In December, mst committed ec2e2ee5a0f010fe09d57e0176717b6b4f5671a2, which
removes the meta-attr object as the third argument to triggers, saying
unsupport passing meta-attr object to triggers because (a) it's not tested
(b) it's not documented (c) it makes it impossible to not close over the
meta-attr objects
As far as I can tell, though, it *was* documented and had been for a long time.
(Also, that commit only removed this behavior from inlined accessors and
constructors, so anyone not using make_immutable still got the meta-attr object
passed to their triggers if the constructor invoked them.)
A month later, nothingmuch merged in a big branch that reinstated the passing
of the meta-attr object to triggers in inline constructors, but I couldn't find
any comment about why this was done. (59f5bbde66d61d15b684be88d138fd798ba851d0)
Today, http://rt.cpan.org/Public/Bug/Display.html?id=44429 summarizes the
problem: writers (inline or not) invoking a trigger don't pass in the
meta-attr. Moose::Manual::Attributes document that they do and Moose document
that they don't. There aren't tests one way or another.
We need to pick one and stick to it. I don't want another fix (in whichever
direction) that someone else undoes a month later.
If it's "change the code to match the docs", that's done in
http://github.com/hdp/moose/commit/b0871377ee1fb633784cf69cd4d096a0c3183493
If it's "change the docs to match the code", that's easy; remove the mention of
the meta-attribute object from Moose::Manual::Attributes.
I don't know how to weigh these two concerns, though:
* It has been documented and working for as long as I can remember that
triggers receive the meta-attr object.
* Passing the meta-attr object to triggers rules out some theoretical
performance improvements.
(When I say it like it that, it sounds like we should change the code to match
the docs, but optimizing Moose performance is something I have no experience
with.)
hdp.
