Update convert.jl

[githubtomtom]

supports promote() to SArray

[andyferris]

OK, this is pretty interesting - what is the desired behavior here? That operations like SVector(1,2,3) + [1,2,3] returns an SVector? Or manual usage of promote?

I note that promote was designed for Number and operations such as * can invoke automatically promote - while the definitions here might not make great sense for that case. Also, should the length be checked to be compatible, or the size, or the axes (remembering we are coming from AbstractArray)?

[githubtomtom]

actually I was try to do SVector(1,2,3) .+ [1,2,3] (broadcasting) returning a SVector. Unfortunately overriding promote_rule() seems not able to achieve that automatically. I’m not familiar with broadcasting interface at the moment, would try to extend that piece later. By the way, now we can manually use promote() which is nice I think.

…

On Feb 19, 2020, at 6:58 PM, Andy Ferris ***@***.***> wrote: OK, this is pretty interesting - what is the desired behavior here? That operations like SVector(1,2,3) + [1,2,3] returns an SVector? Or manual usage of promote? I note that promote was designed for Number and operations such as * can invoke automatically promote - while the definitions here might not make great sense for that case. Also, should the length be checked to be compatible, or the size, or the axes (remembering we are coming from AbstractArray)? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#740?email_source=notifications&email_token=AIABXJPBFDZQTG7KMCYR3HTRDUGFVA5CNFSM4KXWOA42YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEMHKIOQ#issuecomment-588162106>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AIABXJPQHXCTUWQB4UDUKDLRDUGFVANCNFSM4KXWOA4Q>.

[c42f]

actually I was try to do SVector(1,2,3) .+ [1,2,3] (broadcasting) returning a SVector

Oh right! great to know thanks.

In that case I suggest we fix broadcasting rather than messing with promote which is generally for use with scalars. Unfortunately it may not be simple; the broadcasting machinery is already quite complex.

[c42f]

I've created #744 to track fixing broadcast as I think that's the correct way to fix your underlying issue.