Inventory list: inv:set_list() not working anymore according to docs

[BluebirdGreycoat]

Obligatory Minetest version

Minetest 5.8.0-dev-65692ad1b (Linux)

OS / Hardware

Operating system: 6.2.6-76060206-generic
CPU: dunno, why?

Summary

Use of inv:set_list("listname", {}) inside on_metadata_inventory_take causes error from Lua:

Runtime error from mod 'bleh' in callback nodemeta_inventory_OnTake(): InventoryList 'whatev' is currently in use and cannot be deleted or resized.

This is a problem, because the docs for inv:set_list() state:

set full list (size will not change)

There's a mod running on my server which expected inv:set_list() to do what it used to do a couple Minetest versions ago: that is, to set all (or just most) of the itemstacks in a list at once, without changing its size. (IIRC, I think Minetest would clear out the bits of the inventory list not present in the method's table argument, in the case that the table contained too few elements to fill the whole list?)

Pedantic steps to reproduce

minetest.register_node("invlist:test", {
  description = "Test",
  tiles = {"default_stone.png"},

  on_construct = function(pos)
    local formspec =
      "size[8,2]" ..
      "list[context;test;0,0;8,1;]"..
      "list[current_player;main;0,1;8,1;]"

    local meta = minetest.get_meta(pos)
    meta:set_string("formspec", formspec)

    local inv = meta:get_inventory()
    inv:set_size("test", 5)
    inv:set_stack("test", 5, "default:mese")
  end,

  on_metadata_inventory_take = function(pos, listname, index, stack, player)
    local meta = minetest.get_meta(pos)
    local inv = meta:get_inventory()
    inv:set_list("test", {"default:wood", "default:stick"})
    
    -- It doesn't work even if the number of entries is exactly right >:(
    inv:set_list("test", {"default:wood", "default:stick", "default:apple", "default:chest", "default:tree"})
  end,
})

[sfan5]

dup #13465

[BluebirdGreycoat]

How is this a dup? The mentioned issue (#13465) was already closed with a merged solution, but I encountered this bug with the latest git pull.

In the meantime I have to work around this everywhere by using a inv:set_stack() loop in place of inv:set_list().

[Desour]

#13465 is similar, but it was fixed by allowing resizing in more situations (i.e. if the list is the destination in a listring operation).
The non-resizing set_list() can, and should still be made to work though.

[sfan5]

but I encountered this bug with the latest git pull.

I admit I didn't check the commit hash and assumed that we'd already fixed this entirely.

[SmallJoker]

The solution here might be to refresh all inventory list pointers within the inventory action code after each callback execution to avoid bad memory access in subsequent callbacks. That will again make the code slower and more complicated/bloated than it already is but eventually solves the issue.

[SmallJoker]

Cannot reproduce in current master, either due to #13497 or follow-up PRs.