above now yields the expected results.
Best regards
Johannes

[divral]

Hello Johannes,

I just re-made this query (using templates, so different syntax): "Combined total revenue of the top 3 big tech companies (when ordered by descending number of employees)"

In QLever, the LIMIT is ignored again: QLever, with sum
The resulting sum ( 1,223,190,000,000 ) includes Meta Platforms (but not Google): QLever, without sum

I checked by hand that the following is correct:
The ranking by employees is: Amazon > Microsoft > Apple > Google > Meta Platforms
Sum: 1,120,226,000,000

Wikidata Query Service outputs the expected values:
WDQS, with sum
WDQS, without sum

This is one curiosity I found.

The other is that the links in my original post still work regarding the LIMIT in the subquery. However, for some reason, they do not use the most recent total revenue values. For Microsoft, a newer, higher value from 2023 is now available.
old QLever, with sum
old QLever, without sum - I just added labels manually for this one

Just wanted to point it out. It's late so I might have made a mistake somewhere.

Best,
Christina

[hannahbast]

Can you reproduce this with a smaller example?

It's kind of hard to say where the mistake lies for such a monster query.

If it's really a bug in QLever, there should be a smaller query reproducing it.

[divral]

For the first mistake, I can narrow it down:

Query 1 - The LIMIT is minded
Query 2 - The LIMIT is ignored

As the comments mention, Query 2 is just Query 1 with an additional level of nesting.

[divral]

About the second one:

Query 3
It seems that the total revenue statement of Microsoft with point in time 2023 is just not used. It is the only such statement that is missing. In WDQS, that statement is returned.

EDIT: This appears to have been fixed now.

[hannahbast]

Great, thanks, here is an even simpler (and indeed minimal) example. When the second SELECT is commented out (lines 3 and 5), the query has three results as it should, otherwise it has many more.

PREFIX wdt: <http://www.wikidata.org/prop/direct/>
SELECT ?company WHERE {
  { SELECT ?company WHERE {
    { SELECT ?company WHERE { ?company wdt:P1128 ?employees } LIMIT 3 }
  } }
}

https://qlever.cs.uni-freiburg.de/wikidata/o4LHyz