Joe Wallwork
Joe Wallwork
One discussion to have about these changes is whether we're happy that the examples are becoming more verbose. The versions without assertions will still be used in the docs but...
Huh! The CI has spontaneously hit a shape error again https://github.com/Cambridge-ICCS/FTorch/actions/runs/10880645825/job/30187925536 As suggested by @TomMelt, we should use valgrind to see if there's a memory leak.
Merged in @TomMelt's fixes from #175 and extended the test utils to consider other shapes than just 1D arrays. Now ready for review!
Could possibly use the `matrix` concept to support different build approaches within a single workflow definition, along with `spack` and `fpm` (to be added in #271, #429).
One way to do this is with Sphinx, which I've used before, although that's not necessarily the best tool to use for FTorch.
There's also this GitHub Action: https://github.com/gaurav-nelson/github-action-markdown-link-check.
To investigate: is Torch underneath smart enough to do this, or will we have to loop?
Might still be worth comparing performance between implementing this on the Fortran side vs Torch.
> This seems sensible, and impressive 😄 What is the argument for making `torch_tensor_from_array()` no longer return a value but instead operate internally? We can chat over lunch. Thanks! If...
I created a new branch [`autograd_toarray`](https://github.com/Cambridge-ICCS/FTorch/tree/autograd_toarray) off this one and merged in changes from #161. Can confirm the tests pass!