But it seems bizarre to me that it "can't" do this because some fairly rare programs dynamically write and name a subroutine. That can't happen very often.

The counter example in the thread you linked is not that. It is that a module which exports or even just provides a given sub may only be dynamically loaded at runtime. This happens more often than you appear to presume.

If your test suite doesn't have 100% coverage then this is the sort of thing that will bite you occasionally. The only guaranteed solution is ... improve the coverage of your test suite. Not the advice you wanted to hear, no doubt.

Than use AUTOLOAD to catch the call before crashing.

You haven't told us which behavior you expect, so like this you're free to implement it yourself...

Update

To catch a typo at compile time, drop the parens

See Re^3: How to identify incorrect subroutine calls in a perl program

Generally it's better to import the sub too, instead of a full qualified call.

For best results, use use v5.28;/no feature qw( indirect );/no indirect; along with omitting parens. If you don't, omitting parens won't always detect the problem at compile-time, and new weird issues could be introduced.

Like I already updated, you can drop the parentheses (and &-sigils if any) to enforce compile time checking.

FWIW: If you want this to happen with parentheses too, it's theoretically possible to write a module/pragma parsing the OP tree at the CHECK or UNITCHECK phase to abort compilation in case of absent subs in the namespaces. (See perlmod for phases)

I think I know how to implement this particular feature, but that's none of my itches to scratch. 🤷🏻‍♂️

Maybe worth checking CPAN before reinventing the wheel 🛞 (I haven't)

”I don’t want to write a test…”

I wonder what actually speaks against this: ok(__PACKAGE__->can("acme"), "acme");

Avoiding parens introduces new issues. It seems to me a third-party module could check that all non-method calls call a defined sub. Also, does perlcritic have a relevant rule?

Its not a matter of how often people write it: The syntax allows it, so its not a syntax error. This has a deeper meaning than you think - because it is not possible to detect this problem unless in the scope of an execution. Consider this:

my $name = 'est';
*{$name} = sub { print "hello\n"; };
est();
[download]

Even if strict is on, who is to say the assignment of the glob doesn't happen under the scope of no strict 'refs';? You could argue that anything you don't personally use... should be a syntax error?