Compiler Panic with multiple group transforms

[richb-hanover]

What happened?

Using a recent Playground from main 9cc5bc8 in a Dev Container (not the deployed Playground) the PRQL compiler generates an error. PR #3864 now displays an attractive error message when a panic is encountered.

Can I provide any further diagnostic info? Thanks.

PRQL input

from inv=invoices
join item=invoice_items (==invoice_id)

group { inv.billing_city } (

  group { item.name } (
    aggregate {
      ct1 = count inv.name,
    }
  )
)

SQL output

N/A

Expected SQL output

N/A

MVCE confirmation

• Minimal example
• New issue

Anything else?

No response

[vanillajonathan]

I've investigated this a bit using a breakpoint in the browser dev tools.

The PRQL query gets passed (as the variable named value) to the prql.compile method on line 71:

sql = prql.compile(value);

However, no exception gets thrown because the compiler successfully compiles the query (I can compile this query with prqlc too) and returns the SQL query:

SELECT
  item.name,
  COUNT(*) AS ct1
FROM
  invoices AS inv
  JOIN invoice_items AS item ON inv.invoice_id = item.invoice_id
GROUP BY
  item.name

-- Generated by PRQL compiler version:0.10.1 (https://prql-lang.org)

It then passes the SQL query to DuckDB through the c.query method on line 121.

arrow = await c.query(sql);

which throws an Error with the message:

Binder Error: Table "item" does not have a column named "name"
LINE 8:   item.name
          ^

which gets caught and sets the React state duckdbError which is shown if you change tab to the "Query Results" tab.

[aljazerzen]

That's right @vanillajonathan, this is not a compiler panic, but bad PRQL program.

[max-sixty]

@vanillajonathan great debugging!

So the issue is that the old version of the compiler didn't have a panic, so the playground doesn't panic. Only the new version of the compiler panics.

So we're all good on #3864 — that works great. Thanks @vanillajonathan .

The remaining piece here is the panic from the compiler.

---

The error messages Binder Error: Table "item" does not have a column named "name" are because we try and run every query on the duckdb backend. We could adjust that to only run queries when the "Query Results" tab is displayed...

[richb-hanover]

I ask more about the "bad program" in #3832 (comment)

[richb-hanover]

Here's a screenshot showing the error in main:

[haydenflinner]

I managed to recreate this with this query:

from experiment
select started
append (from experiment | sort started | take 5 | select started)

[kgutwin]

With PR #4582, @haydenflinner 's query returns the error message:

Error:
   ╭─[:3:32]
   │
 3 │ append (from experiment | sort started | take 5 | select started)
   │                                ───┬───
   │                                   ╰───── internal compiler error; tracked at https://github.com/PRQL/prql/issues/3870
───╯

Which makes me think it's related to the sort issue identified in #4063 (which is of course related to this issue somehow as well...)