Change in import() behaviour for Perl > 5.39.1

[zmughal]

Fixes #167.

See also Perl/perl5#21269.

[zmughal]

To test this:

docker run -v $PWD:/work -w /work --rm -it perl:5.39.1 \
	bash -c '
		curl -fsSL https://raw.githubusercontent.com/skaji/cpm/main/cpm | perl - install -g App::cpm;
		cpm install -g Module::Install::AuthorTests Module::Install::DOAPChangeSets Module::Signature;
		cpanm -nq --showdeps . | xargs cpm install -g ;
		cpanm --verbose --test-only .;
		bash '

[kjetilk]

I'm baffled that this would break, but then, given that the test suite passes all the way back to 5.14, there seems to be no harm to merging it.

[kasei]

Is there an explanation for what the underlying issue here is? The linked Perl issue left things unclear to me. This particular change seems fine (since, as @kjetilk mentions, the tests still pass). But the overall trend of having to remove constructs like this seems contrary to the API design, and I'd like to understand the implications before changing code.

[zmughal]

Thanks for reviewing the code @kjetilk and @kasei. Yeah, I didn't fully
understand the change myself. I was trying different ways of getting import()
into the package… but it looks like I get it now.

After inspecting the output of Devel::TraceUse, I think that what was going
on was that Attean.pm was being required by Attean/RDF.pm and this was in
turn loading other packages before the Exporter::Tiny::import() was set up.
So I moved it before that and changed use Attean into require Attean.

It should pass via the Docker command and the GitHub Actions (once it is
re-run).

[kasei]

Great investigative work! And that does provide a good explanation for both the problem and why this is a good fix.