To s or not to s, that is the question

[appgurueu]

Summary

I ran across this while working on BuckarooBanzay/luacheck#1.

Our docs say

Registered callback tables

All callbacks registered with [Global callback registration functions] are added to corresponding minetest.registered_* tables.

But our code does

minetest/builtin/game/register.lua

Lines 530 to 563 in 4843890

	core.registered_on_chat_messages, core.register_on_chat_message = make_registration()
	core.registered_on_chatcommands, core.register_on_chatcommand = make_registration()
	core.registered_globalsteps, core.register_globalstep = make_registration()
	core.registered_playerevents, core.register_playerevent = make_registration()
	core.registered_on_mods_loaded, core.register_on_mods_loaded = make_registration()
	core.registered_on_shutdown, core.register_on_shutdown = make_registration()
	core.registered_on_punchnodes, core.register_on_punchnode = make_registration()
	core.registered_on_placenodes, core.register_on_placenode = make_registration()
	core.registered_on_dignodes, core.register_on_dignode = make_registration()
	core.registered_on_generateds, core.register_on_generated = make_registration()
	core.registered_on_newplayers, core.register_on_newplayer = make_registration()
	core.registered_on_dieplayers, core.register_on_dieplayer = make_registration()
	core.registered_on_respawnplayers, core.register_on_respawnplayer = make_registration()
	core.registered_on_prejoinplayers, core.register_on_prejoinplayer = make_registration()
	core.registered_on_joinplayers, core.register_on_joinplayer = make_registration()
	core.registered_on_leaveplayers, core.register_on_leaveplayer = make_registration()
	core.registered_on_player_receive_fields, core.register_on_player_receive_fields = make_registration_reverse()
	core.registered_on_cheats, core.register_on_cheat = make_registration()
	core.registered_on_crafts, core.register_on_craft = make_registration()
	core.registered_craft_predicts, core.register_craft_predict = make_registration()
	core.registered_on_protection_violation, core.register_on_protection_violation = make_registration()
	core.registered_on_item_eats, core.register_on_item_eat = make_registration()
	core.registered_on_item_pickups, core.register_on_item_pickup = make_registration()
	core.registered_on_punchplayers, core.register_on_punchplayer = make_registration()
	core.registered_on_priv_grant, core.register_on_priv_grant = make_registration()
	core.registered_on_priv_revoke, core.register_on_priv_revoke = make_registration()
	core.registered_on_authplayers, core.register_on_authplayer = make_registration()
	core.registered_can_bypass_userlimit, core.register_can_bypass_userlimit = make_registration()
	core.registered_on_modchannel_message, core.register_on_modchannel_message = make_registration()
	core.registered_on_player_inventory_actions, core.register_on_player_inventory_action = make_registration()
	core.registered_allow_player_inventory_actions, core.register_allow_player_inventory_action = make_registration()
	core.registered_on_rightclickplayers, core.register_on_rightclickplayer = make_registration()
	core.registered_on_liquid_transformed, core.register_on_liquid_transformed = make_registration()
	core.registered_on_mapblocks_changed, core.register_on_mapblocks_changed = make_registration()

(the attentive reader may have spotted a subtle difference: some of our registered_* tables have an s suffix, like registered_on_placenodes, while others do not, like registered_on_priv_grant; there seems to be no consistent rule being followed here)

Steps to reproduce

Access minetest.registered_*s vs minetest.registered_*, or just 👁️ 👄 👁️

What we can do

1. Create aliases for the -s tables without the s to match what the docs say.
2. Deprecate the -s tables.

(or the other way around)

Other suggestions?

[rubenwardy]

For the most part, it looks like s is added if the thing ends in a noun, and no s otherwise. But there's the exception of registered_on_generateds

Proposal

I'd say no s for callbacks, and an s for object types like registered_items.

registered_globalsteps should also become registered_on_globalstep, so all callbacks are on_*, with the register function matching as well

[Desour]

Another Proposal

Prefer consistency. core.register_<thing> -> core.registered_<thing>s, always.

One exception though: If only one can be registered (e.g. register_authentication_handler), it's singular core.registered_<thing>.

(And obviously add aliases for the old names.)

[rubenwardy]

My issue with that is that registered_on_generateds does not make grammatical sense, "generated" isn't a thing that can be plural

[alerque]

If you want everything to be grammatical I think you'd need to change the scheme entirely and use registered_*_hooks instead of registered_*s. Some projects and programmers care about natural language grammar more than others, some even prefer the less linguistic/natural more technical/mechanical approach.

Not my circus not my monkeys, I just came here as the maintainer of luacheck looking to check if there was an official stance and/or community consensus and what linter rules should be canonical.

[fluxionary]

My issue with that is that registered_on_generateds does not make grammatical sense, "generated" isn't a thing that can be plural

it makes grammatical sense if you read the "on_generated" part in the quotative sense, as in "the registered 'on_generated' callbacks", similar to how "there are two 'the's in the sentence 'the man went in the house'" is grammatical even though 'the' isn't a noun and can't be plural.

that said, i'm not in favor of changing the status quo - other than documenting what's already there - because it'll mostly be a headache for mod maintainers with very little practical use except for not annoying pedants.

of the two existing proposals, i prefer Desour's.

[grorp]

I agree with fluxionary that it doesn't make much sense to rename all these tables just to have consistent "s". We'll have to keep the old names anyway for backwards compatibility, and then we'll have two names for the same thing, with the only difference being an "s" at the end.