format and int generic for x86 #13
Conversation
|
We will also see if the error observed on nightly has been fixed |
Codecov Report
@@ Coverage Diff @@
## master #13 +/- ##
=======================================
Coverage 82.97% 82.97%
=======================================
Files 5 5
Lines 276 276
=======================================
Hits 229 229
Misses 47 47
Continue to review full report at Codecov.
|
|
So COSMO.jl seems to be the cause of test failure on x86 architecture. And latest has some broken JLL artifacts. |
src/conic_form.jl
Outdated
| @@ -26,14 +26,14 @@ _set_type(::MOI.ConstraintIndex{F,S}) where {F,S} = S | |||
| cons_offset(conic::MOI.Zeros) = conic.dimension | |||
| cons_offset(conic::MOI.Nonnegatives) = conic.dimension | |||
| cons_offset(conic::MOI.SecondOrderCone) = conic.dimension | |||
| cons_offset(conic::MOI.PositiveSemidefiniteConeTriangle) = Int64((conic.side_dimension*(conic.side_dimension+1))/2) | |||
| cons_offset(conic::MOI.PositiveSemidefiniteConeTriangle) = Int((conic.side_dimension*(conic.side_dimension+1))/2) | |||
There was a problem hiding this comment.
This should be div(conic.side_dimension * (conic.side_dimension+1), 2) or even better, we should replace this con_offset by MOI.dimension
There was a problem hiding this comment.
I replaced the implementation with MOI.dimension. I didn't want to remove cons_offset altogether in the same PR, maybe do so in a follow-up
| @@ -26,14 +26,14 @@ _set_type(::MOI.ConstraintIndex{F,S}) where {F,S} = S | |||
| cons_offset(conic::MOI.Zeros) = conic.dimension | |||
| cons_offset(conic::MOI.Nonnegatives) = conic.dimension | |||
| cons_offset(conic::MOI.SecondOrderCone) = conic.dimension | |||
| cons_offset(conic::MOI.PositiveSemidefiniteConeTriangle) = Int64((conic.side_dimension*(conic.side_dimension+1))/2) | |||
| cons_offset(conic::MOI.PositiveSemidefiniteConeTriangle) = MOI.dimension(conic) | |||
There was a problem hiding this comment.
All cons_offset method are equivalent to MOI.dimension so we should just remove this cons_offset function.
There was a problem hiding this comment.
Missed your other comment, let's do it in a separate PR indeed
This should resolve the current fail on x86, hard-coded Int64 blocked conversions