Optimize Cursor.fetchall

[rostan-t]

The method Cursor.fetchall has the following docstring :

Fetch all (remaining) rows of a query result, returning them as a
sequence of sequences (e.g. a list of tuples). Note that the cursor's
arraysize attribute can affect the performance of this operation.
However, the method's implementation is simply :

return list(self)

which creates a list by calling self.fetchone, i.e. self._results.pop(0), once for each element of the list self._results.

This is a performance issue and can be done more straightforwardly.

[navina]

@RostanTabet : Iiuc, you are proposing to return a copy of the internal list object (_results) instead of letting the python's in-built list to construct a list instance by iterating through the objects. Is that right?

I am not that familiar with python internals. Does results = self._results create a copy of the list and assign to the lhs results variable ? Or is it just returning a reference to the list _results ?

[zhtaoxiang]

IIUC, this change may not be safe

1. this change the value of self._results to None, will affect other methods which need this result
2. the returned result is changed, previously, we change self to a list, now it's just part of self (i.e., self._results).

Can we try to fix those?

[zhtaoxiang]

To keep backward compatibility, maybe we can add a new method fetchresults? and keep the existing one unchanged?

[rostan-t]

Thanks for taking the time to review this pull request.

It is actually an expected behaviour that a call to fetchall affects subsequent calls to other methods. In fact, this is already the case since returning list(self) will call __iter__, which will itself call fetchone until the list is emptied.

To sum up, after a call to fetchall, fetchone has been called n times with n the size of the resulting list and self._results is empty.

There is actually a mistake in my code, I should have set self._results to [] instead of None¹. Appart from this, the bahavior is strictly the same, except from the fact that the result is directly returned as is instead of resulting from multiple calls to fetchone.

Footnotes

1. Edit: this is now fixed ↩

[navina]

@RostanTabet it might be useful to add a simple unit test to show that this is not an incompatible change in the function's behavior. Can you please add a unit test?

[HassanAITBRIK]

The method is already tested in :

pinot-dbapi/tests/unit/test_db.py

Line 618 in ef93a90

def test_fetches_all_results(self):

It needs more ?

[zhtaoxiang]

LGTM

[xiangfu0]

Can you please rebase and we can merge this.