Closed BoundaryValueProblems closed 6 years ago
Yeah, I suspected we'd have an issue with this. I think the best (easiest) course of action is to change all the Int
s to Int64
s.
@jpfairbanks - aside from killing performance on 32-bit machines, do you see any other drawbacks?
On second thought, these just look like errors in the test code. Let me look into this a bit more.
@BoundaryValueProblems - we're working on this. sorry for the delay.
32-bit tests are passing in https://github.com/JuliaGraphs/LightGraphs.jl/pull/999 . It is a bit more complicated than making everything Int
since overflow seems to be possible as indicated by some failing accuracy tests, so there are some hardcoded default upconversions to Int64
in that PR in places which are required. In addition, the splitrange
issue is a bad interaction in the Julia stdlib which got a PR ( https://github.com/JuliaLang/julia/pull/28865) but for now we can work around it by doing an Int
conversion on any collection that is @distributed for
looped over.
@ChrisRackauckas and @sbromberger , I just want to point out that the current master branch still gives me one error on my 32 bit linux machine as I reported at https://github.com/JuliaGraphs/LightGraphs.jl/pull/999#issuecomment-416012707 Thanks for your help!
This should fix it https://github.com/JuliaGraphs/LightGraphs.jl/pull/1004 , but it's hard to test since CI doesn't have 32-bit Linux. Can you test that PR locally?
@ChrisRackauckas , I just tested it, and it worked, i.e., the test passed without any error! Thanks a lot!!
I could successfully added LightGraphs.jl on my laptops after upgrading Julia from v0.6.4 to v0.7.0/v1.0.0. Then, I tried
Pkg.test("LightGraphs")
, and it completed without any error on my MacBook Pro (64 bit, OS X 10.11.6). However, on my 32 bit Linux machine (Ubuntu 16.04.4 LTS), the test failed with the following errors, which all seem to me related to Int32 vs Int64 mismatches.Thanks for your help!