Support explicit NULL values in timevectors #429
Conversation
WireBaron
force-pushed
the
br/timevector_explicit_null
branch
from
837e01e
to
0faaa44
Compare
|
I think that the update tests failed because this branch is too far behind |
WireBaron
force-pushed
the
br/timevector_explicit_null
branch
from
12ff8dc
to
f23066b
Compare
I did add a NULL value to the test in the sort pipeline element since I was most concerned about that behavior. It did suss out a few issues, but I'm not sure adding another test would exercise this code any differently. |
|
Looks like you need another rebase onto |
WireBaron
force-pushed
the
br/timevector_explicit_null
branch
from
17dd718
to
75ca499
Compare
| flags: time_vector::FLAG_IS_SORTED, | ||
| internal_padding: [0; 3], | ||
| points: (&*downsampled).into(), | ||
| null_val: std::vec::from_elem(0_u8, (downsampled.len() + 7) / 8).into() |
There was a problem hiding this comment.
What is from_elem()? I've never heard of the function before, and I can't find any refrences to it in the docs...
If it's what I think it is, it's probably more idiomatic to do
null_val: vec![0_u8; (downsampled.len() + 7) / 8].into()
| || state.points.as_slice().last().unwrap().ts | ||
| > series.points.as_slice().first().unwrap().ts) | ||
| { | ||
| state.flags ^= FLAG_IS_SORTED |
There was a problem hiding this comment.
is ^= really better than state.flags &= !FLAG_IS_SORTED here?
| points: [TSPoint; self.num_points], | ||
| null_val: [u8; (self.num_points + 7)/ 8], // bit vector, must be last element for alignment purposes |
There was a problem hiding this comment.
Ah, if only div_ceil() was stable, then we could use that instead of doing this manually. I do wonder if we should extract this out into a type, or at least extract out (num_points + 7)/ 8 into a function
| // Bit flags stored in Timevector flags | ||
| const FLAG_HAS_NULLS : u8 = 0x01; |
There was a problem hiding this comment.
The bitflags library may be useful for this
|
bors r+ |
|
Build succeeded: |
This change modifies the timevector structure by replacing the is_sorted field with a set of bit flags covering both whether a timevector is known to be sorted and whether it may contain NULL values. It also adds a bitvector with one bit per entry indicating whether that entry is NULL.
Fixes #425