Regression: No such method 'op_props' for invocant of type 'Block'

[cfa]

The Problem

The following example from the documentation fails to compile in 2023.02 (and the current HEAD),

multi sub infix:<||=>($a, $b) is equiv(&infix:<+=>) { $a || $b }
 
my $foo = 0;
$foo ||= 1;
say $foo; # OUTPUT: 1 

with the following error

===SORRY!=== Error while compiling /Users/cfa/tmp/fail.raku
No such method 'op_props' for invocant of type 'Block'
at /Users/cfa/tmp/fail.raku:1

Expected Behavior

Compilation without error.

Relatedly, the documentation expects $foo to be 1 but bisection indicates that when this compiled (≤2022.12), the result was consistently 0.

A few relevant bisection logs,

• https://gist.github.com/76206dee2e26324b4b0b30ccad04701e
• https://gist.github.com/9516f200dc2befcf57d085c3743e9935
• https://gist.github.com/Whateverable/eb01bf61dcb88f5a7a0eab9425b60cc8

Environment

• Operating system: macOS 13.2.1
• Compiler version (perl6 -v or raku -v): 2023.12

[thundergnat]

Seems to me there are two separate issues in play with this.

First: += is a synthetic operator. It is not explicitly present in the grammar so the precedence level needs to be derived separately. That seems to have broken with commit a7ccfc6 "Move .prec from Routine to Code " That's definitely a regression.

Second: I don't see how that ever would have returned 1. Since the equal sign is part of the operator, there is no synthetic assignment operator, and there is no assignment within the operator routine, so the value 1 would never have been assigned to $foo. It will still be 0. And it is.

Leaving aside the problem with the precedence of synthetic operators, (just use <+> for now), I would expect that to be written something like:

multi sub infix:<||=>($a is raw, $b) is equiv(&infix:<+>) { $a = ($a || $b) }

my $foo = 0;
$foo ||= 1;
say $foo; # OUTPUT: 1

So regression and bad example.

[cfa]

Thanks @thundergnat.

Regarding the example, should the documentation be updated to use your amended version? If so, would you like to make the change?

[2colours]

That example is really bad anyway - ||= does exist by default, what value does it add to re-implement it badly? That's surely not "making your own operator"... I'd change that (and if nobody objects, I will) to something like |-| for distance (absolute value of difference).

By the way: if a7ccfc6 broke the precedence, how come the regression only happened more than 5 years later? That's surely not the whole story - although it's good to know that apparently the issue affects precedence derivation of synthetic operators, thank you.

[cfa]

Thanks @2colours.