Implement Filter::mergeWith for MultiRange about velox HOT 7 CLOSED

@mbasmanova

I am a person who is interested in contributing to open source. Looks like this issue is a good one for a starter. May I work on this issue?

from velox.

Welcome, @amaliujia . Sure, feel free to pick this up. CC: @atanu1991 who is working on a related issue #153

from velox.

@mbasmanova

I want to discuss a bit on high level before start to implement.

MultiRange is a class that contains a list of Filter, with nan_allowed and null_allowed. So my current idea of the implementation is:

1. Two MultiRange can merge only if they have the same nan_allowed and null_allowed.
2. Two MultiRange can merge only if they have the same data type on all the Filters.
3. If 1 and 2 are true, then simply have a nested loop to do

 for f1 in MultiRangeA.getFilters():
      for f2 in MultiRangeB.getFilters():
           f1.mergeWith(f2);

Can you give me some suggestions?

from velox.

Rui,

MultiRange represents and OR of multiple filters. NaN values pass only if MultiRange::nanAllowed is true. Null values pass only if nullAllowed is true.

Merging two filters is combining the filters with an AND. A value passes the new filter only if it passes both the original filters.

Here is how I would represent a merge of two MultiRange filters (with just 2 filters in each for simplicity):

(a OR b) AND (c OR d) = (a AND c) OR (a AND d) OR (b AND c) OR (b AND d)

Therefore, what you have proposed is very close.

1. Calculate nullAllowed for the combined filter: bool bothNullAllowed = nullAllowed_ && other->testNull();
2. Calculate nanAllowed for the combined filter: {same as above}
3. Loop over all combinations of filters from left and right side and merge these.
4. Construct a new filter using flags and list of filters from previous steps.

In step (3), make sure to check if a merge of two filter is kAlwaysFalse, kIsNull, kIsNotNull. In this case, the filter can be dropped from the result (null and NaN flags in the nested filters do not count for MultiRange filter; only top-level flags take effect).

Step (3) may result in an empty list of filters or a list of one. If the list is empty, return kAlwaysFalse, kIsNull, or kIsNotNull as appropriate (see nullOrFalse(bothNullAllowed) in Filter.cpp). If the list contains just one filter, return that filter after incorporating the null flag, e.g. if bothNullAllowed && !filter->testNull() return filter.mergeWith(IsNull) else return filter.

That should be it.

(Check out #233 that implements merge logic for BytesValues and BytesRanges.)

from velox.

Thanks @mbasmanova!

I prepared a WIP impl based on your suggestions: #299.

Can you comment on the PR to see if I am on the right track? I haven't added test cases and I can add those later.

Sorry to occupy too much of your time. I am a beginner so probably need more guidance at this moment (for example on how to deal with nullness in Volex)? Maybe later I could help to contribute a doc about nullness, nan, always true, always false, is null, is not null with Filter.

from velox.

Is this still relevant? If so, what is blocking it? Is there anything you can do to help move it forward?

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs.

from velox.

Looks like this issue was resolved with #299 . Closing.

from velox.