RevBank reads the new products file on every interaction (e.g. pressing
enter), and then fires hooks like `product_changed`. Every running
instance gets those hooks, but the price tage should be generated only
once.
Was already implicitly required (since 59387ddb) because RevBank::Amount
uses the "isa" feature, which was introduced in Perl 5.32 (but no longer
experimental since 5.36, not 5.32 as the old comment said).
Perl 5.32 was released in June 2020, and ships with Debian bullseye
("oldstable") which was released in August 2021.
A space had a custom plugin that died during hook_checkout, which caused
the CHECKOUT lines to be logged without the corresponding BALANCE, and
indeed no account balances were updated. While the plugin had a bug, it
should not cause a half transaction in RevBank.
After some hesitation, I went with ON ERROR RESUME NEXT because if a
hook throws an exception, that should not interfere with other plugins
(the hook can return ABORT if this it was intentional), including the
calling plugin. An error message is printed (but not logged... TODO: add
hook_plugin_fail to plugins/log) but the show must go on.
During hook_checkout_prepare, however, nothing is set in stone yet, so
this could be used for something that might die, and this instance of
call_hooks() is now the one place where a failing hook should result in
the transaction getting aborted. For this, call_hooks() now returns a
success status boolean. Maybe it would make sense in more places, but I
didn't identify any such calls yet.
RevBank::Cart->checkout used to return a success status boolean, but it
could just as well just die (indirectly, to abort the transaction) since
it can't be called a second time within the same transaction anyway
(because ->set_user must be called exactly once), so continuing with the
same transaction can't result in anything useful anyway.
In some places, error messages were slightly improved to contain a bit
more information.
Fixes deadlock if hook_checkout returns ABORT.
One of these days I want to implement the abort mechanism through
exceptions, even though that means handling it explicitly in more
places. Or maybe *because* that means handling it explicitly in more
places.
