There is currently a bug found in our materialized view implementation
that happens when we delete a row, and then re-insert another row with
the same primary key.
Our insert code needs to detect updates and generate a DELETE +
INSERT. But in this case, after the initial DELETE, the fresh insert
generates another delete.
We ended up with the wrong response for aggregations (and I am pretty
sure even filter-only views would manifest the bug as well), where
groups that should still be present just disappeared because of the
extra delete.
A new test case is added that fails without the fix.
SQLite supports complex expressions in group by columns - because of
course it does...
So we need to make sure that a column is created for this expression if
it doesn't exist already, and compute it, the same way we compute
pre-projections in the filter operator.
Fixes#3363Fixes#3366Fixes#3365
In comparisons for joins, we were assuming that the left column belonged
to the left join (and vice-versa). That is incorrect, because you can
write the comparison condition in any order.
Fixes#3368
We had code for this, but the code had a fatal flaw: it tried to detect
a complex operation (an operation that needs projection), and return
false (no need for projection), for the others.
This is the exact opposite of what we should do: we should identify the
*simple* operations, and then return true (needs projection) for the
rest.
CAST is a special beast, since it is not a function, but rather, a
special opcode. Everything else above is the true just the same. But for
CAST, we have to do the extra work to capture it in the logical plan and
pass it down.
Fixes#3372Fixes#3370Fixes#3369
My goal with this patch is to be able to implement the ProjectOperator
for DBSP circuits using VDBE for expression evaluation.
*not* doing so is dangerous for the following reason: we will end up
with different, subtle, and incompatible behavior between SQLite
expressions if they are used in views versus outside of views.
In fact, even in our prototype had them: our projection tests, which
used to pass, were actually wrong =) (sqlite would return something
different if those functions were executed outside the view context)
For optimization reasons, we single out trivial expressions: they don't
have go through VDBE. Trivial expressions are expressions that only
involve Columns, Literals, and simple operators on elements of the same
type. Even type coercion takes this out of the realm of trivial.
Everything that is not trivial, is then translated with translate_expr -
in the same way SQLite will, and then compiled with VDBE.
We can, over time, make this process much better. There are essentially
infinite opportunities for optimization here. But for now, the main
warts are:
* VDBE execution needs a connection
* There is no good way in VDBE to pass parameters to a program.
* It is almost trivial to pollute the original connection. For example,
we need to issue HALT for the program to stop, but seeing that halt
will usually cause the program to try and halt the original program.
Subprograms, like the ones we use in triggers are a possible solution,
but they are much more expensive to execute, especially given that our
execution would essentially have to have a program with no other role
than to wrap the subprogram.
Therefore, what I am doing is:
* There is an in-memory database inside the projection operator (an
obvious optimization is to share it with *all* projection operators).
* We obtain a connection to that database when the operator is created
* We use that connection to execute our VDBE, which offers a clean, safe
and isolated way to execute the expression.
* We feed the values to the program manually by editing the registers
directly.
The op_insert function was incorrectly trying to capture an "old record"
for fresh INSERT operations when a table had dependent materialized views.
This caused a "Cannot delete: no current row" error because the cursor
wasn't positioned on any row for new inserts.
The issue was introduced in commit f38333b3 which refactored the state
machine for incremental view handling but didn't properly distinguish
between:
- Fresh INSERT operations (no old record exists)
- UPDATE operations without rowid change (old record should be captured)
- UPDATE operations with rowid change (already handled by DELETE)
This fix checks if cursor.rowid() returns a value before attempting to
capture the old record. If no row exists (fresh INSERT), we correctly
set old_record to None instead of erroring out.
I am also including tests to make sure this doesn't break. The reason I
didn't include tests earlier is that I didn't know it was possible to
run the tests under a flag. But in here, I am just adding the flag to
the execution script.
