[BUG] Variables in "block scope parameter" declarations should not default to `in` (const)

[Neil Henderson]

Describe the bug
(Apologies if this should be a suggestion rather than a bug!)

I want to write a loop and also use a counter, but I want the counter variable to be scoped with the loop, and not accessible outside. When I use the "block scope parameter" feature (not sure if that's the correct name!) the variable defaults to in and is therefore const, so I can't update the counter inside the loop.

To Reproduce
Run cppfront on this code:

main: () -> int = {
    words: std::vector<std::string> = ("Adam", "Betty");

    (i: int = 0)
    for words do (word)
    {
        std::cout << i++ << ": " << word << "\n";
    }

    // I don't want access to `i` here

    return 0;
}

See Godbolt.

The same applies when using the next syntax:

(i: int = 0)
for words next i++ do (word)

The relevant code lowers to:

{
  cpp2::in<int> i{0};
  for ( auto const& word : std::move(words) ) 
  {
    std::cout << ++i << ": " << word << "\n";
  }
}

and causes a C++ compiler error: cannot assign to variable 'i' with const-qualified type 'cpp2::in<int>' (aka 'const int')

My first thought was to try inout since I wanted to mutate i, but that doesn't work (see below).

I get the desired behaviour by writing:

(copy i: int = 0)
for ...

The 3 currently supported options are:

Syntax	Lowers to	Compiles?
in (default)	cpp2::in<int> i{0};	No
copy	int i{0};	Yes
inout	int& i{0};	No

Desired behaviour
There's a lot to like about unifying functions and local/block scope parameters, but I think variables declared in the "block scope parameter" list should have different defaults to function parameters.

My reasoning:

1. The purpose is different. When declaring a variable in a block scope parameter list, the intention is to declare a local variable that has a specific scope, as opposed to declaring a parameter that refers to a caller's argument.
2. Cpp2 local variables default to mutable. If they defaulted to const then the current block scope parameter behaviour would be consistent with them, but as it is there's a difference.
3. Having to write copy is not intuitive and also reads awkwardly.

[Herb Sutter]

Thanks! I hear you... I've switched the default for statement scope parameters between in and copy a couple of times, and I appreciate more usage experience-based feedback like this.

The most recent switch to the in default (and requiring explicit copy) is in this commit where you can see the effect in 7 places in reflect.h2 as it was then: d215985#diff-9124d88032e1de131eaa643c46e63344e15dd90e07eea20b2c0fcc5e3b9c84d5

See also the terse commit comments on that comment.

But I've continued to hit this myself, where I have to be reminded to write copy.

I'm not sure how to attach a poll to this issue... I'd love to get a quick poll of how many people think in vs copy should be the default. Absent a poll, perhaps it's time to switch back.

[Neil Henderson]

GitHub Discussions support polls but I think you have to enable them.

[Herb Sutter]

Ah, I finally found it -- enabled, thanks.

[Herb Sutter]

#1001

[Herb Sutter]

Thanks, everyone! Based on the results of #1000 I'll close this as "status quo" for now, and we can always reopen it again in the future if we get new information.