In perl.git, the branch smoke-me/avar/universal-no-import has been created
<http://perl5.git.perl.org/perl.git/commitdiff/8804321d1a2f6e0eec133ffb898f7b73d39de547?hp=0000000000000000000000000000000000000000>
at 8804321d1a2f6e0eec133ffb898f7b73d39de547 (commit)
- Log -----------------------------------------------------------------
commit 8804321d1a2f6e0eec133ffb898f7b73d39de547
Author: Ãvar Arnfjörð Bjarmason <[email protected]>
Date: Mon Sep 26 10:59:13 2016 +0000
UNIVERSAL: Make "use UNIVERSAL" croak without making $pkg->can("import")
true everywhere
In 5.12.0 doing "use UNIVERSAL (...)" would start warning that you
shouldn't use it (1d9f57d), then since v5.22.0 it's been dying when
you do that (v5.21.2-13-g1178d2c).
http://perldoc.perl.org/perl5120delta.html#Deprecations
http://perldoc.perl.org/perl5220delta.html#use-UNIVERSAL-'...'-is-now-a-fatal-error
This change makes both "use UNIVERSAL" and "no UNIVERSAL" die. This
doesn't currently die and *will* break some code on the CPAN:
http://grep.cpan.me/?q=use%5Cs*UNIVERSAL%3B
The reason I'm doing this is because I had UNIVERSAL::import in my
code by just doing:
use UNIVERSAL ();
As some distros like IPC::Run pointlessly
do (https://rt.cpan.org/Ticket/Display.html?id=118082) would make
$pkg->can("import") true for any value of $pkg, which trips up any
code that for whatever reason wants to know if there's an existing
"import" method in a package, e.g. my own Constant::Export::Lazy:
https://metacpan.org/pod/Constant::Export::Lazy#wrap_existing_import
The reason for the comment about the import() being a historical
accident that we can't remove is because of (via #p5p on Fri Jul 15
2016):
08:05 < avar> ilmari: Why is the existence of UNIVERSAL::import()
something that can't be removed without breaking code? You
mentioned that in 1178d2c
[...]
14:38 < ilmari> avar: I should have reworded that comment
better. the import needs to exist to get an error message when
trying to import something
14:39 < ilmari> otherwise use UNIVERSAL 'foo'; would just silently
do nothing (unless we delete UNIVERSAL.pm)
14:39 < ilmari> that might be something to consider doing after
it's been erroring for a sufficient number major releases
The reason I'm not just deleting UNIVERSAL.pm (and moving the docs to
a UNIVERSAL.pod) is because that would then die with the way more
confusing:
$ ./miniperl -e 'use UNIVERSAL'
Can't locate UNIVERSAL.pm in @INC (you may need to install the
UNIVERSAL module) (@INC contains: .) at -e line 1.
As opposed to now:
$ ./miniperl -Ilib -e 'use UNIVERSAL'
Can't locate object method "import" via package "UNIVERSAL" at -e line
1.
I *could* patch the core to make *that* a special case, but not only
would that cause even more unexpected nastyness (e.g. you couldn't
provide your own UNIVERSAL.pm if you wanted in your $PERL5LIB), but
more importantly "use UNIVERSAL ()" while pointless wasn't the thing
that was deprecated in v5.12.0 & v5.22.0, only actually calling the
"import" method with an argument.
I'm making the "unimport" method have the same error while I'm at it,
while that was never technically deprecated, I can't imagine that
anyone will mind, and it makes the patch to gv.c easier.
-----------------------------------------------------------------------
--
Perl5 Master Repository
