Fix _sub_attrs for perl 5.37.9

[demerphq]

This fixes an issue with _sub_attrs() where it does not localize $SIG{__DIE__} when it does its eval probe. A fix to Perl 5.37.9 made it so that $SIG{__DIE__} is always called on a compile error, unlike before where it depended on the nature of the error and how many errors.

See:
Perl/perl5#20357
https://rt.cpan.org/Ticket/Display.html?id=146848
Perl/perl5#20885
fany/MIME-Signature#3

This also fixes the spelling issues reported in https://rt.cpan.org/Ticket/Display.html?id=146847

It also bumps the version. Once this is released with the _sub_attrs fix, MIME::Signature needs to be updated to depend on the new version. Please coordinate with fany/MIME-Signature#3 and fany/MIME-Signature#4

[haarg]

Can you change this to only include the __DIE__ change? The spelling changes are unfortunately dependent on how your local environment is set up, but there's nothing wrong with the existing text. "Behavior" is correct for US English. And the version numbers for this dist are automatically incremented by Dist::Zilla on release - the next release will be 2.15.

[demerphq]

Done.

[karenetheridge]

I am Canadian and did the last release; using the en_US, en_CA and en_UK aspell dictionaries there are no errors reported.

[demerphq]

On Thu, 2 Mar 2023 at 18:16, Karen Etheridge ***@***.***> wrote: I am Canadian and did the last release; using the en_US, en_CA and en_UK aspell dictionaries there are no errors reported.

I am too. But you are using the US spelling. What are your env vars? I suspect the tests are using 'LANG'. Graham said that Test::Spelling "doesn't have a way to provide a language. i'm sure an env var could help, but it's still dependent on what dictionaries you have installed. and then autodetects one of 4 different spell checkers, which have different behavior regarding contractions and other stuff." $ env | grep 'LANG\|LC' | sort GDM_LANG=en_US LANG=en_CA.utf8 LANGUAGE=en_US LC_ADDRESS=nl_NL.UTF-8 LC_IDENTIFICATION=nl_NL.UTF-8 LC_MEASUREMENT=nl_NL.UTF-8 LC_MONETARY=nl_NL.UTF-8 LC_NAME=nl_NL.UTF-8 LC_NUMERIC=en_CA.UTF-8 LC_PAPER=nl_NL.UTF-8 LC_TELEPHONE=nl_NL.UTF-8 cheers, Yves

…

-- perl -Mre=debug -e "/just|another|perl|hacker/"

[karenetheridge]

LANG=en_CA.UTF-8
LC_ALL=en_CA.UTF-8
LC_CTYPE=en_CA.UTF-8

[karenetheridge]

Can you change this to only include the DIE change?

Done.

This doesn't seem to have happened -- those two commits are still here. You would need to git rebase -i HEAD~3, delete the two commits interactively, and then git push --force-with-lease to update the PR.

[demerphq]

"force-with-lease"? Never used that, and it wasn't necessary FWIW. I just fat fingered it earlier and pushed to my master and not to my branch.

[demerphq]

LANG=en_CA.UTF-8

Huh. I don't get it, did you add "behavior" to your dictionary? That isn't UK/Canadian spelling: https://www.btb.termiumplus.gc.ca/tpv2guides/guides/wrtps/index-eng.html?lang=eng&lettr=indx_catlog_b&page=9U6Ot1RIV-vo.html

Anyway, doesn't matter, except if you are interested in getting to the bottom of why your tests and mine for the same thing don't produce the same results.

[karenetheridge]

As I said, I have the en_US, en_CA and en_UK dictionaries installed for aspell.

git push --force-with-lease is the safer version of git push --force.

[karenetheridge]

Is there a simple way of demonstrating the problem this solves with a test? What is the repro case?

[haarg]

It would be rather annoying to replicate. The problem in MIME::Signature is a script that does the equivalent of:

$SIG{__DIE__} = sub {
    warn @_;
    exit 111;
};

This is a broken __DIE__ handler, but wasn't previously being triggered due to a core bug. Testing this would require a subprocess.

I don't really think it needs its own test.