Implement the return keyword

[smores56]

Closes #7104

This adds the return keyword to Roc, which allows returning early from anywhere in a function body. It unblocks the new try keyword, which is implemented using return : #7087

Here's a demo of the feature:

module []

staticValueDef =
    someVal =
        if 10 > 5 then
            x = 5
            return x
        else
            6

    someVal + 2

expect staticValueDef == 5

earlyReturnWithUnusedStatements = \n ->
    return ["$(Num.toStr n)"]

    if n > 3 then
        ["first", "with", "some", "items"]
    else
        ["other", "with", "multiple", "items"]

expect earlyReturnWithUnusedStatements 7 == ["7"]
expect earlyReturnWithUnusedStatements 3 == ["3"]

earlyReturnNested = \n ->
    (x, y) =
        if n > 5 then
            z = "early return!"
            return z
            (n, n - 2)
        else
            (n, n + 2)

    "did not early return with $(Num.toStr x), $(Num.toStr y)"

expect earlyReturnNested 7 == "early return!"
expect earlyReturnNested 3 == "did not early return with 3, 5"

earlyReturnOnlyReturn = \x ->
    return x

expect earlyReturnOnlyReturn 20 == 20
expect earlyReturnOnlyReturn "abc" == "abc"

Let's test this:

smores@smortress:~/dev/roc$ cargo run --bin roc -- test ~/dev/return-test.roc
...

── UNREACHABLE CODE in /home/smores/dev/return-test.roc ────────────────────────

This code won't run because it follows a return statement:

18│>      if n > 3 then
19│>          ["first", "with", "some", "items"]
20│>      else
21│>          ["other", "with", "multiple", "items"]


── UNREACHABLE CODE in /home/smores/dev/return-test.roc ────────────────────────

This code won't run because it follows a return statement:

31│              (n, n - 2)
                 ^^^^^^^^^^


── RETURN OUTSIDE OF FUNCTION in /home/smores/dev/return-test.roc ──────────────

This return statement doesn't belong to a function:

7│              return x
                ^^^^^^^^

────────────────────────────────────────────────────────────────────────────────

0 errors and 3 warnings found in 266 ms
.

Running tests…

────────────────────────────────────────────────────────────────────────────────
0 failed and 7 passed in 266 ms.

You'll notice that there's a static value that has a return in it. This PR marks those as warnings because we can still treat them as "early returns" to the end of the def block, pretty convenient that the implementation worked out that way!

[joshuawarner32]

Parser changes LGTM

[agu-z]

minor: this should probably say "but I need every branch of this function..."

[smores56]

I see what you're getting at, but I think this is misleading when we have if expressions/statements and early returns. For example, in this function:

myFunc = \x ->
    y =
        if x == 5 then
            return "early return"
        else
            x + 2

    if y == 4 then
        (y, 1)
    else
        (y, 2)

Your suggested error would imply that the branches of the y expression should both be (Num *, Num *) even though the branches of y are actually Num *.

I'm not sure of a better way to express this, so I'll leave it for now.

[agu-z]

I'm not sure I follow, those are not the branches of the function, they are the branches of a def inside the function.

[smores56]

I agree. I think since "branches" refers usually to if/when branches, referring to the function's "branches" is a little confusing I expect, even if I personally can figure out what you mean. It seems like the most precise phrasing would be "every place your function can return from should return XYZ". Would that be worse or better in your eyes?

[agu-z]

We should probably add tests to test_reporting to make sure the errors don't break when we make other changes.

[smores56]

I've added tests for test_reporting, meaning all PR comments have been addressed (or at least responded to).

[smores56]

These were removed during my attempts to fix tail calls, but didn't need to be. However, their default value is None anyway, so I kept them out.

[smores56]

@ayazhafiz I've responded to all of your comments, let me know if there's anything else you want to have fixed.

P.S. I used all of your wording suggestions, but there are a couple articles different, I hope you don't mind.

[smores56]

@ayazhafiz thanks for the approval! I had a one-line clippy issue that's been fixed, please re-approve when you get the chance.