LAST phaser doesn't fire with return

[alabamenhu]

The Problem

Here's the code that has issues with proper calling of LAST.

sub foo ($x) { 
    ENTER say "entered"; 
    LEAVE say "left"; 
    for ^$x { 
        FIRST say "first"; 
        LAST say "last"; 
        return if $_ == 0; 
    }
}
foo 42

Expected Behavior

The output should read in order, entered, first, last, left

Actual Behavior

The output reads in order, entered, first, left.

[lizmat]

For the record, it is documented to work: https://docs.raku.org/language/phasers#LAST

[lizmat]

Apart from the fact that you could argue that the LAST phaser should only be fired after the last iteration, it turns out to be very hard to implement this feature.

At a technical level, it's because the nqp::handle that handles the body of the loop, is not able to catch the return control message, unless it's within the lexical scope of the body.

So:

$ raku -e 'use nqp; sub a($a) { nqp::handle({return $a}(), "RETURN", nqp::stmts((say "caught return"), (return $a))) }; dd a 42'           
caught return
42

works, but

$ raku -e 'use nqp; my $b := { return 42 }; sub a($a) { nqp::handle($b(), "RETURN", nqp::stmts((say "caught return"), (return $a))) }; dd a 42'
Attempt to return outside of any Routine

does not.

Since to be able to support this, it would require same serious redesign with regards to return handling, and the fact that it could be considered correct that the LAST phaser won't fire until you've actually done the last iteration or done a last, I've adapted the documentation.

I propose to close this issue with a WONTFIX.

[alabamenhu]

Since to be able to support this, it would require same serious redesign with regards to return handling, and the fact that it could be considered correct that the LAST phaser won't fire until you've actually done the last iteration or done a last, I've adapted the documentation.

I propose to close this issue with a WONTFIX.

When we say a WONTFIX, are we saying it's going to be considered canonically correct as designed, or a "not sufficiently common and too difficult to fix at present"?

I'd update in the meantime the documentation to say that if there is a need for LAST to be fired with return, as a workaround, one can do last() && return, last() and return, or last() andthen return. All these ensure that last is called (although sadly parens are necessary)

[lizmat]

When we say a WONTFIX, are we saying it's going to be considered canonically correct as designed, or a "not sufficiently common and too difficult to fix at present"?

I'd focus on the "canonically correct" since the last iteration was never reached with return. And you could consider last not only as exiting the loop, but also as marking the current iteration as the last one.