[Chapel Merge] Add test locking in behavior of chaining calls ret

Notifications Commits
1

Branch: refs/heads/master
Revision: 661aaa0
Author: bradcray
Log Message:

Merge pull request #17106 from bradcray/chain-void-calls

Add test locking in behavior of chaining calls returning ‘void’ type

[trivial, and reviewed by @mppf]

While describing @leekillough’s recent work in PR #17007 to someone this week, I got myself re-stewing on a question that came up while it was being developed about whether we should permit chaining of void-returning routines, like:

proc foo(): void {
}

proc bar(): void {
  return foo();
}

bar();

At the time, there was a lot of uncertain mmmmmaybe-ing in the chatroom (myself included), but after having to describe the effort out loud this week, I became more convinced that we should not permit it. This matches the current behavior of the compiler, but I couldn’t find a test that explicitly locked the behavior in, so am doing so in this PR.

Specifically, the reason I think we split our original void type into void vs. none/nothing was to distinguish between:

  • procedures that return nothing interesting but do return some value/expression (none/nothing — the “unit type”) vs.
  • those that simply return nothing (void — the “bottom type”) — that is, all return statements are simply return; (or we fall out of the end of the routine)

An implication of this is that if I wanted to write the pattern above, I could do so, but would need to use the none/nothing types to do so:

proc foo(): nothing {
  return none;
}

proc bar(): nothing {
  return foo();
}

bar();

Yes, it’s a bit more typing, but I think nothing significant is lost and it avoids slippery slopes from the void case like "why can’t I capture the result of foo() into a temp variable and then return that variable? (answer: that’s what none/nothing are for).

Modified Files:
A test/types/void/void-chain.chpl
A test/types/void/void-chain.good

Compare: Comparing bdeac1efe70a...661aaa074c6f · chapel-lang/chapel · GitHub