Output from forward in `test_reverse` has crazy size

[kshyatt]

QuantumKitHub/MatrixAlgebraKit.jl#86 for a reproducer

From within my augmented_primal function, right before it returns:

size.(primal) = ((19, 17), (17, 17), (17, 17), (1,))

From within test_reverse, after the forward from autodiff_thunk is run:

size.(y_ad) = ((19, 17), (1, 1), (17, 4818589264), (17,))

IDK what happened here but it looks like very bad things are happening to memory layout

[kshyatt]

OK some further testing seems to show this is only happening on 1.12, on 1.11 Enzyme correctly flagged the actual problem in my rule

[wsmoses]

A related cause of issues on 1.12 has been resolved, can you check if this still errs?

[wsmoses]

Is this still an issue?

[kshyatt]

I think not

[kshyatt]

Oh nope I am still seeing this, I'll try again to make a MWE...

[kshyatt]

The underlying rule seems to work because I don't see any problems on 1.10...

[wsmoses]

can you retry on latest? I think that should fix some rule related errors