fix Vector{Bool} branches and factorize interped_data - #95
Conversation
|
Alternatively, I could just undo the factorization of function interped_data(rawdata, rawoffsets, ::Type{T}, ::Type{Nojagg}) where {T<:Bool}
return map(ntoh,reinterpret(T, rawdata))
endbut then it's less nice to read the long if-else chain. |
Codecov Report
@@ Coverage Diff @@
## master #95 +/- ##
==========================================
+ Coverage 84.61% 84.63% +0.02%
==========================================
Files 11 11
Lines 1287 1328 +41
==========================================
+ Hits 1089 1124 +35
- Misses 198 204 +6
Continue to review full report at Codecov.
|
interped_data
|
Tamas probably uses more |
|
I am back on track. Thanks for this PR, I will check it and report back, but I think it should be fine since the test coverage is pretty good for custom stuff. That's why I wanted to leave it there in first place ;) |
|
I tried it but I get ambiguities. I probably need to change the way I deal with |
|
Ah wait, I am dumb. This is colliding with an existing implementation in the core library... |
|
OK, I refactored |
|
I screwed up the merge, so I replaced this PR with #106 . ;) |
The change in #93 results in
BitVectorinstead ofVector{Bool}which breaks LazyBranch. We didn't have a test for it, but I added one here. I split up the largeinterped_datainto smaller chunks for non-, singly-, doubly-jagged branches so that it's easier to specialize on Bool. Evidently this breaks some LV64 test cases which I'm still figuring out.