Are we committed to supporting the separate assignment of the real/imaginary components of a complex?

[bradcray]

In a meeting today, we were discussing how most Chapel types are "singletons" that can't be partially modified, but have to be modified via re-assignment: integers, ranges, dense rectangular domains, strings, bytes. It was asserted that complexes had this property as well, and a few of us confirmed that that was our understanding / recollection, but checking, it seems we were wrong about that. For example, you can write:

var c: complex;
c.re = 1.0;

which makes complexes somewhat of an outlier compared to other built-in types (but if viewed as being similar to a record or tuple, perhaps doesn't make it an outlier at all).

This issue asks whether we're sure that this is what we want given that some of us had it wrong.

[bradcray]

Tagging @npadmana on this issue as one of our biggest users of complex values to see how he feels about this (my guess: wouldn't want it to change).

Tagging @gbtitus , @e-kayrakli , @mstrout as people who I recall as speaking up during this part of the discussion.

[damianmoz]

How is works currently is perfectly fine with me. Certainly this is how I use it in my code and have for as long as I can remember. You would break my code and numerous algorithms is you did not allow it, or at least provide a method to update the real or imaginary field on its own. I do not consider it an outlier. It seems perfectly natural.

[npadmana]

Agreed. I would also prefer to keep the ability to separately assign the real and imaginary parts.

[bradcray]

OK, thanks. I'm going to close this then, as I don't think there's any real argument for changing the behavior; I just wanted to note it because a few people asserted that it couldn't be changed piece-wise, and I wanted to make sure we hadn't enabled it without consideration.