isassigned: incorrect bounds check for ndims > 1

[stevengj]

This looks like a long-buried bug, recently identified on discourse:

julia> isassigned(zeros(161,517), 500, 514) # should be false
true

The problem is that the isassigned code is checking only the linear index and not the multidimensional index, which is sufficient to prevent a crash but insufficient to check bounds for multidimensional indices.

cc @timholy, since this code originated in #11167, and @yuyichao who optimized this code (but did not change the semantics) in #20890.

[shlemas]

I'm the OP from that thread. I started looking at this on a whim, learning a bit about Julia dev if nothing else.

I have a naive fix here: master...shlemas:julia:isassigned-multidim

It's so simple that I feel like it must be non-ideal. For example, isassigned was doing its own simple bounds check instead of calling checkbounds. I see other versions doing the same. I don't know if there is an intent behind that (e.g. optimization) or if it's because the check is so trivial that the author just wrote it out. Interested to know either way.

[oscardssmith]

Looks good to me, can you make a pull request out of the branch? (press green button "Create pull request"). The only comment I have is that I think the second checkbounds isn't needed.

[shlemas]

Done. I removed the second checkbounds, tests were still good.