specialize tweight on FixedPoint #89
Conversation
Codecov Report
@@ Coverage Diff @@
## master #89 +/- ##
==========================================
+ Coverage 93.15% 93.19% +0.04%
==========================================
Files 7 7
Lines 292 294 +2
==========================================
+ Hits 272 274 +2
Misses 20 20
Continue to review full report at Codecov.
|
|
I then realize that I don't have write permission to this repository when I'm about to merge this PR.. |
If upstream package (e.g., Interpolations, FixedPointNumbers) doesn't provide such specialization, then we add one here. This trick avoids incremental compilation warnings
| end) | ||
| # Reached when upstream package (e.g., FixedPoint) doesn't provide a specialization | ||
| Interpolations.tweight(A::AbstractArray{T}) where T<:FixedPoint = T | ||
| end |
There was a problem hiding this comment.
This is what I can come up with so far to minimize the danger of type piracy (and incremental compilation). Do you have any better ideas? @timholy
There was a problem hiding this comment.
If this is really an issue, I suspect we need to add a Requires dependency to Interpolations and put this in a @require block there. This solution seems likely to be fragile with respect to load order, and may not be fully correct wrt precompilation.
There was a problem hiding this comment.
I suspect we need to add a Requires dependency to Interpolations and put this in a
@requireblock there
This is more reasonable for me; I'll open a PR there.
may not be fully correct wrt precompilation.
Requires will disable the precompilation for this line of code, and thus won't get any trouble, am I understanding it correctly?
There was a problem hiding this comment.
Correct. I'm also not completely clear on whether the precompiled module would retain this runtime check of the method tables; I think it wouldn't, it would just store the answer at the time when it happened to run. In contrast, by putting it into the __init__ method it's guaranteed to be correct.
|
This one is possibly no longer needed by Augmentor, so I'm closing it. |
Currently, The return eltype of
ImageTransformations.box_extrapolation(img, Flat())isGray{N0f8}forGray{N0f8}input butFloat64forN0f8. To keep consistency, I add a specialization method here. (Technically it's a type piracy but I think it's safe to do so here)C.f. Evizero/Augmentor.jl#39
cc: @Evizero