Fix broadcast failure for VectorOfArray with SVector{1}
#406
Conversation
|
Approval is required @ChrisRackauckas |
|
Yes it worked locally and I added it to the tests. But the CI seems need approval to be triggered. |
|
Thanks for checking! Lets be patient; Chris is very responsive to PRs and will approve the workflow, he just gets a ton of PRs/issues/messages between all the packages in the SciML ecosystem. |
|
Sure 👌 |
|
Looks like this gets a test failure. Look at LTS since the non-LTS inference failure has not been investigated. |
|
Thanks! Looks like my version doesn’t work for constant broadcasting. I’ll fix it tomorrow. |
| @@ -905,7 +905,9 @@ for (type, N_expr) in [ | |||
| else | |||
| unpacked = unpack_voa(bc, i) | |||
| arr_type = StaticArraysCore.similar_type(dest[:, i]) | |||
| dest[:, i] = if length(unpacked) == 1 | |||
| dest[:, i] = if length(unpacked) == 1 && length(dest[:, i]) == 1 | |||
| arr_type(unpacked[1]) | |||
There was a problem hiding this comment.
I added one more condition to deal with our special case - it worked in my place please rerun CI @ChrisRackauckas
There might be room for improvement, but for now, I just need to make it pass all tests.
|
https://github.com/SciML/RecursiveArrayTools.jl/actions/runs/11562374105/job/32187348125?pr=406 |
Just to double check, does the same error show up on a “dummy” PR too? |
Yes before I ran this branch I also tested master branch forked from upstream master - it showed this error |
|
It just needed a rebase, and seems fine now on v1.10. I'll track down the inference issue separately. |
Fix #378.
The corner case was considered but was never included in tests, so the error was not caught during development.