Support promotion against Base AbstractUnitRanges #213
Conversation
It's hard to support all `AbstractUnitRange` types unless one decides the types here should "lose" to all other types. But we can support the two most commonly used types.
Codecov Report
@@ Coverage Diff @@
## master #213 +/- ##
==========================================
+ Coverage 92.28% 92.29% +0.01%
==========================================
Files 14 14
Lines 1374 1376 +2
==========================================
+ Hits 1268 1270 +2
Misses 106 106
Continue to review full report at Codecov.
|
Previously `ZeroArrayPromise` required homogeneous axes, this supports promotion prior to construction. The test requires BlockArrays 0.16.17 (specifically JuliaArrays/BlockArrays.jl#213) but it's not a dependency of the package, so let's just trust in the lack of conflicts.
|
Could we get a new release? Tests are failing in JuliaImages/ImageCore.jl#182 |
|
Just tagged |
|
Thanks! I was unsure whether #126 was breaking. Looks like there were some changes from |
|
I wouldn’t consider that breaking, rather that the test was too strict: for a kronecker product I don’t think code should depend on getting the numbers correct up to floating point arithmetic. That is, if an algorithm is faster with a different order of multiplication that should be allowed. |
|
Thanks for making the right call! I didn't follow the details so was unsure of the core cause. Agreed 100% with your reasoning. |
Previously `ZeroArrayPromise` required homogeneous axes (all `AbstractUnitRange`s having the same type). This supports promotion prior to construction. The test requires BlockArrays 0.16.17 (specifically JuliaArrays/BlockArrays.jl#213) but it's not a dependency of the package, so let's just trust in the lack of conflicts.
It's hard to support all
AbstractUnitRangetypes unless one decidesthe types here should "lose" to all other types. But we can support
the two most commonly used types.