Inbounds tuple iteration and reduction

[maleadt]

I noticed prod(length, (1,)) performed a bounds check, caused by the reduction not specifying @inbounds (which should be safe, because it is guarded by a call to done) and the tuple iteration functions not propagating it. Not sure about the correctness of this, so going for limited scope first.

[nalimilan]

Unfortunately I'm not sure this is allowed in general. See discussion at #21402.

[maleadt]

Hmm, I based it on existing code:

julia/base/reduce.jl

Line 46 in 74f5328

@inbounds (x, i) = next(itr, i)

[mbauman]

Yes, we're inconsistent here… and at a standoff between two warring factions.

Part of the trouble is that next kinda got grandfathered in for iterating over Array since @inbounds used to propagate through all inlined functions for Array back in the dark ages. So when that got fixed, we hit perf regressions without them.

I'm on side @inbounds next.

[nalimilan]

If we merge this I'll request merging #21402 too (which is very useful to to make checking the presence of missing values in a non-nullable array a no-op). :-p

[vchuravy]

Hm somehow I can't dismiss my review ;)

he can't do it himself

[haampie]

Can be closed now that #28847 is merged